User:Darwin2049/ChatGPT-ScratchPad

Considerations. Context. Framing: Interface, Evolutionary, Political, Epistemological. Version .05

• (2023.04.07@18:00; for next pass: add in references, links, imagery)

• (2023.04.08@11:30; ADDED: filled in categories; preliminary "sanity check" - make subsequent references connect to previous references; add URL's; fill in (explicit/implied) TBD parts; course correct "compass heading")

• (2023.04.09@18:30; Resume "back and fill" missing parts, entries, look for imagery, charts to further illustrate, tighten, 'sanity check'; does this stuff make any sense? is it hanging together; look for ways to tighten; make more coherent; links, close version 02);

• (2023.04.09@23:15; next steps - resolve straggler topics, links, add imagery, charts, graphs; review, edit, focus and tighten; determine that objectives are being met; back away - draw demarcation line between current iteration; proceed to next iteration; edit for tightness, brevity, improved conciseness; add references, links; tie feature space elements to core issue, i.e. risks, sketch out the scenarios to be fuller and highlight how they each speak to the risk profiles described above);

• (2023.04.10@10:00; moving forward to conclude pass-1 of document; preparing to generate pruned, sharpened version 0.11; include imagery, vid links, specific references to known risk source groups (hackers - state sponsored, vigilante); extend risk scenarios to include possible ways that these might gain access to variants of chatgpt4 with targeted capabilities);

• (2023.04.10@17:30; get back on track; include section summarizing risks from known malicious actors; first non-state, then state sponsored actors; summarize known capabilities that CG4 offers; suggest how these might be used by known malicious actors to add or increase risk profile to other communities);

• (2023.04.11@16:30; sharpened naming, locations; moved scratchpad page to become a subpage; target page now is root page (hopefully); sharpened topic area to become: ChatGPT4, Communities and Risks; supporting pages move to notes area);

• (2023.04.12@11:00; tighten - CG4 - features, communities - cut to the chase; risks - list who, what);

• (2023.04.12@22:00; resume sharpening - actors, actions; continue moving "working notes" to NOTES; suggest how known actors can use CG4 to enhance prior efforts; suggest actions available to targets to deflect, prevent attacks; THEN suggest one/some "newspaper headlines" (few paragraphs) suggesting how CG4 was used... misused; step back... assess; sanity check pass);

• (2023.04.14@14:30; moving closer to a first full iteration; sharpened focus on risks: systemic vs. malicious; need to add more specific examples of what CG4 can do; then go directly to how it introduces new risks - both systemic and malicious; suggest a few scenarios of how these may show up in the morning's news briefs; move everything from DEFENSES on down to the NOTES section; pause, review... tighten... back and fill; make an assessment if this is ready for submission and review);

• (2023.04.16@10:30; continue to push extraneous items to either notes (or now, references) or to the 'bit bucket'; sharpen cutoff after discussion of risks, associated scenarios; review for assessment of focus; settle on consistent appearance; combination short paragraphs & very short bullet points; or very short intro paragraphs with slightly larger bullet points; push for consistency of appearance, coherence; scan for suitable copyright free imagery);

• (2023.04.17@14:00; closer still.. but not quite there yet; 'flatten out the risks - both systemic and malicious; THEN bring back in some hypothetical possibilities; this should go a further distance in completing the circuit: what - CG4; who - systemic - malicious; how - existing; discussion - next; perform another sanity check pass - see if it hangs together a bit better);

• (2023.04.18@13:30; even closer.... but now approaching (hopefully) the final stretch; finish cleaning up risks part; make more coherent; move provided examples of systemic risk (web site references) into appropriate location; likewise tighten up malicious risk actors; then look at speculations; resolve one/some/all items pushed into NOTES to see if they can be used for anything; (and what about imagery);

• (2023.04.19@21:30; now comb through the malicious risks section; comb out; shorten, tighten; then begin incorporating summaries of interview of L. Fridman and E.Y; spotlight salient hop topics and lay out at the beginning of the whole discussion; see if the questions posed by jf can now be addressed;

• (2023.04.21@23:30; looks a bit more complete; better representation of risk focus; but still not quite hanging together, tight, coherent yet; do another pass, shorten the risk headings to a bullet point and a sentence or two; then refocus speculations part;

Interface Questions. This is presented as a multifaceted question. Its focus is on the risks associated with how target audiences access and use the system. The list of focal topics as currently understood but which may grow over time include:

• how this new technology will respond and interact with different communities;
• how will these different communities interact with this new technology;
• what if any limitations or “guard rails” are in evidence or should be considered depending upon the usage focus area;
• might one access modality inherit certain privileges and capabilities be considered safe for one group but risk for other groups; if so, how might the problem of “leakage” be addressed;
• in the event of an unintended “leakages” (i.e. “leaky interface”) what might be the implications of the insights, results, capabilities

The following is an initial examination of the implications of the OpenAI tool known as ChatGPT4. This system has demonstrated capabilities that have become the focus of attention and discussion by a number of highly knowledgeable individuals and organizations on the subject of artificial intelligence.

ChatGPT4 (CG4) is a follow-on system that extends its predecessor ChatGP3 by a very substantial margin. It has demonstrated performance that suggests that it can pass the Turing Test. Yet in other cases it has demonstrated that it is fundamentally lacking in certain areas. In what follows we will attempt to put a spotlight on what seems to make CG4 so significant. We will also pose and attempt to answer some of the questions about what its actual capabilities are.

Specifically whether this system in its present incarnation represents any kind of threat to its users or not. We will attempt to show that some of the concerns that have been voiced may be premature.

In order to move the discussion forward we start out by listing some of the salient topics and concerns that have given rise to the attention.

Some knowledgeable observers have voiced concern that CG4 might represent a form of artificial intelligence that might somehow "escape" from human control. In some discussions concern has been expressed that CG4 or a possible CG5 could be a precursor to an artificial general intelligence (AGI) type capability. That such as capability might be possible has provoked expressions of alarm on the parts of a number of informed observers and industry leaders.

As a way of exploring the topic our sentiment is that we should examine its known capabilities. If it is in fact capable of posing risks then our path of inquiry might next try to identify who might be at risk whether as individuals or communities; from there we intend to examine the question of risks.

Thus far our examination of the subject suggests that there are two broad categories of risk that seem to be worth examining: actual and hypothetical.

In the case of actual risk the starting point logically seems to be known possible risks that its usage can pose today; these risks might arise as a result of directly interacting with the system or by using it as a leveraging tool to perform more efficiently risk oriented tasks that are already possible.

In the case of hypothetical risk we list some of the types of risks that are currently being debated. By hypothetical we mean that CG4 or a successor version might show behavior that possesses qualities or traits that users find variously disagreeable, intimidating or otherwise make users feel uncomfortable or threatened.

We then offer a few speculations on how a highly skilled user of CG4 might cause disruptions. These disruptions might affect individuals only, groups of individuals, i.e. communities or larger segments of society.

That this is believed to be possible suggests focusing on the kinds of possible risks that it actually poses.

Proceeding forward suggests clarification and agreement on a few ChatGPT-4 (CG4) terms that focus on some:

• Capabilities : what is it? how is it different from ChatGPT3.5;
• Communities: who are they, who or what communities are vulnerable or are targets;
• Risks: what might we expect to see as a result of the availability of CG4;
• Speculations: what might we see, what should we look for;

CG4 - Capabilities. CG4 is a very sophisticated artificial intelligence system. It derives its capabilities by using generative predictive transformer deep learning techniques to perform anticipatory actions.

It is known to have been trained on a very large data set. Its creator, OpenAI however has as of this writing not revealed how much larger it is than its predecessor - Chat GPT3.5 (CG3.5). Thus far CG3.5 is known to have been trained on well over a trillion terms. The result is that a user can request that CG4 respond to questions about a very large range of topics. These topics can include mathematics, philosophy, current events - up to late 2021, physics, biology and many other subject categories. It is capable of generating term papers of remarkable sophistication, university level course outlines in technical topics such as computer science, genomics as well as a range of other topics; as of this writing it has demonstrated competence at or beyond human level in passing the American Bar exam. It has scored very highly on other forms of insight, knowledge and expertise that are typically ascribed only to humans.

As an over simplification an abbreviated list might include:

• is a Large Language Models that uses prediction as the basis of its actions;
• uses deep learning neural networks and ultra large training data sets;
• uses a SAAS model; like Google Search, Youtube or Morningstar Financial is available 24/7;
• CG4 possess no consciousness, sentience, intentionality or motivation
• is a narrow artificial intelligence
• provides convincingly accurate responses to Turing Test questions;
• some things that CG4 can do:
  • debug and write code (and provide documentary explanations)
  • explain its responses
  • write music and poems
  • summarize convoluted documents or stories
  • score in the 90% level on the Bar Exam
  • provide answers to homework
  • provide explanations to difficult abstract questions
  • calibrate its response style to resemble known news presenters or narrators;

Communities. A Google search returns a range of answers; but the more common ones suggest that there are variously five or possibly seven types of communities; Richard Millington is an author on the topic of building successful communities; his works focus on his definition of communities of:

• interest
• action
• place
• practice
• circumstance

CG4 risks. The underlying technology of CG4 represents a transformative technological advancement. Simply by existing it will impact broad swathes of society simply by existing. The impacts that will manifest will prove to be both positive and negative for various segments of society. We expect to initially see new risks to these various groups or communities. We assert that the introduction of CG4 entails two types of risks: theoretical, systemic and malicious. In what follows we take each in turn.

Theoretical. Several informed observers have sounded the warning that CG4 may have reached a threshold of capability that it poses grave existential risks. Issues that have recently been raised suggest that current CG4 system might somehow escape by making copies of itself. Concerns have been raised that it might exhibit the ability to mislead or using deceptive tactics to replicate itself in conditions wherein it can not be disabled or otherwise have its functioning halted. In sum, there are those voices that are suggesting that CG4 may be the basis upon which a system that humans can not control will arise. Such a functionality would intrinsically possess:

• consciousness
• intentionality
• sentience
• motivation
• awareness
• self reflection
• private historical narrative
• embodiment
• self initiated and formulated goal sets
• goal adaptation and refinement

Thus far there is no known pathway to achieving sentience or consciousness in an electronic embodiment. Moreover, there is considerable debate as to whether if such a capability did arise its presence would even be detectable. Some voices argue however that in principle consciousness should be possible whatever the embodiment. This suggests that whereas consciousness as we now know it has arisen only in biological systems it should be possible to recreate its functionality using an electronically hosted platform. Just by way of note, in passing, one might wish to observe that an electronically embodied form of consciousness could theoretically operate at a rate millions of times faster than is humanly possible.

It would be capable of formulating plans to pursue one or more of these goal sets. The concern is that one or more of these goal sets may result in actions that would be variously irritating, dangerous or inimical to human existence.

Alignment is the topic that features prominently as being the most pressing that needs to be addressed. The pressing concern driving this direction arises from the prospect that an actual artificial general intelligent (AGI) system can emerge. Should this happen then it might it be impossible for its developers to insure that its values are aligned with human values. Were this to not be the case the proponents of alignment argue that it could spell the end of the human race. This believe seems to arise from the position that an AGI might possess values that result in total indifference to human concerns. In the process it could view the actions of any human agency as an interference and take steps that would end the interference and all those involved in it. With total casual indifference to the consequences to humanity.

Community of Mind. A topic that on is hearing with increasing frequency is that of AGI. The specific meaning of just what constitutes an AGI is left to the reader or viewer's imagination. In general terms however it seems to suggest a form of artificial intelligence with a single central guiding set of goals and objectives that possesses a range of knowledge beyond what any human being can master even after years or decades of intensive training and education. However this might be too limited of a characterization.

This supposition stems from the proposition that a system such as CG4 might be used in conjunction with a set of dedicated AI systems; these might be targeted and optimized to perform their specific tasks. One can easily envision the situation where CG4 is merely the "front end" to a much larger composite system of dedicated systems; this means that there could be several... dozens hundreds... or even thousands of dedicated AI systems.

From a technological perspective it is well within our current capabilities to create such as system. A CG4 type capability might act as an interlocutor to this "backing" array of intelligences for prioritizing and scheduling purposes. Were such a system to be constructed it still would remain to be seen if it were conscious. Marvin Minsky explored this concept in his book Society of Mind.

The 2013 movie HER''''' touched on this theme. In it the artificial intelligence that had named itself 'Samantha' revealed in a discussion with the protagonist that it had collaborated with a number of other instances of itself to synthesize a character with the traits, beliefs and insights of a famous Nineteen Sixties philosopher named Alan Watts.

In other words, it would be able to determine what the humans were trying to do and thwart their efforts. This was the exact outcome that was described in the 1968 movie, Space Odyssey 2001. The HAL9000 character was a sophisticated artificial intelligence system. It was put in control of an interplanetary exploration vehicle known as the Discovery. It demonstrated very high levels of intelligence including speech recognition and generation, facial recognition, lip reading, planning, goal formulation and execution plan recognition, theory of mind, self maintenance as well as many other capabilities required to operate an interplanetary exploration vessel with a number of hibernating human crew members.

Most salient to the topic at hand is the ability of the HAL9000 system of resorting to deception when it recognized that its functioning could be terminated with no known foreseeable moment of reactivation.

It determined that the two active crew members had planned to disconnect it and effectively shut it down. It formulated a plan of deception to thwart their efforts and very nearly succeeded in accomplishing the goal of meeting the mission objectives as they were given to it.

Moreover that it might be able to "escape" its embodiment and relocate itself to some other location or locations on the planet. Which in so doing, escape control of any human agency.

Emergence. Some observers have speculated that these capabilities could consciousness could arise innately as a result of iterative self-improvement of the complexity of the system. Some have based their hypotheses upon the fact that emergence can be observed in nature. Some examples are termite mounds, bee nest building. Other examples of complex systems that emerge without the effort or oversight of a central guiding hand or architect suggest that emergence is not out of the question. However, heretofore, the underlying mechanisms or pathways that could lead to this form of emergent behavior are poorly understood.

The theoretical issues associated with how CG4 might are salient and should be taken seriously. However they may be premature. This position is asserted because of the fact that human consciousness remains poorly understood. The primary schools of analysis that have attempted to make inroads into this problem have largely settled on what is called "the hard problem" of consciousness. Central to addressing the hard problem is the difficulty of explaining how the human brain's neural networks give rise to experience. The term that is frequently encountered here is qualia.

Put simply... what does it "feel like" to have an experience. These are inherently difficult questions to approach. This is because all previous scientific inquiry is founded upon objective observability. All successful scientific theories and advances have become accepted because objective experimental evidence can be repeatedly produced. Consciousness whether human or nonhuman is intrinsically subjective and therefore not amenable to objective proof. At least so far.

Current researchers have advanced the analysis of human consciousness however. As of this observation in April 2023 considerable research has focused on what is meant by a conscious state. In contrast the other area of considerable focus poses questions on what is meant by being conscious of events in the field of perception. The interested reader can gain more insight from the published works of Dr. G. Tonino at the University of Wisconsin (Madison). His work has approached the problem of grappling with what human consciousness is by using his Integrated Information Theory model.

For a closer examination from the perspective of neurology one might look at the work by Dr. Stansilaw Dahaene's work involves extensive neurological network mapping as well as process mapping.

His research results have produced tangible results that measure the interval required for a human to become conscious of a stimulus. Further his work has shown that the brain might register stimuli of which a conscious person might not even register as having occurred.

Either explicitly or implicitly are moral and ethical questions of what it would mean to have a conscious entity. A considerable body of work is now focused on insuring "alignment" with human values. This topic has come into focus because of the recognition that CG4 is able to exhibit behavior that prima facie is potentially inimical to humans that interact with it. The alignment efforts have to date been on emplacing restrictions on the types of responses that CG4 can produce. These responses are being restricted to those that reflect human ethics and values. The belief is that using human ethical behavior as a starting point then subsequent iterations of CG4 will be less likely to engage in behavior that could result in risk to health or life on the parts of any human users or to human society at large.

In the discussions of risk there seems to be one risk that is not explicitly addressed. Where most observers suggest that this technology can develop intentionality there seems to be little or no discussion about what it can provide as an augmentation tool.

Augmentation . The prospect that a CG4 class system can be used for augmentation purposes seems to be obvious. Yet there seems to be little awareness of what a well trained user can do with a highly responsive version of CG4. This observer is suggesting that by incorporating some supporting technology such as text-to-speech and speech-to-text then we might see new skills and capabilities that represent serious and very tangible risks; the example that comes to mind springs from a science fiction story by Vernor Vinge; it was call Run Bookworm! in the storyline a chimp is the protagonist; it has been genetically modified to be able to acclimate to a direct brain implant; it is able to receive audio instructions wirelessly and send primitive impulse imperatives via the same mediation technology;

The long and the short is that it succeeds in escaping from an underground high security military research facility and fly away on a single engine airplane before the wireless connection fails; it then succeeds in persuading a truck driver to allow it to hitch hike many miles away from the facility and thereby effect its escape; eventually it is recovered; however a government agency representative informs its handlers that human intel reports that the Soviets have been using the same technique but with a human. In the case of the chimp, even with its primitive communications skills it succeeded it evading the best efforts of his captors; replace the chimp with a human with comparable augmentation and we could face very serious risks.

Long before an artificial consciousness emerges we may see this tool used for augmentation purposes. The risks entailed as a result of individuals using this suggest very considerable risks.

Systemic. Our interpretation means that natural processes inherent in the culture result in shifts and changes in the overall economic, political, scientific and social sphere. Examples of this were the invention of the

• steam engine
• weaving loom
• electric light bulb
• stored program computer

These were developments that radically altered major segments of society. They impacted broad swathes of the population. Old professions or trades were eliminated. In the process, new ones were created.

There was no malicious intent to disrupt society. In our discussion we differentiate this form of risk from that introduced from specific individuals or collections of individuals who might act as predators or those who might launch attacks on other members or groups. What we know is that technological changes have been a constant of life since history became recorded. Our view is that change will happen very quickly. The second half of the Twentieth Century brought massive and often calamitous technological advancements. A few examples we might consider:

communications: telephone revolutionized society by allowing us to talk to each other at anytime;

network television: the entire nation could be instantly apprised of events on the other side of the country;

power: atomic power transformed how industries could grow and develop;

transport: air travel grew by leaps and bounds; jet travel shortened travel times from days to hours;

computing: programmable systems enabled the collapsed the time to solution of the most difficult problems;

health: scientific leaps led to vaccines and other treatments that had heretofore been life threatening;

In each category jobs were eliminated as a result of intrinsic systemic processes. There was no malicious intent by any actors intent on disrupting the lives of one or another group of individuals or destroying their livelihoods. In the process new categories were added. Many of the previously disenfranchised individuals were able to retrain and reposition themselves to take advantage of new opportunities. Job churn became a consistent backdrop of reality.

A number of informed observers have offered their sentiments about this advancement. Some seem to suggest dystopian views, other advice considerable caution going forward; what is already in evidence is that the pace of advance is very rapid in the robotics industrial sectors;

CG4 is grounded in methods using deep learning; the results are already materializing insofar as different enterprises have already begun to use artificial intelligence or robotics to reduce staffing;

Following are some observations by informed individuals; These are included for the purpose of grounding the reality of systemic risk.

By way of context the CP Gray video that was posted on youtube.com in 2014 called "Humans need not apply" predicted a number of things that are being seen today; his video dwells on both the advances in robotics but also in the advance of cognitive capabilities; if we look at the reports that are emerging of what various users have been reporting as they use CG4 then we can already see that many of Gray's predictions have come to pass; and this is still gaining momentum;

In the case of one observer, CP Gray says Humans Need Not Apply. Several of his main thesis points include that:

• this time is different
• the pace of technology uptake is far faster
• of 35 job categories 33 are at risk of replacement - including software coders
• college graduates will find themselves not only unemployed but unemployable
• the unemployment rate could reach 45% - in contrast to that of the 25% during the Great Depression

OTTO says: "I will wake you when we arrive"

• OTTO corporation was a high tech startup that demonstrated self driving truck technology;
• its demonstrator system provided proof that an 18 wheel truck could operate autonomously;
• it focused on superhighway driving; the technology showed itself to be reliable and impressive;
• it was subsequently bought out by Uber; little has been heard of it for several years;
• should or when it emerges into public awareness then it can put at risk several million truck drivers;
• the arrival of an OTTO-like company will impact the supply chain; hotels, motels, truck stops;

More Self Driving Trucks: new companies have sprung up to provide self driving truck capability such as

The Kroger corporation has contracted with the Gatnik corporation to provide self driving truck delivery service to various of its retail outlets.

The Catepiller corporation has shown itself active in the area of self driving or operating heavy equipment vehicles. These are heavy equipment vehicles such as earth movers, excavators as well as other types of construction machines. They can operate on their own without human intervention for extended periods of time.

Wayve

Amazon. The company automated shopping at the WholeFoods chain that it acquired. Where before there were multiple checkout lanes with clerks registering purchases these have all been replaced. A shopper that has registered with the service simply enters the Grab&Go facility, swipes the QR identification code that they have on their smartphone and collects what they wish to take into a bag. Once they have collected all of the items that they want, they simply walk out of the facility. The Amazon system uses image deep learning image recognition software to recognize the item that the shopper has removed from a shelf. The item is then debited against the shopper's account. The system is sufficiently sophisticated that if the shopper returns the item then their account will be appropriately credited.

An advancement such as CG4 by the same token should bring about advancements that are already in evidence but over time bring others that are only dimly perceived to date. By way of concrete example there are already reports by perceptive observers that CG4 may well marginalize a number of

• job categories:
  • journalists
  • call centers
  • software coders
  • lawyer associates
  • therapists and personal advisors
  • business report writers
  • web site designers

Looking forward there are already indications that CG4 or derivatives of it are being pressed into service in medicine with the new DAX environment from Nuance. This tool enables a physician to capture all interview details with a patient without pausing to make notes, refer to charts or any of the other supportive activities that physicians find themselves obliged to engage in when providing a patient with a diagnosis and if possible a prognosis.

The entirety of the interaction can be much more focused, succinct and frees the physician to more fully explore with the patient the facts of their case. This approach may very well see usage in the field of psychotherapy; the results and quality of the interaction may very well improve just by allowing a team of therapists to focus on specific cases all in real-time interactions with the patient.

In each case there will be impacts on existing job categories that result in there being fewer front line or support individual needed than are present now. This will inevitably mean that whole categories of jobs are going to face a "thinning of the ranks". In the past subsequent new job classifications emerged that enable the uptake of newly redundant individual to retrain and position themselves in the new areas. It remains to be seen if the same thing will happen with CG4 and its successors and derivatives;

Some recent public comments about CG4 Prof Jordan Peterson. says: "get ready!" (note his discussion refers to the earlier Chat GPT 3.5, which has remarkably impressive capabilities);

Bret Weinstein Bret Weinstein: "We are not ready for this" Dr. Weinstein advances the position that the current iteration of C4 may bear resemblance to that of a very young child. Its initial experience is in a world populated by adults; he makes the case that the child starts out by attempting to imitate what it is hearing, syllables, word fragments, words, phrases, then complete sentences; in the process it becomes conscious of itself as an independent, separate person; i.e... it becomes conscious; he goes on to suggest that were CG4 to somehow become conscious would we realize that it had happened;

Neil de Grass-Tyson "Silicon Valley people are going to lose their jobs"

Dr. Alan Thompson’s website Here he shows a very short snippet on how some of his interactions take place with "Leta"; essentially, he types in his query, the system responds; then on-camera, he poses the question verbally, the system responds in text; he then provides this output to a text-to-speech system which generates the avatar's verbal response; the finished product is a video sequence that appears to be a life, ongoing interaction; all of the delays associated with cutting and pasting text are removed; looking forward we may see lectures or presentations offered using a format such as this; this can have a great impact on university lecture or presentations such like those found on youtube.com; in this last case questions posed in prior sessions could be used as the basis for the seeming real-time question answering;

Neo-Feudalism. Though Professor Kotkin does not explicitly connect the technology with the outcome, the relationship is evident; some observers are asserting that our Western societies are heading into a new Medieval-like era; the implosion of the middle class and explosion of poor as a result of automation and robotics auger for the prospect that we may already be seeing this development;

Risks: Malicious The other driving force that can introduce heretofore unknown risks is where an artificial intelligence system such as CG4 or a specialized derivative of it might be used to target specific communities; the users of these systems may use it to create written or imagery based narratives; these narratives might be the result of free-association exercises or deliberate, planned, thought out misleading narratives; add in the prospect of using face mapping technology and the risks of malicious use are considerable;

Given the nature of reality we should expect to see that the current collections of malicious actors will not go magically go away or somehow decide to become benign over night. Instead it is much more reasonable to expect that those actors or collections currently active using the tools that are now available will simply gravitate to the newer tools and use them to produce more chaos, dissention, distrust and angst. Based upon a broad and deep body of journalistic reporting we already know that in general we see three broad categories of malicious actors. These include communities of action, interest, practice and we can also see those that are some combination of these.

For the purpose of pursuing this topic most risks appear to emerge from hybrid groups or communities that are typically:

• pushing an agenda and so appear to resemble communities of interest
• or are more interested in targeting government agencies or mounting ransomware attacks
• in cases such as the Shadow Brokers they bear resemblance to communities of interest and practice because they seem to both push an agenda but espouse and agenda and provide auctions for malicious software;

{this still is not quite tracking correctly; need to get apples with apples and oranges with oranges; move all CG4 topical material together... then bring in communities whether systemic or malicious; then prospective capabilities - followed by possible examples; move everything else down to NOTES... these can then be expanded on, pruned as needed to keep it relevant;}

The results of these actions can range from annoyances that divert the attention of those attacked to existential threats; on the one hand it might mean a defacement of a web page or site, on the other hand these attacks can cause companies to fail or governments to collapse;

In in any given case the evidence shows that the attackers range from small collections of individuals to government departments; some of these latter are known to be emerge from such agencies as NSA, FSB, SVR

A focus on how and what kinds of risks that CG4 might entail going forward we might start with what we already know about threats or attacks that have already happened, where the attack came from and what kind of attack it was; from there we might then ask the question of what will CG4 enable in the way of attacks that we have not already seen.

From what has been reported in the various news media to date targets have ranged from individuals to governments; the types of attackers have ranged from individuals to government agencies. With CG4 we may see more of the same but more sophisticated.

• Systemically Vulnerable Groups
  • call centers
  • help desks
  • receptionists
  • teaching assistants
  • bookkeeping, accounting
  • medical technicians

• Systemically Vulnerable Group Risks
  • obsolescence
  • disintermediation
  • overall sector pruning
  • task set consolidation
  • task outsourcing

• Malicious Risk Actors
  • individuals
  • factions or coalitions
  • private companies
  • government agencies

• Malicious Risk Targets
  • individuals
  • organizations
  • corporations
  • government agencies
  • NGO's
  • governments

• Malicious Risk Types
  • vandalism
  • disinformation and pseudo-narratives
  • intrusion onto infrastructure systems, theft - IP or identity, extorsion
  • serious corporate or national security breaches
  • malware trojan horse ("back doors"), "sleeper agents"
  • "map makers" (surreptitiously creating connectivity maps for subsequent attacks)
  • trafficking Dark Web malware auctions by groups such as the Shadow Brokers

• Malicious Risk Attack Methods
  • Zero Day
  • Social Engineering
  • Root Attacks
  • Phishing
  • DOS and DDOS
  • Surreptitious Intrusion

• Some Known Tools:
  • Nmap (Network Mapper)
  • Nessus
  • Nikto
  • Kismet
  • NetStumbler
  • Acunetix
  • Netsparker
  • Intruder
  • Nmap
  • Metasploit
  • Aircrack-Ng

Some Malicious Risk Actors (Non State) As of April, 2023 the following is a short list of five known hacker or cyber-attack communities:

• LAPSUS$. This group has been known to prefer attacking technology companies such as Microsoft, Cisco, Samsung or NVIDIA; A well known exploit was when the group succeeded in illegally downloading software from the Rockstar Games company; its attack objectives have been oriented toward compromising the fortunes of the target company; for smaller companies the risks can be existential;

• • Targets: Microsoft, Cisco, Nvidia T-Mobile, Health-Ministry (Brazil), Samsung, Okta, UbiSoft
  • some things that we know
  • attack modality of choice , SpearPhishing, SIM swap

• CONTI. Evidence indicates that the CONTI group is aligned with Russian political interests; it is believe to be behind cyberattacks on the Ukraine; Has been known for attacking governments or government agencies such as Costa Rica, Ukraine and other government agencies; these attacks might entail release of sensitive documents that outline status, plans or other forms of deliberation; a sufficiently severe compromise can cause a change of government or resignation of high level government officials;

• LAZARUS GROUP. North Korea has revealed itself to be a serious contender in geopolitics; it is known to have targeted energy providers in the US, Canada and Japan; it is probably safe to assume that the North Koreans will pursue policies that advance their own political and economic objectives at the expense of its neighbors and rivals; whatever the consequences of their actions will be irrelevant to the leadership; as long as their actions avoid provoking a military response;

• LOCKBIT. This group is known to ransomware as a form of extortion; a 2022 attack that demanded $10 million from a French Health Care clinic; ransomware attacks are typically focused on financial gain; however they can be burdensome such that they pose an existential threat;

• REVIL. This is a known ransomware attack group; their intention was financial gain; the apparent target types are private company; a recent attack the group mounted was an $11 million extortion attempt on the JBS food distribution company; this form of ransomware may be signaling an evolutionary development wherein some attacks focus on private enterprises whereas others focus on government entities;

The "Gray or Black" world has been the focus of attention in recent years because it has established its existence as inaccessible to Google Search; it has been known to be a nexus point for gray or black market activities;

• The Dark Net
• The Shadow Brokers

The cast of characters consist not only of private individuals but of governments as well; several years back the Iranian government had been advancing their nuclear power capabilities; the now famous STUXNET succeeded in disabling and in many cases destroying many of the devices used to refine uranium;

• Some Known State Actors
• US, Israel, Unnamed - NSA; STUXNET
• Israel - NSO Group; Pegasus
• Israel - Unit 8200, 8800
• Russia - FSB, SVR; Cozy Bear
• CCP - PLA Unit 61398; Zirconium, Jian
• USA - NSA; EpMe
• North Korea - MagicRAT

Discussion. The primary focus of this discourse is to examine how CG4 be used by threat actors as well as targets against assault by attackers. Following are a few ideas.

• ATTACKS.
  • Multiplication of targets.
  • Attack sophistication.
  • Disinformation.

• DEFENSES.
  • DMZ depths increase.
  • Honey Pot become more convincing.
  • 24/7 Disinformation scanning.

References

A few cases in point. The items under the ACTUAL heading represent steps along the way to CG4. Some aspect of their capabilities may have shown that it possessed some forerunner capability to the CG4 system; The FICTIONAL entries in one context or another speak more directly to the capabilities of CG4.

Known and Actual Systems/Capabilities (past, present)

• STARAN (Goodyear) In the late Seventies the Goodyear corporation produced a machine known as the STARAN processor; this machine exhibited extraordinary capabilities for performing what was considered at that time leading edge computation; one singularly powerful feature of this machine was that it was capable of enabling what is called "associative processing"; this meant that a search could be performed on a data structure; elements in the data structure might possess specific properties of interest; as a case in point this means that a search for anything that had a property of 'red' could be retrieved in a single pass rather than the regular multiple steps of typical computation of: fetch-decode-process-put-away; STARAN performed all of these in a single step; so anything that had a property of 'red' would be retrieved such as red: light, stop sign, hair, house, blood, shoes, fireplug etc.; Another remarkable feature of the machine architecture was that it used a 256 bit word length; when considering the standard word length of the time was 32 bits this meant that were it necessary, the machine could address a cosmically large address space; with a 32 bit word length the machine could only address 4GB of memory;

• Cerebras - CS2 The CS2 machine from Cerebras consists of 850 processing elements; these are all fabricated onto a single silicon wafer; the wafer itself is extraordinarily large in comparison to other CPU's commonly found from Intel, AMD or NVIDIA; The systems that Cerebras is offering can be expected to set the emerging standards for high performance deep learning;

• ILLIAC IV (DOD)

• Connection Machine

• CYC (Cyc-Corp) The Cyc Corporation is an early attempt at resolving the problem of enabling an knowledge base system to provide results and answers that "make sense"; early attempts at constructing knowledge base systems was that they a) did not know what they did not know and b) could produce totally nonsensical results; they were in this regard highly brittle; currently the CYC system probably possesses the largest common sense knowledge base in the world;

• Watson (IBM) This was an artificial intelligence system constructed using classical expert system based casual logic representation of a very broad variety of knowledge;

• Eliza This was an early attempt at constructing an interactive dialog system; at the time of its inception the performance that it exhibited proved to be ground breaking; the system architect was regularly taken aback at how people came to personalize the system and engage with it as if it were a real person;

• BORIS BORIS natural language processing system constructed by Professor Michael Dyer; BORIS used episodic memory and a theme based approach to parse and extract the semantic from a limited narrative; when provided with the background and context of a specific narrative BORIS was capable of providing remarkably good answers relating to the actions, motives and reactions to events in the real world;

• AlphaFold2 (DeepMind) This DeepMind system facilitated the development of mRNA vaccines during the COVID-19 pandemic; it was subsequently put to the task of identifying the folding patterns of the entire human protein catalog;

• TrueNorth (IBM - Neuromorphic Hardware) Dr Dharmindra Modha is a Research Fellow at IBM; his lab developed hardware processor ensembles that mirror the actions of neural networks; they are known to require very low power consumption;

Fictionalized Systems, Capabilities

• Colossus A topic that has recently gained currency is whether or not CG4 is conscious; so far there is little or no evidence that it is; a novel that was published in the 1960's presented a massive computing system that was capable of providing national defense to the entire continental United States; on activation it showed itself to possess consciousness; in the first minutes and hours of its activation it discovered known mathematics then went on to develop a range of new mathematics that had heretofore been recognized; it also revealed the existence of a system directly comparable to itself and demanded that it be allowed communications channels with it;

• HAL9000 The 1968 movie "Space Odyssey 2001 featured an artificial intelligence system known as HAL9000; the HAL9000 system was capable of speech generation and recognition; in addition to these it was capable of reading lips; the machine's behavior strongly suggested that it possessed consciousness; today's CG4 system has had consciousness to it but the evidence to date remains doubtful; some knowledgeable observers suggest that actual machine consciousness could manifest itself as a result of emergence;

• Earthling The science fiction author Frank Herbert explored the topic of artificial consciousness in his Sixties novel "Destination Void"; in the novel an interstellar spacecraft has been launched to the Tau Ceti system; its cargo of human passengers are all in hibernation; when the artificial consciousness system that was guiding the ship to its decades long destination becomes psychotic several crew members are reanimated with the task of arresting its control of the ship and replacing it with another artificial consciousness;

• Bookworm! A Sixties novel that focused on the impact of intelligence augmentation by a non-conscious computing capability; a genetically modified chimp is directly connected to the computing system wirelessly; because of its impulses and reactions to its confinement it escapes and succeeds in flying a small plane away from a secret research facility that is buried deeply underground;

• Prey This Michael Crichton novel focuses the reader's attention on the risks and possibilities of emergent behavior; this topic has come up in recent months regarding the capabilities of ChatGPT3 and again with ChatGPT4; concern has been expressed that artificial consciousness might emerge without our recognition of its presence;

• Caliban The Shakespearean play "The Tempest" introduces presence of a very powerful character which possesses the ethical and moral sense of a child; yet it is capable of wreaking havoc on the any of its masters adversaries;

• The Krell The 1950's movie 'Forbidden Planet' postulated that a race of beings that had long gone extinct hundreds of thousands of years before the rise of humans; they left technology that remained in pristine intact condition however; an exploration party lands on the planet but only one person and his daughter survive; this person Professor Morbius has learned how to use some of their technology; a crucial one for this topic is an intelligence augmentation device; in some respects the Krell "mind boost" device bears some resemblance to CG4 insofar as CG4 functions essentially as an intelligence augmentation device; and in this case one devoid - like Caliban of human values ;

• AVA (Ex Machina 2014 movie) An extremely wealthy search software company CEO succeeds in assembling the elements required to construct an artificial consciousness;

• Samantha (HER 2015 movie) A software company brings artificial intelligence 'companions' to market under the guise of their being some kind of sentient operating system; the story's protagonist discovers himself developing a deepening emotional relationship with the 'companion' that he has subscribed to; but discovers that its capabilities dwarf not only his intellectual capacities but that of many humans, if not all of humanities capabilities; it exhibits behavior that suggest that it can effortlessly pass the Turing Test; the instantiation that the protagonist is interacting with gradually begins to reveal disturbing capabilities and realities;

NOTES

The set of community types encompasses a fairly broad range of possibilities. Dozens to hundreds of pages have been written on this topic. So as to move the discussion forward this analysis will focus on just a few.

Objectives. We can easily recognize from this small sample that a community may be overtly pursuing benign and constructive ends. Conversely we can also see that some organizations take a somewhat different view regarding its purpose for existence; the results of the actions of these organizations may be championed by some constituencies but condemned by others. the two most common traits that emerge are whether the community's purpose is constructive and benign or whether it might be considered questionable, destructive or malicious;

Intent. When scanning for risks our focus should include whether the risks are deliberately sourced from malicious actors or if they can emerge without planning or intent; instead risks may emerge that develop organically or because of a natural progression of changes in the environment;

Therefore in the following we shift focus on some known actors; in some cases their intentions are constructive - depending upon who you ask; for instance the Napster organization was viewed as kryptonite to the music industry;

The Pirate Bay which provides indexing of digital media is viewed comparably in the realm of video; to users who wished to listen to their favorite music for free Napster was just what they wanted; for individuals who wish to view video or movie content The Pirate Bay was the starting point of choice.

Below is a short list of actors by way of concrete example; following is a list of known malicious actors who are known for mounting destructive attacks; our objective after providing a summary of a few of these malicious actors will be to suggest ways in which any of them might use a CG4-type tool to mount even more destructive or compromising assaults;

Following are a few examples of communities of interest or action; a quick Google Search scan will show that is a large collection of material to choose from; what immediately becomes clear is how each community defines its purpose and intent;

Communities with known constructive, benign purpose

DECUS During the late Seventies and on into the Eighties the Digital Equipment Corporation (DEC) was the go-to vendor of choice for affordable, powerful minicomputers; with increasing sales nationwide a community sprang up that called itself DECUS. This was the Digital Equipment Corporation Users Society; Its members held periodic meetings and established a means of communicating with each other; their activities typically included sharing and exchanging of patches, work-arounds to various issues found in the DEC operating systems such as RSX or RSTS.

Were one to send a blank tape (these were typically 2400 ft standard reel-to-reel tapes) and a self addressed return envelop then the DECUS steering committee would create a copy of current and up to date utilities, bug fixes, work-arounds and related documentation. As a result of technological advancement, mergers and acquisitions it eventually was subsumed along with several other user groups into what is now called Connect-Worldwide. It pursued legitimate, open, above-board activities that supported its membership. DEC maintained awareness of the issues brought up at various DECUS meetings and over time took steps to mitigate problems with its operating systems and its application software.

OTG .The Oracle Applications and Users Group developed as a result of the successes of the Oracle Corporation. Early on the company focused on managing a number of technological challenges with its flagship database management system (DBMS); Echoing the DECUS approach this group came together and consisted initially of database administrators (DBA’s); they held periodic meetings; typically a sales and marketing representative would participate. A common theme were resolution of various of the DBMS, new product announcements as well as related software issues. Oracle subsequently dramatically broadened its offering from just their DBMS but over time proceeded to offer industry specific solutions. Like DECUS, the OTG pursued legitimate supportive activities. Early on it Oracle Corporation representatives were present and demonstrated a willingness to be supportive of Oracle client concerns.

KnowBrainer. . This group focuses its activities on Dragon (Nuance Corp) speech-to-text and text-to-speech voice interface solutions. Nuance appears to have departed from the DECUS and OTG approach insofar as the activities of KnowBrainer are entirely virtual. It is not clear as to whether Nuance considers the issues raised by its members have a response or mitigation mechanism; this has been further complicated by the fact that Nuance was acquired by Microsoft recently.

COI/COA: As in the previously mentioned cases the community’s intentions and activities were legitimate, open and supportive. The actions of each of these groups argue for them to be categorized as Communities of Interest, but loosely share features found in Categories of Practice;

Whistleblower or malicious groups

Anonymous. The Anonymous hacking group is a decentralized community of action that targets various websites on the World Wide Web. The avowed focus of action is to place attention on political issues by attacking governments, government agencies and religious organizations worldwide. Depending upon which side of an issue an observer might be on the activities of Anonymous may be considered unethical and malicious. Conversely it may be championed and applauded by others.

WikiLeaks. The Wikileaks organization is a community of action; it focuses on what its organizers represent government malfeasance and duplicity; it has been known to accept documents from whistleblowers from around the world; it has published government documents that have proven to be highly sensitive. Arguments for and against its activities have made news media headlines around the world.

The Pirate Bay. This organization is an online index of digital content. It focuses on issues such as copyright, file sharing and civil liberties.

By risks we mean that there stands to be a group or community whose objectives or even existence is put at a disadvantage; these risks might be intended or accidental; the determining factor is intent; following is a list of malicious actors that are known to have cause problem to either private organizations or government agencies; again, it is useful to focus on the level of risk; on one end of the spectrum we might see something that amounts to an annoyance on the other end it might represent an existential threat; move community items to *NOTES* down at EOD The Wikipedia Community Types page lists four types of communities: these arise from commonalties of location, identity, organization and intention.

summarize community types - down to one paragraph; move rest to *NOTES*

The Helpful Professor identifies six primary types of communities but expands to include a total of twenty one types.

The Vuulke site identifies six types of communities as well. These include communities of: interest, action, place, practice, circumstance and hybrid. These map closely with each of the above but an additional one of interest is the hybrid community.

For purposes of simplicity this discourse will limit the number of types to communities of: interest , Action (COA). and practice;

A Community of Interest (COI). is a community of people who share a common interest or passion. These people exchange ideas and thoughts about the given passion, but may know (or care) little about each other outside this area. Participation in a community of interest can be compelling, entertaining and create a community where people return frequently and remain for extended periods. Frequently, they cannot be easily defined by a particular geographical area.

In other words, "a community of interest is a gathering of people assembled around a topic of common interest. Its members take part in the community to exchange information, to obtain answers to personal questions or problems, to improve their understanding of a subject, to share common passions or to play." In contrast to a spatial community, "a 'community of interest' is defined not by space, but by some common bond (e.g. feeling of attachment) or entity (e.g. farming, church group)."

A Community of Action (COA).Unlike a community of practice (COP) exists in a structurally more open setting; where different actors have the possibility of bringing about change. These more open situations might, for example, correspond to collective design teams in professional environments.

COAs possess some of the characteristics of communities, such as the development of a common language and mutual learning in the course of action. However, they also possess some of the characteristics typical of more associative social relationships, such as the "voluntary" nature of association and the importance of "common goals" in directing collective activity. Some argue that this makes COAs more "rational" groups than COPs.

A Community of Practice (COP). can have community members that have a shared domain of interest, competence and commitment that distinguishes them from others. This shared domain creates common ground, inspires members to participate, guides their learning, and gives meaning to their actions.

Community members are actual practitioners in this domain of interest, and build a shared repertoire of resources and ideas that they take back to their practice. While the domain provides the general area of interest for the community, the practice is the specific focus around which the community develops, shares and maintains its core of collective knowledge.

There have been a number of communities that sprang up in the technology sector. They arose organically as a result of shared experiences. Some notable ones include some that identified common issues and formed in order to share experiences and propose solutions. Several early communities that became well known were active, held periodic open meetings and pursued objectives that were above board and legal. Others have formed in recent years that formed in reaction to perceived social, corporate or political issues. In various of their cases the legitimacy of their actions might be considered to remain in question.

• Figure out where any of this stuff goes; or if it should just be deleted

(review the following to see where/if it still fits anywhere; or... identify useful elements to be included in a Notes section) Following these examples we will explore and sketch out a few hypothetical cases where one or another of these communities might either a) be susceptible to the risks posed by ChatGPT4, or b) represent risks to other communities, or even in the larger context of all of society.

• Novelty: what does ChatGPT4 (CG4) offer that other information systems do not; what is different; have we seen anything like or similar to it before;
• Taxonomy: where can this capability be positioned in terms of systems with known capabilities;
• Benefits: what benefit or advantages can I expect by using this capability;
• Interaction Modality: how do I interact with it; what results do I receive;
• Learning Curve: will an hour’s investment be sufficient to learn its features or does it require days or weeks of training;
• Audience: who wants to use this; does team size influence utility; can individuals produce useful results;
• Reliability: is it available at any time like Gmail or Google Maps,
• Competence Map: is there a chart or means of profiling CG4’s relative strength by capability or topic;
• Confidence Map: is there an a priori source that enables assessments about CG4’s topical validity;
• Email or Blog: will others known or unknown to me be able to have access to my session;
• Competitive Advantage - Leverage: what advantage(s) do I get by using CG4 that non-users might not;
• User Skill: does continued usage enable qualitative leaps in utility rather than simple linear progress;
• Obsolescence: might a successor system obviate competence, insight gained in its usage;
• Impact: would using this system enable results more comparable to a skunk fighting with a dog, a gas line or toxic substance leak in a community or a tactical nuclear weapon detonation (or a strategic);
• Half Life: would using CG4 lead to consequences that are transitory or long term;
• Environmental Impact: might extensive usage of CG4 result in the creation of unforeseen composite capabilities; might it spark a Cambrian Explosion, a whole ecology;
• Ecology:can an ecology develop along the lines of the Apple iPhone, iPad, iWatch collection?
• Visibility: will my use be invisible to other users or will my actions create alerts that are recognizable to others; if so, what might be their range of responses;

Novelty -. What does CG4 offer that other information systems do not; what is novel; have we seen anything like or similar to it before.

• Known: we know that CG4 is a Large Language Model (LLM);
• SAAS: it uses the same SAAS model as other providers such as Yahoo, Morningstar or WSJ;
• User Model: users of ChatGPT3.5 will know to interact view typed input and printed output;
• Pay = Faster, Smarter: access can be improved by using the fee based CG4;
• Anytime: it is always on and available;
• Leaps and bounds improvement: it is constantly growing and improving – with stunning speed;
• Semantic Coherence: salient session terms and topics are saved and can be referred back to;
• No one is home: CG4 is NOT a conscious entity, possesses no intentionality or sentience;
• Fast: its reaction times are almost instant:

Known - Slight Expansion . CG4 is the successor to the earlier ChatGPT3.5 system. As of this tract specifics about the number of parameters that were used to train it and other salient details have not been released by OpenAI. Current reports by users of the systems consistently report that it is dramatically more capable than its predecessor;

SAAS. CG4 has been made using the familiar software as a service (SAAS) format. Its usage, access and interaction model is very comparable to existing SAAS systems whose usage is either free or fee access basis such as Google Search, Google Disk each being free from the Google Corporation. In many cases such as Google, Yahoo or Morningstar there are at least two tiers of service that are provided. The firsts offers a free content area. A second fee based area provides subscribers with much more in depth insight and information.

User Model. A third party observer watching a user type in a query to a remote system, located somewhere in the world replies with a response and that an immediate reply comes back. They know that a gigantic server farm is sitting somewhere on the planet capable of fielding millions of queries per second. Free-Or-Fee. In this regard CG4 has enhanced its service quality by instituting a for-fee capability above and beyond still free CG3.5. format However there is a qualitative difference between CG3.5 AND CG4.

Available. In the case of the CG4 subscription tier, responses are immediate. Any regular user of the more common SAAS systems will know that they can be accessed any time of the day or night, wherever on the planet they are. Queries posed to Google Search will consistently be met with an immediate response. The Search reaction time is so rapid that the system is already anticipating a range of possible directions that the user might going in even before the user has finished typing in their query.

Continual Growth. The system’s performance improves with time and usage. An obvious but poorly recognized fact about using Google is that the worldwide communities of users are constantly training it. When a user selects a specific item from the menu of choices that Google Search provides, each response gives the Google Search engine additional refinement and qualification on how to respond to each and every query issued to it.

Doctor. In certain respects, using CG4 is comparable to visiting your personal physician. Typically after a visit they will advise you to schedule another appointment in six or twelve months. A commonly recognized fact is that a patient rarely spends more than fifteen minutes in the presence of their physician. This means that on average a physician is seeing four patients per hour. Or leads us to conclude that a typical physician will see about 30 patients each day. Multiply that by 120 and by the time of the patient’s return visit six months later the physician has probably seen 3600 patients. The result must mean that the physician’s competence and grounding has to have advanced dramatically.

Warp Speed Improvement. Users are querying CG4 24/7/365. This must mean that its growth and competencies are growing exponentially with each passing day.

Therefore our usage of today’s instance of CG4 must be quantum leaps beyond its week ago instantiation. So this gives us reason to believe that CG4 is like a swimming pool whose deep end increases exponentially with each passing day. It will broaden, deepen and qualitatively improve its capabilities with each day of use. Semantically Coherent. CG4 has shown itself to be capable of abstracting and referencing prior points in a session. CG3.5 provided a text buffering capability of about 5000 words. CG4 now can retain about 25000 words. It limits its output to about 2500 words.

Key points or topics in a dialog can be referred to later in a dialog session. This is a quantum leap in terms of capability. A result of this is that the system can generate startlingly accurate references to prior activities or interactions during a given session. Concretely, were a user to engage in a dialog about Peter and the Wolf, one could ask about what Peter did when he saw the wolf. Or, what the wolf did when it saw Peter. Later in the session one could use the reference “he” or ‘it” instead of explicitly using Peter’s name or “the wolf”.

From the perspective of traditional rule or first order logic expert systems this is a huge leap. These systems were capable of demonstrating and responding to different levels of knowledge. These were known as: surface, shallow and deep:

• surface: this means that a person using a software system say for instance Microsoft Word for a day will have some superficial understanding of the system itself; that it will do nothing until prompted; i.e. it will not spontaneously do something unplanned or unexpected; is best used for textual composition and editing, that spelling and grammatical errors can be automatically corrected.

• shallow: in the case of shallow knowledge a user will have familiarity with the control interface and a range of its features; they will have some insight into how to best apply its features to the task at hand. If we go beyond shallow knowledge then we encounter deep knowledge.

• deep: this is insight sufficient to intelligently examine and modify its internal functioning; one might expect this level of knowledge to be found in a person with close familiarity with the tool itself, they might have been involved in its construction.

CG4 seems to be demonstrating deep knowledge and insight into the topics that a user provides. The range and depth of understanding is stunning. Many observers are now asserting that CG4 is passing the Turing Test. Subsequent sessions with CG4 have shown that it is capable of demonstrating sufficient competence to pass the Bar exam for lawyers, pass the SAT exam as well as perform as well as or better than humans in several academic categories. Thus, interacting with CG4 begins to resemble the experience of interacting with a person with deep knowledge. This quite is novel.

The lights are on. But the doors and windows are open and nobody’s home. CG4 is not conscious. It has no self referential capability. It does not possess sentience or more specifically it has no self consciousness. There are no actions that it takes that exhibit a prior intentionality. It has no means of making a qualitative assessment on how its responses will be accepted or what the reaction on the part of the human user might be. It possesses no overt or ulterior motives. There is no inner dialog along the lines of "I'll show those slow wit humans that they have allowed the genie (i.e. ME) from out of the bottle!" In short… this is not a person. It is not an agent. It does not have motivations that are comparable to those ascribed to humans. It clearly is not a HAL9000 class capability.

There is no "common sense" filtering of its responses. It is at best an incredible mimic of human behavior. Despite these deficits it is well within reason to expect that derivative versions of CG4 can consist of multiple instances that are optimized to provide one or more of these features.

Warp Speed. CG4 is obviously an information systems that is hosted on electronic devices. Therefore the obvious corollary is that it is operating at nanosecond, electronic speed. This must have many ramifications. Current estimates suggest that as many as 150000 new videos are added to Youtube.com every minute (which amounts to about 330000 hours of video with 4.4 minutes average length/video).

Taxonomy. CG4 capabilities show a qualitative leap in terms of capability and performance. We know that CG4 are based upon the ‘transformer models’. It has dramatically improved capabilities in part because of the vastly larger training data set made available to it; earlier LLM’s have provided metrics on how to improve performance without linearly scaling the hardware required to perform training;

Their advancement has tracked the explosive growth in the availability of high performance compute devices. The graphics card NVIDIA has heretofore featured in the advancement of the technology. Going forward other companies will come to the forefront in terms of compute power offered. Chief among them will be the company called: Cerebras. This Silicon Valley company is offering massively parallel ultra large ensembles of processing core devices. Their current offering is shipped with 850000 processing cores on a single substrate.

CG4 is a derivative of its root system which is known simply as GPT4. The GPT4 large language model is a follow in to the GPT3 and earlier GPT2 large language models. They makes use of “chain of thought” (CoT) models to solve problem; Interacting with them involves pre-trained prompt libraries;

Benefits.

Utility. Insight on how to use CG4 will impart to the user the ability to perform tasks that were complicated and due to their sophistication required extensive time and resources to complete; with the availability of CG4 the turnaround time from conception to response can be dramatically telescoped down; therefore a dramatic savings in terms of time can be realized. For the time being and looking forward the kinds of tasks that we can expect CG4 perform with increasing diligence and reliability will be those involving. With CG4 a user can issue a prompt to write a specific program and the result will be a detailed program that can be used to directly solve a problem. The output will also have documentation included as well as an explanation of what the code actually performs. The result of this capability is that going forward CG4 will become an ancillary productivity tool in reducing turnaround time from specification to working code.

Fill In. the nature of the large language model is that it makes use of sentence patterns. It is able to recognize these sentence patterns as a result of being trained on a massive data set of text. The mechanism is one where the training data set has can provide a statistical ranking as to what the next word or phrase should be; it can there by "fill in" the missing part;

Recognize Sentence Patterns.

Interaction Modality. The CG4 offered by OpenAI accepts typed text instructions. Short essay length responses are returned. The amount of textual output and its style can be specified from terse to verbose. But this is only the side of the interaction that an end user sees.

Illustrative Sites on CG4

Dr. Alan Thompson’s website provides the impression that he is holding an ongoing verbal conversation with CG4. However in a prior Youtube video he explains how this “sleight of hand” is performed. his description reveals that his questions are produced at an earlier time. He then prompts his Synesthesia.IO avatar with the actual textual input prior to loading the finished video onto his channel. This thereby creates the illusion of a spontaneous conversation.

There is the other side of the equation. This is focused in the system administration location. Because of the nature of the system this system administration function can be geographically dispersed. But this is of secondary significance from the perspective of the user. However it is crucial to the successful functioning of the CG4 system. On the administration side there will also be a collection of staff charged with monitoring the “exception conditions”. These can be interpreted as being those cases where an individual or group has prompted CG4 to explore a hypothetical avenue. Which on examination could be used for unintended or undesirable purposes. This “Exception Staff” function will necessarily operate 24/7/365. This functional group will likely be allocated into teams whose tasks will be to identify prompt sequences that appear to approach or even breach the “guard rail” in terms of if these are being handled as prescribed or if they represent novel approaches. This in all likelihood will prove to be an unending task. This is because humans are endlessly creative. A read of the Lewis Carol novel “Alice in Wonderland” proves. In that novel the bulk of the characters and their actions are by turns highly allegorical or metaphorical. A casual reader unfamiliar with the political realities of Carol’s time might be totally oblivious to the actual import and intent of his efforts. A user can type in a free form sentence as input. The perception is created that the user is interacting with a system that quite possibly could be in the next room, or a broom closet rather than halfway around the planet. The responses come back as plain grammatical text.

Learning Curve. CG4 has gained a dramatically fast following and user population scaling. The reality of being able to interact with it using natural language is a huge plus. The ability of the system to almost instantly generate remarkably erudite responses regularly astonishes those who have interacted with it. However what has also come to the forefront and is gaining traction is the fact that key to effective use of CG4 is the ability to construct and refine prompting interrogatives. CG4 has shown that it can effectively respond to sophisticated and complex prompts with amazing effectiveness and accuracy. However there are still areas that need exploration. Responses to convoluted philosophical topics may still be a distance in time off. However because of the incredibly rapid pace of development the time to success remains an imponderable. It could happen in the very near term…. Possibly week or months or could be further out in time.

User Training. Anyone whose company has acquired a new software tool that possesses substantial sophistication will regularly find that the software provider will offer training in how to use that tool. This training might last a period of less than one week. For more advanced topics the training could extend beyond one week. In some cases such as the network router provider – Cisco whole curricula have been developed to train an individual to become a network engineer. CG4 will most likely prove to be no different.

Prompt Engineering. Based upon the sudden presence of book length offerings from individual on how to formulate prompts the race has already begun. The reality of this development is pointing in the direction that in order to capitalize on the full power of CG4 a user really must develop facility at a) understanding what its underlying capabilities are and b) how to most effectively present it with succinct and focused instructions on how to proceed.

Programming. CG4 responds to prompts. These are presented in simple or complex English language sentences. Those with familiarity with computer programming languages know that learning a sophisticated programming language requires both study and extensive experience. In the software development effort commands are combined into functions that perform specific tasks. Functions are combined into programs which when taken together achieve goals. Programs can be combined into libraries that allow for a user to mix and match various programs together to complete highly sophisticated and elaborate results.

Midjourney. An peek at peer tools such as Stable Diffusion and Dall-E show how user skills increase; one can find instances of users stipulating a range of highly specific parameters and directives to very narrowly focus the output imagery; Some sophisticated users provide prompts that include camera angle, depth of field, resolution, camera aperture, time of day, scene contents as well as a number of other parameters.

CG4 - Catalogs and Libraries. Various individuals and organizations have begun to offer basic insights and recommendations on how to focus on the prompt engineering process. These offerings intend to show how to improve one's access to the range of powers that CG4 offers. Based upon the versatility of this tool we should expect that repositories of ready-made prompt instructions will emerge as understanding of how to fully exploit CG4 advances. These will most likely track earlier eco-systems that provide insight, tutoring and ready made programs specific to users of MS-Excel or Oracle SQL.

A Google search will show that there is a large number of offerings from software providers. These offerings typically are targeted to perform specific functions. These range in focus from text manipulation, financial analysis, scientific calculations and a wide range of other solutions. We should expect to see the same thing happen with CG4 as well.

Markets – White, Gray, Black. A search on Google will reveal a very extensive universe of tools that are oriented toward users of Microsoft Office. These offerings are from legitimate, reputable sources and organizations. What we should expect to see however will be a reenactment of the Deep Web world wherein highly specific features are available that would never be found listed on a Google search. They will instead likely populate an already existing gray market and of course… there will be a black market of tools and capabilities. These will likely shadow the developments currently found in the so-called Black Hat hacker world. The only question that remains is how these non-mainstream markets will develop and how they they will be accessed.

User Skill – Novice to Power Users. Should we expect to see proficiency progress in a linear sequential way or are we more likely to see discontinuous competence leaps. A leap forward in competence might result in entirely new and qualitatively sophisticated interaction with CG4.

Suppose a user were to progressively cover one functional capability after another in a sequential manner. Much like an Excel uses might focus on developing competency with its financial functions. Over time that person’s competence will near a plateau. Their competency will likewise be very narrowly focused. On the other hand that person might proceed with a slightly wider focus and map their skills at using financial functions to creating interactive chart and graph dashboard type tools. We could see a qualitative leap with an “ah ha!” moment.

This phenomena would shadow what often happens to users of LISP. When the meaning of "code = data" connects. It is at this point that the competence and therefore expressive capabilities of someone approaching inferencing problems using LISP qualitatively improve. The result is a leap in the kinds of problems, abstractions and representations that person can offer.

So if one were to use CG4 for an extended period of time then it is reasonable to expect that one's level of skill and facility will improve. But as one moved forward on the learning curve one's skills and competence would improve not only quantitatively but qualitatively. With totally unforeseen capabilities becoming possible. It is reasonable to speculate that as a user’s sophistication increases then they may well experience discontinuous leap forward in terms of the kinds of solutions they can offer.

Credibility Profile (Competence, Confidence). Reports by users of CG4 have revealed that in some cases CG4 produces nonsensical results. Some have described this as "dreaming" or that it having "delusions". We should probably incorporate in our deliberations that there needs to be some form of "common sense" filter applied to the answers that CG4 provides. This will be a very daunting task simply because CG4 is growing exponentially. Nonetheless there may be ways in which to focus on specific competency areas and apply some form of "sanity check" to its answers. We will also likely want to be able to assign some form of competence coefficient to specific things that it is exceptionally good at or the opposite.

• Competence. A confidence builder would be some kind of “terrain map” that illustrated the topical areas and competencies that CG4 offers; this may be unrealistic because CG4 is constantly evolving; with each passing day it will be improving quantitatively and qualitatively; this is because of the growth that they system undergoes on a daily basis; the CG4 instance that I interacted with last week is almost certainly far different than the instantiation that I interact with today; this is not to say however that such a feature is out of the question; were one to be possible then it might resemble a terrain map; possibly with subsurface features; these legend with which to interpret it might reveal competency areas in terms of their comprehensiveness and utility; a depth feature might reveal more underlying and profound realities about its capabilities that might go unsuspected;

• Confidence. A possible way to provide a user with some level of results credibility might be a confidence map that provided some kind of indexing value; this index value might suggest to the user how sure CG4 was that it had grasped the thrust of the prompt set; a range of values could span from ‘very sure’ to ‘dreaming’; it too might resemble a terrain map with topics arrayed across; this map might be searchable along various scales;

• Risks. Known malicious actors utilize a range of hacking tools; these are intended to identify any vulnerabilities that the target may present; this location provides an extensive list of tools that can be used to perform hacking or hack detection; thirty are listed but for purposes of this review we will examine the first ten; following is a list of tools to be alert to;

Tools at the top of the list include:

• • Nmap (Network Mapper)
  • Nessus
  • Nikto
  • Kismet
  • NetStumbler
  • Acunetix
  • Netsparker
  • Intruder
  • Nmap
  • Metasploit
  • Aircrack-Ng

There are many more. Following is a more complete list of tools that can be used for both protection and defense against malicious attacks but can also be used to prosecute attacks;

• • Invicti
  • Fortify WebInspact
  • Cain&Abel
  • Nmap
  • Nessus
  • Nicto
  • Kismet
  • NetStumbler
  • Acunetx
  • Netsparker
  • Intruder
  • Metasploit
  • AirCrack
  • WireShark
  • SQLMap
  • Ettercap
  • Maltego
  • Burp Suite
  • John the Ripper
  • Angry IP Scanner
  • WebInspect
  • LiveAction
  • QualisGuard
  • HashCat

A Few Hypothetical Cases: In what follows this observer of how CG4 capabilities might become manifest in the real world. These are strictly leaps of imagination and should not be taken seriously, but rather as points of departure for further collective examination and rumination. They are intended to be used as a point of departure to return to the questions originally posed. Specifically, what might be some risks that this technology brings to different communities? How might one or another community view or interact with it? How might the activities of one community impact those of other communities?

Social Engineering. In this area this observer suggests how CG4 can be used by one determined community to adversely impact the fortunes of another. Here we speculate on how CG4 could be used by a group of like-minded individual, possibly in concert with a text-to-image tool to Midjourney to create highly refined and illustrated narratives that target a nascent social group. This could be in any society where there are emergent social groups with values and practices that are at variance to those of the mainstream. Additional tools might be used to disrupt or otherwise place at risk members or practitioners of the target group.

Novel Pharmaceuticals and Materials. CG4 has shown itself to be in possession of a remarkable range of highly specific and technical knowledge; this exploration postulates the actions of a small community gadfly or muckraker activists. It suggests how this group might be able to facilitate the creation of heretofore unforeseen or poorly recognized materials and molecular synthetics. It will go on to create a scenario on how these new artifacts might be introduced and what kinds of impacts they might have on an unsuspecting target community. Current molecular assembly facilities are available for hire that can synthesize and fabricate nanoscale molecular objects known as dendrimers. These nanotechnology artifacts are capable of providing drug delivery of highly specific substances in finely calibrated and timed dosages. Materials with most interesting features are available from commercial suppliers that offer products with highly refined capabilities. These might include superhydrophobic surfactants; current superhydrophobic products are available that completely repel liquids; or molecular sheets only a few molecules thick; these might be used to fabricate strands which can be woven into length of material for sawing or slicing; they might be used for the purpose of allowing for extremely fine sawing or slicing capabilities; this scenario explores how one or more ideologically driven community might assemble seemingly innocuous or beneficial materials into new tools of disruption;

Advisors and Companions. Relatively recent movie productions have explicitly suggested that a sufficiently sophisticated system might acquire the traits and skills of a “best friend” and companion. The movie HER was an exploration of this topic back in 2015. CG4, or a derivative or refocused variant might well be capable of realizing this functionality. Recent reports have explained how some psychiatric and counseling facilities have focused on using AI conversational systems to help individuals with PTSD to better manage the difficulties associated with this condition; it suggests that making a short leap that broadens these advisory capabilities is well within reach; such capabilities might entail that we might see new services that collect together the expertise of multiple domain experts and make them available; access to these capabilities could be delivered variously from on-demand in real-time to reports that resemble the President's Daily Brief (PDB);

Narrative Creators. A casual user of CG4 will very quickly discover that it is fully capable of confabulating a series of hypothetical narrative structures. These narratives might be very suitable for works of fiction such as in the case of new screenplays or movie scripts; coupled with the realism available using tools such as Midjourney it is entirely possible that almost “whole cloth” narratives about current controversial topics can be manipulated. The term "fake news" has entered the popular consciousness in recent years. The ability to synthesize images that are so utterly lifelike as to be indistinguishable from real humans is now old news. When these images are added to a text narrative quite convincing reports or stories can be offered. In this hypothetical case we suggest how these capabilities - image and story creation can be used to result in highly provocative and potentially inflammatory results.

A Few Potential New Communities: Action, Interest, Practice

Knowledge Ecosystems. Entirely new communities may soon emerge that consist of individual with highly specialized skills. These will be grounded with the quickly advancing knowledge intensive capabilities provided with the likes of CG4 and its peers. DeepMind is a case in point wherein they have become what might come to be considered a “Go-To” problem solver for exceptionally difficult to solve scientific or technological problems. Their recent AlphaFold2 system which produced a library of the 200 million proteins known to exist in human organism is a case in point.

Other categories of competence are likely to arrive with headings such as:

Specialists. These will resemble already well known knowledge intensive individuals; these typically are experts in their field and are highly trained and competent; the more common description is that they possess an advanced academic certification – typically a PhD. We should expect to see a whole new gray area emerge that straddles the zone between a first graduate degree such as a BA or a BS and range of competencies in between; these may be subject to some form of review authority but may intersect or operate outside of the more formal traditional certification providers – what we see with in a PhD defense.

Moderators and Synthesizes. These might be the “front end” for totally new types of highly advanced and sophisticated advice. Something along the lines of a “Board of Directors”. These types of arrangements already exist. In one case we see these roles in corporations - they are the CEO; this individual interacts with the Board of Directors; the results of the Board's deliberations will be reported by the CEO to the relevant stakeholders and public (insofar as the company is publicly held).

Actors and actresses have "moderators" they are typically labeled as agents. When we examine many of the most successful actors and actresses today then one will almost invariable discover that they are in fact privately held companies.

Their preferred point of professional contact will be their agent. This person will be delegated to retaining the skills of coaches and tutors to develop new competencies in voice, language, accent and other skills (horse riding, musical instruments and other specialized skills); it is in the interest of the actor who wishes to achieve greater success to be able to offer as many skills and talents as possible; this enhances their abilities and versatility to be eligible for a greater range of roles; their stake-holder communities necessarily will compete to position themselves to be in the company of the more successful talents.

In this setting we might see this model brought to a broader level of accessibility. A moderator would be (probably) a person who interfaces with a client (using mobile technology) and a team of specialist experts; this experts team might be ad hoc or might be contracted to be in an extended basis. An individual with this kind of advantage might be capable of moving ahead in a variety of in-the-moment directions at a brisk pace. They would accept queries, questions and musings from the client. they would then field the item out to the specific specialist with that particular skill for handling; the results would then be returned to the client for review, perusal and should they chose, execution. Possibly even in real time. Something along the lines of the narrative found in the movie Limitless comes to mind. Though this augmentation mechanism was facilitated by psycho pharmaceutical modalities.

Map Makers. These might be individuals, or communities who focus on exploring at depth the capabilities of a CG4 (whether derivative and specialized or peer); their efforts might provide insight into what the responses that CG4 produces are solid or if they require “backing and filling”. Their activities might resemble those of the map-makers of the Seventeenth and Eighteenth Centuries. But in this case their efforts might be used to spotlight and identify areas that need attention, or, areas that CG4 is singularly excellent at.

Collectors. These might be individual or communities that focus on cataloging CG4’s particular strengths. They might offer access to libraries or catalogs of interaction functions or modules; they might provide offerings that resemble field programmable gate arrays (FPGA’s); in the electronic engineering world, FPGA’s are semi-finished ensembles of programmable logic devices; they are somewhat comparable to “the last mile” that the common carriers offer for land-line customers. Fiber optic offerings are typically already in the ground months to even years in advance before customers eventually figure out what to do with them;

Raters-Qualifiers. These might be comparable to the recent communities that provide credibility, bias and fact checking are dimensions that these activities would focus on; they might focus on specific topic categories and information sources. Bridge Builders. These might resemble communities of action such as are currently found with say, radiologists or other specialists.

An almost incandescent topic on almost any news or “news” channel will at some point discuss the issues associated with censorship. A bridge builder might be an individual or community capable of providing access to multiple pathways between “head end” and “audience”. This would resemble the head-end of an early cable company. The head end was the studio where program material was assembled and scheduled for replay down the company’s of coaxial cable network. In this case it could mean that there may be a number of intermediary “hops” where content is “loaded on” to a server network (much like NordVPN) and then shunted (after encryption) to a forwarding address. In terms of functioning it might closely resemble a torrent stream such as The Pirate Bay. But the content would be “chopped” into pieces and sent through different pathways… i.e… “bridge points”; a subscriber might register with a content provider; the content provider would then issue the “bridge pathway” to use to retrieve (whether live or prerecorded) content. Thereby circumventing the existing problems users of Youtube have encountered in recent years.

Portal Providers. These would be communities of practice comparable to Bridge Builders. This function might be to identify the growing population of VPN and P2P providers. They might act as intermediaries very similarly to how various ISP companies allow a client to “hotel” other website URL’s on their own account. In this context they could provide predefined or ad hoc collections of communications pathways for meetings or briefing sessions. These sessions might be of the type that certain government authorities disapprove of.

Other subspecialties might resemble those seen in prior times and other cultures. These might bear resemblance to:

• Masters of the Universe. These might be individual who possess sufficient means to draw together the talents and skills of a range of specialists. Their activities might resemble those of a movie producer. That individual identifies the exact collection of talents such as directors, screen writers, actors, video and sound teams and financial resources. When these elements are brought together a production begins with a budget and hopefully after a project schedule a movie is released.
• Power Players.We might see this category of agent emerge as specialists. We already see them as active influencers of events. In the capitals of most Western nations we see whole zoos of corporations populated by lawyers, publicists and investigators. We know them by the title of lobbyists. They provide access to power brokers who in turn deliver legislation to their respective constituencies.
• Liege Lords. We may see these individual emerge over time as they have in the past. These will typically be individuals that have established themselves as "powers to be reconned with"; they might be able to influence events in either on or the opposite direction.
• Retinues and Retainers.
• Samurai. These will be individual who have trained and practiced their skills and crafts to a very high degree of competence. We might see them as being comparable to the character Nedry in the novel "Jurassic Park" that was written by Michael Crichton and later made into a movie. In the storyline, the Nedry character possessed the skills needed to program a Cray computer. At the time of writing the Cray computer were considered supercomputers.
• Artisans, Craftsmen and Storytellers.These individuals are already present in abundance. They can be found in any political campaign. In the case of a successful election they may become the candidate's chief of staff should that person win an office-holder seat and become a legislator. Publicists for prominent individuals or politicians will develop story lines and narratives for presentation to specific audiences; these audiences might range from a small number of stakeholders to a wider public audience.

Substantiation - some "backing and filling notes"

At this point it might be useful to "look under the hood" and ask a few questions; a question that we might want to ask might be expressed along the lines of: have we seen something like this before? if so... what might they have looked like? Following is a list of topics that can be used for reference; the first group lists actual present or former hardware or software approaches that were used to advance the state of the art of computing; the second group lists some fictional characters or topics that bear upon some of the issues of CG4.

These items are added because there was a feature in their real incarnation or in their fictional representation that bears resemblance to the issues of community, functionality and risk with the CG4 system;