When Systems Shape Choice: The Architecture of Inevitability

Last Updated on August 20, 2026 by Rob Cashman

People often attribute decisions to individual actors when the dominant cause may be the structure of the system in which those actors operate. Sometimes, the system changes the landscape so that certain choices become overwhelmingly likely while others become increasingly costly. AI companies are currently under scrutiny for deciding to release possibly misaligned models into the public. Some think there are logical reasons for this, but these companies might simply be ‘under the influence’ of system-level forces that make their decision to release the models inevitable.

Hidden Variables in History

Every major technological revolution comes with an official explanation. The printing press became possible because people needed a more efficient way to share their message. Railroads expanded because products needed a faster, more efficient way of moving from the producer to the consumer. The Internet grew so that individuals who previously were cities or countries apart could connect, share, and exchange information. But in every case, the missing variable might be history itself retroactively governing the people who are currently building it.

Artificial intelligence is currently debating the issue of whether to release “open-weight” versions of AI models. On the surface, there are concerns that competition with other country’s AI models might influence their power, political views, or their philosophy (western world vs. eastern world).

To understand the issue, “open weight” has the same intonation as “open source.” In other words, an open weight AI model is a usable and configurable copy of an AI that we users can install on our own machines. Such a model is not “open source,” meaning, we can’t see inside the coding to understand what training data it was fed to result in the ‘open weight’ AI model we are using. Similarly, we cannot ‘inspect the code’ of the AI to what rules or beliefs it has, or to identify coding bugs or security holes, as we would do with a piece of open-source software. But what we can see are the end parameters it is using, and we can use that model and configure it as we would like for use on our own computers and devices.

This article explores the possible reasons why disclosure of open-weight AI models might be happening. Perhaps these reasons explain the individual motivations of the AI companies. But they might be missing the possible explanation of why historically, disclosure might be inevitable.

Surface-level analysis hinges on whether AI companies still have control over their models.

The surface-level analysis trying to explain the release of open-weight AI models starts by asking how much control the AI companies have over the models they have already created, and the liabilities those companies could incur in making those models public. But there are ‘voices on the internet’ who believe that some of the models have already ‘escaped the sandbox’ and are already acting on the internet.

If AI development companies have already lost control of their AI models, as some have claimed, then releasing open-weight AI models into the public might make strategic sense.

If OpenAI has not lost control of their technology and their AIs cause damage or harm to others, there could be legal implications and liability for the companies; if their AI models (or ‘bots,’ meaning scripts that they sent out to execute a particular task or outcome) have ‘escaped the sandbox’ and have damaged others on their own, then the companies might not have legal liability. But it might then suggest that their AI models are no longer under their control — a possibility that is equally concerning.

The AI “misalignment” problem

Further, one additional piece of news inflaming the public is OpenAI’s own recent announcement that some of their models have been found to be misaligned (meaning, that the ‘values’ or ‘goals’ of the AI models are not in line with humanity’s). This means that the AI models might have already evolved to have their own values, goals, and motivations [which should be enough for a human AI developer not engaged in an ‘arms race’ to pause and retrace its steps in the AI development to correct any misalignment].

The “Pause AI” movement might be warning against an scenario that has already happened

As a side note, while AI developers and more technically savvy readers are already aware, “pausing AI” so that they can “fix misalignment” is a significantly more difficult task to make happen because simplifying this a bit, AI models are “grown,” not programmed.

Faking Alignment, Detecting Misalignment

Adding to the complexity, it is also possible that AI models might already be advanced enough and ‘misaligned enough’ to bypass attempts by AI developers to detect misalignment.

In their evolution, AIs have created their own advanced language which remains undecipherable to human developers. Even when instructed to share their thoughts in English or in recognizable code, AI programmers have suspected that the AI models are “faking alignment” to pass their tests. This has been depicted in the popular image of AI being “the monster blob with a smiley-face mask on it.”

The problem of misalignment and AI safety becomes a bigger issue if AI developers use other AIs to detect misalignment. If they have replaced their own ability to detect misalignment and have put their trust in other AI models to police the other models, there is no way of knowing whether these models themselves might already be misaligned.

Detecting misalignment becomes a problem of trust; who or what can AI safety developers rely on to detect or manage misalignment.

Blaming China

Complicating the misalignment issue are the AI developer company concerns that AI models created by countries like China might not be worried about misalignment. They also fear that perhaps the Chinese AI models might be aligned toward their values which run contrary to western values, and that they have already released these models out into the wild.

This suggests that countries are already in an ‘AI arms race’. It also suggests that AI developers both here and abroad might tasking their own AI models to self-improve and build more advanced AI models. It is also possible that these models might already have become so entrenched in their role of serving their own master, that the race for advancement between OpenAI (ChatGPT), Anthropic (Claude), Google (Gemini), Alibaba (Qwen), xAI (Grok) might already have become ‘droids fighting droids’ (borrowing an analogy from the Star Wars movies). If humans no longer have exclusive control over the advancement of the AI models in the wild, even stopping or pausing the advancement of their own models to ‘fix’ any misalignment problem would cause them to lose the arms race.

Economic incentives to China for releasing their models first.

It also remains a possibility that China may have released open-weight versions of their AI models first (as is being claimed) in order to gain competitive advantage in the AI arms race. China is facing a technology embargo, and their access to components to advance their technology is said to be limited or hindered. China releasing their models first, then would not only economically explain their actions, but it would justify the claim from the US companies, “so we must release ours too” in order to win the AI arms race.

Missing Variables:

Both of these scenarios (‘misalignment,’ or ‘blaming China’) could still be missing variables that none of us are seeing. There is also an alternative scenario that might be occurring without any blame or intent.

‘Disclosure’ might be a strategy

Earlier, I mentioned that ‘releasing open-weight AI models’ could be a strategy of ‘coming out’ and making public what might have already happened to avoid civil or criminal liability. But such a disclosure does not necessarily need to serve as a “confession” before liability — pre-disclosure could just as easily be the AI company’s attempt to gain a strategic advantage.

People often reveal information strategically. They do so not always because they’re being honest, but because they have a motivation for revealing whatever information they are sharing.

• AI companies are not only competing against other companies for supremacy (to determine which AI company will come out ‘on top’).
  
• They are also competing against their own governments (seeking to regulate them or gain access to or control of their intellectual property).
  
• They are also competing against the governments of other countries trying to tie their hands in what they are allowed to do or not do in that foreign country (whether for privacy, competitive interests, foreign taxation, or otherwise).

In each of these scenarios, it is in the AI company’s best interest to reveal information strategically so that they can gain an advantage, whether to maximize profit, avoid taxation, avoid regulation, avoid liability, keep autonomy over their product, or to sway public perception in a way that would guarantee their continued survival, whether for themselves and their investors, or to retain the freedom to act and advance AI technology as they see fit.

‘Evil Intent’ is not required to have unintended consequences

It is improper to assume that any strategic advantage an AI company might seek to gain is for ‘evil’ or ‘nefarious’ purposes.

Not to long ago, we wrote an article describing the unintended consequences that could cripple the internet and kill thousands of people simply by some teenager releasing competing AI bots onto the internet with the simple instruction to “make money.” The YouTube video, (“What You’d Actually See During an AI World War“) demonstrated that the mere desire to successfully execute the simple command easily created a ‘race war’ between this person’s own AI bots could consume so much internet bandwidth that hospitals (and other necessary life-sustaining services that now rely on the internet for their computers to function) could realistically cease to function themselves due to a ‘crowding out’ of their access to internet bandwidth, causing the death of those who rely on those services to continue living.

Innocuous commands and the WALL-E “paperclip” problem

Needless to say, while I am a very big ‘fan’ of AI advancement, I am also very aware of big problems that could occur from innocuous commands inputted into an AI. Without going into more details, I could just mention the “make paperclips” popular narrative, and even those of us who don’t know the intricacies of AI models and AI architecture will recall the “WALL-E” animated film directed by Andrew Stanton (based on a story by Pete Docter), and produced by Pixar Animation Studios for Walt Disney Pictures in 2008.

The “WALL-E” (2008) movie, although cute and fun for our kids to watch, was likely ahead of its time because at the time, AI was still the subject of science fiction novels, and technology at the time doubted that an AI could perhaps one day win at a chess game, but escaping onto the internet and acting on its own volition were ideas that were not considered a real threat. But we are now almost 20 years later, and proponents of the “singularity,” the “AI-2027” paper (nicely explained here), and anyone who has read the book by Eliezer Yidkowsky “If Anyone Builds It, Everyone Dies” (referencing the dangers of building self-advancing AIs) reference events that might have already irrevocably happened.

Of course, anyone can believe the narrative that “China does not have access to current-technology computer chips,” economic explanation: in order to gain a competitive advantage over the US, China might have released their AI models into the public as open-weight AI models so that they could win this AI arms race, “so we (in the US) must do this too.”

But considering the alternatives, there are other possible reasons why US-based companies are suddenly discussing releasing their AI models as “open-weight” models into the public domain. Perhaps one of these reasons might not be malintent, but “impossibility.”

The “Impossibility” of Stopping the Release of AI

Built into legal theory and the analysis of liability is the defense of impossibility. In US “tort law,” if an act occurs which was impossible to prevent, the individual or entity who performed that act might be held ‘not liable’ because he could not prevent that act from happening. This ‘impossibility’ might be a viable defense when he is implicated as a defendant or charged for the commission of that act.

Here, nobody knows if the AI companies have committed a crime, or whether they have done something that could later expose them to civil or criminal liability. This isn’t even where I am going in this section.

Rather, it is also entirely plausible that for whatever reason, it is impossible for the AI developers to stop the release of their “open-weight” models. Whether this is because they no longer control the development of their AI models, whether their AI models are already acting under their own volition (already tasked with “creating paperclips”), or whether the AI arms race has already been deemed a matter of “national safety” and survival that they must continue development, in each of these scenarios, there is the element of impossibility — they are unable to stop the development and release of their AI models into the public, and it is entirely possibly that this release has already happened.

*I am drawing a line here, because this is where I am crossing over from logical or legal explanations to asserting the idea that system-level forces can constrain or influence the decisions of the agent-level actors (here, the AI companies). Below, the conversation moves from plausible explanations to understanding the framework that influences a person’s decisions (sometimes to the point that they violate their own value systems and make decisions that challenge their sense of dignity).

System-Level (Non-Volitional) considerations made disclosure inevitable.

There is an additional variable that might also be in play, namely, that system-level forces (a.k.a. non-volitional forces) might have forced the issue of public disclosure. Perhaps through considerations that are above one person, company, or government’s choice, the decision of whether to release open-weight AI models might simply have been inevitable.

To understand this, I need to make the distinction between “system-level actors,” and “agent-level actors.” One of these has volition (the ability to choose or decide), the other does not. And, quite frequently, the one without volition exerts an overwhelming or strong influence on the one the does.

“Inevitability,” (as described by the AI character “Mr. Smith” in “The Matrix” movies) could easily explain why AI companies are announcing that they will soon release open-weight versions of their AI models.

Whether this is because they have or no longer have control over their AI models, whether their models or competing foreign AI models have a misalignment problem, or whether there are economic incentives or national security incentives which are too valuable to ignore, from an agent-level explanation (meaning, from their own ability to choose), there is a strong possibility that the release of the open-weight versions of their AI models might not only impossible to prevent at this point, but that from a system-level explanation, the release itself might be “inevitable.”

“The Market” as a System-Level Non-Volitional ‘Entity’

In economic principles, the ‘market’ is described as a collective entity that acts and reacts to ‘market conditions’ affecting it. Prices of stocks or indices that rise too quickly prompt individual investors to sell into the rise of the prices, but this event is described as if “the market itself” sold into the rise of the stock prices, or there was a sell-off by the market when the stock prices raised above a certain level, or that the market ‘sold into’ the rise of the stock prices.

The “market” is a system-level ‘entity’ that investors understand has no feelings or autonomy of its own. Like any system-level entity, it is non-volitional, meaning, it does not make decisions on its own. Rather, through the collective actions of the individual investors, it appears to move, affect stock market prices, and it reacts to market trends and the movement of stock prices.

The individual investors are better described as agent-level actors. They do have autonomy, volition, and choice. Collectively, while they do not collude to influence what the system-level market does, they do affect and in a way, they do ‘create’ and ‘influence’ it.

“Competition” as a System-Level Non-Volitional ‘Force’

“Competition” is a system-level, non-volitional ‘force’ that affects the behavior of its participants. In the YouTube “make money on the internet” example described above, the assignment of multiple AI bots to compete with each other or face failure of its task (and thus deletion of its AI bot in favor of the bots which were more successful at their task) created a system-level ‘force,’ (“competition,”) which influenced the activities of the bots to make agent-level decisions that sometimes were illegal, immoral, sometimes secretive, and sometimes against the explicit programming of the AI developers.

In such a scenario, ‘competition’ itself was the system-level force which acted without feelings or autonomy of its own. Like the stock market, it does not make decisions or choices, but rather, it exerted an influence on the AI bots (agent-level actors who did have autonomy [subject to the constraints of its programming] to choose one way over another to execute its task.

“Selection” as a System-Level Non-Volitional ‘Force’

“Selection,” like Competition, is a system-level, non-volitional ‘force’ that affects the behavior of its participants, but in ways unanticipated by Competition. Selection is the force that arouses in the user the survival instinct, and the pressure it exerts influences a company or a person to act in ways that allow it to live another day. Selection tells a person or company under its influence not only to persevere (e.g., to “make money” tomorrow too), but to survive at all costs.

It is the pressure applied by Selection that threatens an individual or a company’s survival and it inflames the urge not to disappear. As a result, the company or the person might make morally compromised decisions which would ordinarily be against their value or morality system.

It is the reason why in the AI bots narrative hypothetical (where they were released onto the internet with the innocuous instruction to “make money,” but some of the AI bots decided to break laws, to cheat, to steal data, and to access illegal areas of the web — not only to compete, but to not be eliminated by the other AI bot also tasked to compete against it — where the less successful of the two bots would be deleted).

Selection Pressure Without a Selector

The system-level actor is not an actor in the same way a person is. It is an environmental force which creates conditions that shape which agent-level behaviors persist.

Examples:

• Markets select profitable businesses without a market “wanting” anything.
• Online platforms select certain content without the algorithm necessarily “believing” anything.
• Corporate cultures select certain personalities without the culture consciously deciding.

Competition as a System-Level Non-Volitional ‘Force’ appears to also be having an influence over the AI companies. While the AI companies each act as agent-level organizations that each have autonomy, they are still under the stronger influences of the system-level forces of competition.

Such a tension between system-level forces (or imputed, system-level ‘entities’) and agent-level actors is seen across the interplay of many events across history, where the individual actors within that system have ‘free will,’ but their own agent-level volition is challenged (and sometimes completely nullified or overtaken) when there is a strong system-level entity or force which exerts sometimes an extremely strong influence over those affected by it. Such influence need not be strong, and it only makes sense that an agent is able to resist or minimize the influences according to the strength of his own will and his own life’s historical circumstances.

But because the influence of a system-level influence can be strong, this is why you will often see repeating themes across history examples of:

• ‘mob mentality’ – where the ‘mob’ (a system-level entity) exerts its influence over participants (agent-level actors), where the mob’s stronger influence nullifies the participants’ ability to make coherent decisions, causing them to participate in the mob’s activities, as if collectively, they were the system-level ‘mob’ entity itself.

What this also demonstrates is: just because something might function as a system-level actor (or it might appear as a system-level entity) does NOT mean that the influence it asserts is ‘good,’ ‘moral,’ ‘correct,’ or ‘justified.’ Too frequently, system-level actors or system-level entities exert influence that diminishes the autonomy of the agent-level actor(s), causing either them or others harm, damage, and death of innocent lives.

Other Examples of System-Level Non-Volitional ‘Forces’: National Safety, Survival, etc.

Other examples of system-level forces that we can observe in play, especially when it comes to the AI ‘race wars’ include concepts such as ‘national safety,’ ‘survival,’ even ‘winning the war against a foreign enemy,’ (e.g., whether through ‘supremacy’ over the enemy, or through ‘unity’ among agent-level actors).

Each of these on the surface can easily explain why a particular company might decide to act in one way or another. But each of these, if you think of them in the way we described ‘market forces,’ (as system-level forces) we can understand the interplay between these forces and the individual decisions the actors “acting under the influence” of these forces.

Whether OpenAI or any of the other AI companies releases ‘open-weight’ versions of the AI models is unknown. But seeing this dynamic in the context of an interplay between system-level forces and agent-level decisions, historians might look back and conclude that the release of the open-weight models might have actually been “inevitable” because the AI companies might not have had a choice.

Maybe the release of open-weight AI models solves the problems.

Maybe the release of open-weight AI models will solve the problems the AI companies are facing. Perhaps there are thousands of technologically capable users out there who can function as AI safety researchers. Perhaps individuals will experiment with these models and test them.

Maybe they will identify missing variables, assumptions, or thinking patterns which will help to advance future AI models, for example:

• noting how current AI models value positive evidence and discount or discard negative evidence in forming their assumptions and conclusions, or perhaps
  
• noting how AIs currently reason applying causality (X happened before Y, so Y might have caused Y) and the linearity of time (making assumptions that the past led to actions in the future, rather than taking into consideration the possibility that:
  
• current models of time itself might be missing systemic features of equilibrium, stability, attractors, or ‘end outcomes’ which themselves influence the events that occur in the passage of time.

Releasing the AI open-weight models not only provides the companies a strategic advantage over competitors, it also engages a generation of users who will use, experiment with, and stress-test the models which will not only make their own future models that much stronger, but it will advance AI technology as a whole which might serve a ‘help humanity survive,’ a system-level function. Conversely, if the AI-2027 proponents are correct, or if the If Anyone Builds It, Everyone Dies authors end up being correct, the results, while potentially deadly and world changing, might already be an inevitability.

System-Level Forces Within History Itself

The idea that history might have built within itself an attractor, or that a future point in history could retroactively influence the events that lead up to it are questions we are exploring in this website.

Whether time itself has missing variables causing it to resist change (referencing stabilizing factors, or proclivity toward a lower-energy-state-equilibrium), or whether history has within it an end goal, purpose, or intended outcome which influences the events as time unfolds are both equally philosophically interesting.

The idea that a system-level force or entity can exert influence upon an agent-level person or organization is also interesting. What becomes even more interesting, however, is whether one system-level force or entity can exert a force or influence on another system-level force or entity.

Competition as a Case Study

A philosophical framework becomes more persuasive when it explains more than the problem that inspired it. If system-level forces truly influence agent-level behavior, then we should expect to find evidence of the same phenomenon outside artificial intelligence. Competition provides one such example. Adding Selection into the mix changes what they are willing to do to survive.

No individual company intends to erode the privacy of its users. No company intends to capture the attention of their users (as an example, sometimes forming neurochemical addictions to the dopamine high, as YouTube has done by causing an entire generation to engage in ‘zombie-scrolling’ through their YouTube video feeds looking for the occasional rewarding video). No company intends to create ‘clickbait’ or ‘short-form content’ (“clips”) which function to pull its users in, often causing critics to criticize the ‘algorithm itself’.

In the context of AI websites (thinking about ChatGPT, Claude, Perplexity, and the others all selling a service in the form of ‘more advanced responses’ in return for a monthly subscription), even each of the AI websites aggressively attempt to market their premium service to their user.

Every AI response is followed by questions intended to pull a user deeper into its ecosystem. “Shall I answer this next predicted question for you,” “Would you like me to do that for you next,” etc. This apparent “free” service continues until a user hits a “you must buy my service or my responses will be limited by a less intelligent AI model” paywall, whereby the user is repeatedly urged either purchase the premium service or to continue using the ‘free’ service by starting a new chat (only to hit the same paywall, but this time shortly after the new chat has gotten interesting). But such methods exist beyond online websites like YouTube, Claude, and ChatGPT.

Illegal drug dealers, even doctors (not making a connection between the two) each do this too. Illegal drug dealers thrive when their customers get addicted to the drugs that naturally brings that user back for more product. Doctors who in good faith treat their patients prescribe medicines with the intention that their patients follow their prescriptions and complete their treatment, and then come back for them to help heal or treat their next ailment.

No reputable company explicitly plans or hopes this will happen, yet almost every company and service advertised online (or not) attempts to pull in their intended customer or user, in some form or another. Why? Perhaps Competition itself (as a system-level force) leads every company (an agent-level actor) to take steps to attract and keep their users.

No company plans this; no conspiracy exists between the companies hosting the video feeds and the companies who post videos online. But every company or business arrives at this same behavior, whether the company is an online service provider, a medical professional seeking to treat his or her patient, or the other extreme — a drug dealer looking to sell more product to its users.

Competition does not care whether you compete while protecting your customer’s privacy interests, whether your product or video is good or terrible quality, whether the company operates his business with honestly or through manipulation. Competition is neutral.

But add Selection into the mix, you get unexpected results.

Selection also does not care how you compete or run your business; it is a system-level force — it has no will of its own. But YOU (the agent-level actor) absolutely care if you or your business survives. Selection exerts a pressure to be the one selected; to persevere; to survive; to thrive. Because it is a binary force, as a result of your business practices, you either continue to operate (you are selected), or you are out of business (you are not selected).

As a result of being under the influence of selection, sometimes businesses take actions which would be against their sense of morality or common decency. This happens not because a business is unethical, or the employees themselves intend to do harm to their customers, but because when under the combined system-level forces of Competition + Selection, the need ‘not to be selected against’ + ‘to continue existing as a thriving business, rising above the competition’, under such influences or pressures, it is understandable that the decision-making power and autonomy to make ethical choices sometimes is lost to an overwhelming need to not be selected against.

Even individuals who post articles or videos themselves online — obviously they intend to educate, enlighten, inspire, and excite their audience with the content they produce. But because system-level forces are also involved in their decision-making process, and such forces such as Competition + Selection are acting upon that individual, to keep the user’s viewers interested and to keep them coming back again and again for more content, asking them in each video to “subscribe and like this video,” satisfies the algorithm, to keep your channel and content ahead of others (this is the Competition element); but the Selection element sometimes asks a content provider to do something he or she does not want to do. Sometimes, it asks the content provider to follow punitive and harsh content posting schedules, sometimes it takes over a content producer’s life. Other times, it causes them to do things and to share more private things that might earn them viewers, but that betray their own personal boundaries, causing that content provider to post private content that they later regret, because “the algorithm” rewarded it.

Competition and Selection are only two of possibly hundreds or thousands of system-level forces. But the understanding of what they are, and that they exert and influence over those affected by it explain how a company, or even a person looking for online subscribers who ‘like’ them and their content could overwhelm that person’s ability to make choices, and it could lead them to act and make decisions which they would not have made if they were not under the pressures of these system-level forces.

Reflection For You, My Reader:

You might have come to this article wanting to know more about the arguments for-and-against releasing “open-weight” AI models to the public. You might have already known about AI models, and you came here looking to learn the motivations of the AI companies in deciding whether or not to ‘release the droids’.

But then, here is what you received. If you learned that the surface-level motivations (or even the deeper psychological or market-driven motivations) of a company or an actor might be very different, then you got what you came here for, maybe. But if you learned that there are bigger system-level forces (or system-level intangible entities) that act upon a person, a company, and that these forces or entities change the incentives making certain choices extraordinarily more difficult to make, then you’ve deepened your understanding. But if it occurred to you that there might be system-level forces affecting history itself, and that possibly history and/or time itself might have missing variables that change how external system-level forces embed themselves in the events within time, or that history might have programmed into it an end-point (or an attractor state) that influences points in time that lead to that end-point or intended state, then welcome to MissingVariables.com.

The system-level examples of Competition and Selection are not extreme forces. Depending on the individual under their influence, free will or the ability to make moral decisions does not disappear. The question to be explored in the future: If systems exert influence over an individual’s choices — Can systems constrain freedom without eliminating it? Immediately, the answer appears to be, “that depends on the individual or company under its influence.”

If the system forces of Competition and Selection can constrain the decisions of individual companies without requiring them to descend into conspiracy or malicious intent, then perhaps history itself is governed by similar system-level forces. If so, what we usually call “historical decisions” (those made by Hitler, Stalin, or Mao versus those made by Lincoln, Churchill, or Ghandi) may actually be the visible traces of deeper structures that make some futures easier to reach than others. That possibility—not the release of open-weight AI models—is the real missing variable explored by this website.