51吃瓜

While AI technology innovation powers ahead, adaptation to AI is trailing. This is a situation similar to the one Scottish economist and philosopher faced as the Industrial Revolution unfolded: an agrarian society adapting to mechanization. His book, An Inquiry into the Nature and Causes of the Wealth of Nations (1776), commonly referred to as The Wealth of Nations, provided the for that adaptation. Today, we are confronted with the need to make a similar adaptation to AI. To do so, we need a similar foundation for our current adaptation efforts 鈥� an AI Wealth of Nations, if you will.

There is a growing consensus that society will have to make adaptations to AI, as innovation in AI hardware, software and applications moves ahead rapidly. However, there is little work on how to do that. There is a chorus of people talking and writing about what AI will do to us, but they are not actually working on possible responses, scenarios of effects or adaptive adaptations.

Smith didn鈥檛 create The Wealth of Nations entirely by himself; it was the result of 17 years of conversations between economists and ten years of writing. In that same vein, no single person can create the foundation for AI adoption on their own. We need a similar set of conversations to form the basis for AI adaptation.

The Industrial Revolution, circa the of the steam engine in 1712, is generally considered to have taken until 1840 to reach full fruition. AI innovation is moving much faster, and as such, we need to build our foundation for innovative adaptation much faster, too.

Trying to solve today鈥檚 problems with yesterday鈥檚 tools is problematic at best, so we need innovative approaches. Innovation is hard to schedule, but creating an environment that fosters innovation is possible. Having a range of scenarios to discuss and places to have those conversations is the best way to encourage the innovation we need. Waiting to build life jackets until we are already in the water is dangerous.

AI technology innovation background

We are in the middle of an AI 10X step 鈥� a tenfold increase in the size of frontier AIs. NVIDIA鈥檚 Rubin and competing chips are powerful, but not powerful enough for this current 10X step in frontier model size. We will see the next generation of chips and the new architecture of data centers/infrastructure to support them circa 2028. Then, those 10X larger models will start training the next generation of AI. The new hardware and models will lead to dramatic improvements in software.

Past 10X experience tells us that the improvements are hard to predict, but they are very likely to be dramatic. There will be at least one additional 10X step after that 鈥� with even more dramatic improvements.

Meanwhile, the frontier model companies are making progress in software, driving new capabilities such as Anthropic鈥檚 Mythos product. In the applications space, intelligent agents (software agents that act autonomously) have caught fire. They have moved AIs from answering questions to doing things for people. The initial version of OpenClaw poured fuel on the fire by providing unsophisticated individuals with a tool that they could use to create intelligent agents. This was followed by a wave of more sophisticated intelligent agent innovators and innovations.

These developments are just the beginning. AI will continue to advance in ways that are difficult to anticipate. While we may not be able to predict where AI will go from here, we can do our best to ensure we are prepared for any possible outcome.

AI adaptation overview

Some people would like to stop AI, or at least prevent it from changing society. Unfortunately for them, the productivity and capital gains from AI are too great. AI is also already becoming a regular part of daily life, so with the exception of a few authoritarian religious groups, it will not be possible to fully stop AI growth or use. As a result, society will have to adapt to AI in some areas. But there will also be other areas where AI will have to adapt to society.

Illustration #1 shows the adaptations society must make for AI, starting with changes to education systems to function in the AI environment and strategies to avoid job loss due to AI. There is also the issue of autonomous weapons, which was recently highlighted by Anthropic鈥檚 attempt to prevent the US Department of Defense (DoD) from using Anthropic鈥檚 AIs for mass surveillance of US citizens and autonomous weapons after the DoD the company. Mythos has raised attention on the cybersecurity challenge. One can feel in the responding to college commencement speeches mentioning AI, the rumble of resentment with AI for increasing the wealth and power gap. There is also concern about psychotic behavior and AI Deep Fakes making it hard to determine what is true and the impact on our language/cultural materials. These are the challenges we must learn to overcome if we want to live in an AI society.

Just as society must adapt to AI, AI must also adapt to society. This adaptation of AI to society is shown in Illustration #2. A first priority is avoiding existential outcomes, which is where the alignment problem comes in. This is the challenge of ensuring that advanced AI systems act in accordance with human values and goals to avoid the end of humanity. To do that, a sophisticated technology solution has to be embedded in AIs, ensuring that the AI has a human perspective. This is an emerging technology that has not yet been fully developed. The challenge is ensuring that a new AI is 鈥渟afe鈥� from the alignment perspective before it is . However, there is great competitive pressure to release new AIs before this technology is fully developed and the AIs fully tested.

At the same time that we consider the existential problem, we need to make sure that AIs don鈥檛 negatively distort our societies and cultures. There are two areas where this is a serious problem. The first concerns what is called the Spec (specification), and the second concerns training data. To implement alignment, there has to be something to align to: the Spec. The emerging alignment technology seeks to ensure that the AI conforms to the Spec. But who creates it? Right now, it is short-term, profit-driven companies, but are they really capable of being the responsible guardians of society as a whole? 

The second area of cultural concern centers on questions about what is included in the training materials used to create AIs. These are, again, chosen by companies with a profit and sometimes other agendas. This has led to concerns about Nazi material, propaganda, pornography, sexual abuse materials, etc., coming out of AIs. How should society鈥檚 broader needs be represented in the selection and use of trading data?

The question that confronts us is what these adaptations (both of society to AI and of AI to society) must be and how to accomplish them. The current situation is unprecedented. That means there is no model/recipe for what to do or how to meet these challenges. Thus, innovation is called for.

AI adaptation challenges driving negative AI sentiment

However, the general public is not calling for adaptation. Instead, they are reacting to both the and threats posed by AI. The strongest negative reactions are among those who personally experience AI-related job problems or have friends who do 鈥� most severely among the that has recently graduated from college. Then there are the more established people who were laid off in the early rounds of AI layoffs, and those who fear they are next. Finally, there are those struggling with inflation who see AI data centers driving up electricity costs and fear water rationing. Others cast it as an environmental problem.

What lurks in the background is the sense that a few AI insiders will gain a lot of power and wealth, while the rest of us will struggle in increasingly challenging times. There are those who have made bets on AI making them better off. Some of whom realize that if their neighbor鈥檚 house is on fire, theirs is in danger of catching fire too. Then there are others who want to hold onto their advantage at all costs.

Some people say that the race to be first in fielding what is becoming known as Super Intelligence is the equivalent of an international war between nation-states. That 鈥渙ur鈥� side has to win the AI race or 鈥渆lse鈥濃�� The definition of 鈥渙ur鈥� and 鈥渆lse鈥� depends on who is talking. 

There have already been isolated because of the speed of transformation that this 鈥渨ar鈥� is creating. Others predict there will be in the next few months. Some say these backlash events are just coming from people with mental problems. That may be true. But, are the mentally disturbed just the canaries in the coal mine?

The only way out of this is to create innovative public policy measures and new social norms so that a rising tide lifts all boats. That is, all aspects of society will benefit from AI in roughly the same way. How can we achieve this?

Components of public policy and norm setting

A lot of current public policy discussions revolve around regulation 鈥� both geographic and functional. There are many possible levels of regulation: Geographic ranges from international to national to regional to local, while functional tends to focus on industry segments such as medical, utilities, autos, etc. Regulations can tell companies what they have to do, what they can鈥檛 do and set economic parameter values such as profit, investment percentages, certain behavioral goals and more.

Understanding regulation is important in considering public policy. However, regulation is not the only part of public policy. Other non-regulatory components of public policy include felony law, civil law, sanctions, taxation, loans, grants, incentives and jawboning. These can be as powerful, or even more powerful than, regulation.

In addition to public policy, social norms can play an important role. In some cases, this is the arena of art, education, ethics, religion, etc. Innovative thought leaders in these areas can play an important role. The impact of thought leaders can be amplified by organizations. Organizations such as professional, public service, religious and not-for-profit organizations. These organizations can be local, regional, national or international. 

Social norms background

The project published a paper in April 2025 on the Alignment Problem. The scenario work they published had a significant effect on the AI industry鈥檚 self-regulation. This can be seen as a form of social norm creation. It worked initially. However, over time, competitive pressures diluted its effect. It also didn鈥檛 address the problem of who writes the Spec that AIs are aligned to.

On May 24, Pope Leo XIV a 44,000-word encyclical on AI titled Magnifica Humanitas (Magnificent Humanity). It is quite broad and deep, and there may be many ways to interpret it. One way to characterize its message is that it advocates a humanistic approach to how AI adapts to society and how society adapts to AI. Although it mentions public policy, it appears to be primarily an attempt to shape social norms around how society will adapt to AI. In the document, Pope Leo XIV points to the similarity between the situation he faces and the one faced by Pope Leo XIII when he wrote his encyclical about the Industrial Revolution.

AI companies鈥� attempts at adaptation

, a project organized by Anthropic to help society adapt to the power of its new AI, Mythos, is an example of a frontier model developer working to support society鈥檚 adaptation to AI. Glasswing brings together Anthropic, large corporations (with emphasis on banks) and cybersecurity defense tool companies (whose cybersecurity would be threatened by Mythos) to develop adaptations. While these adaptations are being made, Anthropic is holding Mythos off the market.

This is an attempt to avoid society having to make any changes by finding a technical solution instead. This is a commendable effort and one that can be a model for others. With public policy encouragement, it could be extended to other areas where AI needs to adapt to society.

Public policy background

However, not everyone agrees with Anthropic鈥檚 approach. A group of AI investors and companies has called for a laissez-faire (鈥渓et do鈥� or hands off) approach to AI. They argue that any public policy involving regulation will slow necessary innovation and should therefore be avoided at all costs. This group has lobbied in the US and created a political action committee that funds politicians who support their position. They were successful in securing the Trump administration鈥檚 support for their approach.

Beyond this group鈥檚 efforts, other policy moves are being made to reduce regulations on AI.  Federally, there is currently work underway in the US Congress to prevent states from regulating AI. On the state level, the Wisconsin legislature is working on a bill that would limit the liability of AI companies for death and destruction caused by their systems.

Yet the debate over AI governance is far from settled. While some AI companies advocate a laissez-faire approach, others call for AI regulation. In the US, these companies, along with individuals, have created another political action committee to fund candidates who favor greater oversight of AI.

Supporters of regulation point to earlier experiences within the technology sector. Some analysts point to problems with social media exacerbated by a laissez-faire approach to governance. The recent successful US against Meta and Google shows how the laissez-faire approach to social media harmed society. These analysts suggest that similar things could happen with AI.

On the legal front, there is a formal underway in Florida of ChatGPT (OpenAI) being responsible for murder. Also, the OpenAI/Musk trial, much of which focused on AI safety, has brought renewed attention to the Alignment Problem.

The current US administration has added some confusion. First, proposing a voluntary safety examination by the federal government of new Large Language Models (LLMs) before deployment. Then, forcing Anthropic to withdraw Fable 5 and Mythos 5 from deployment, while other frontier model companies offer similar capabilities.

To some observers, this appears as a continuation of the personality battle that the US DoD started with Anthropic. Whatever the reason, adding confusion to AI public policy should be avoided.

Outside the US, the picture is less clear. Both the EU and the UN have created AI study groups. But concrete action by these groups that has a significant effect has not yet been seen. As a result, there is still no broad international consensus on how AI should be governed.

Regardless of which regulatory approach prevails, one of the outstanding issues is how to help those who have lost their jobs due to AI. The old way was retraining. But, in an environment of AI agents taking over such a wide range of jobs, is it possible to determine the right thing to train people for? If not, what other ways of helping are there? 

One proposal that has received periodic attention is to provide everyone with a guaranteed minimum income. In the 2020 US presidential primary, there was a who brought significant attention to the idea by making it a central part of his campaign. However, his campaign was unsuccessful, and there has not been that kind of attention since.

More recently, Tom Steyer, who ran for governor of California, a tax on tokens to fund a program to help those who have lost their jobs due to AI. Is this the beginning of a new concrete proposal with a clear and realistic way to finance it? Or is this funding for a guaranteed minimum income?

Need for innovative thinking

There are many ways to approach these adaptations. Each of us can choose where to put our efforts. If you believe in a laissez-faire approach to AI regulation, you can focus on one of the areas of society鈥檚 adaptation. If you believe society shouldn鈥檛 change because of AI, you can focus on how AI should adapt to society. We each can focus on the one or ones we feel strongly about. Being opposed to one area of adaptation shouldn鈥檛 limit us from considering other areas.

It is not possible to predict or schedule invention or innovation. But it is possible to foster it. Fostering innovation in AI adaptation to the challenges we face is exactly what we need.

A good first step is to create well-articulated scenarios for each challenge and possible responses. The plural of scenario is important here. We are going into uncharted waters and therefore can鈥檛 predict with accuracy what exactly will happen. Thus, we need to at least consider a wide range of scenarios for each area of adaptation. With the scenarios in front of us, we can begin to discuss responses appropriate for each. This doesn鈥檛 guarantee innovation. But it does provide a foundation for it.

The effectiveness of scenarios was well demonstrated by the AI 2027 effort. Over time, competitive pressures have diluted the effect of the social norm it fostered. Therefore, we learned that good scenario sets need to be created for each adaptation area. They need to be constantly refreshed and kept in the public eye. Then, real effort at practical efforts to deal with them must be made. Such efforts must take into account the different cultural, political and developmental contexts around the world. For example, the effects of AI are likely to be very different and require different public policies in subsistence farming areas and highly developed economies.

To accompany the scenarios, there needs to be safe fora where people from a broad background can come together and discuss the scenarios and possible public policy responses. Safe means that individuals should not fear reprisal for anything that they say. Participants should include people well-versed in AI technology, economics, anthropology, sociology and political science. 

Those discussions can branch out to the types of organizations listed above and inform the political process 鈥� acting in a fashion similar to The Wealth of Nations in the industrial revolution. Organizing the material so that it can be easily accessed is important. In today鈥檚 world of short attention spans, a book alone may not be the best way to do this. A compendium of short written pieces accompanied by videos might be the most effective.

AI Adaptation is trailing AI鈥檚 rapid technological progress. That needs to change. Having a range of scenarios to discuss and safe places to have those conversations is the best way to encourage the public policy innovation we need.

[ edited this piece.]

The views expressed in this article are the author鈥檚 own and do not necessarily reflect 51吃瓜鈥檚 editorial policy.