Source: Bankless Podcast

Organizer: Felix, PANews

MIT economist Christian Catalini appeared on the program with Ryan and David to delve into his new paper "Some Simple Economics of General Artificial Intelligence." The paper points out that the scarce resource in AI economics is no longer intelligence, but rather validation: the capacity of humans to examine, judge, and confirm the correctness of AI outputs.

Christian elaborated on two cost curves that are reshaping industries (automation costs and validation costs), explaining why entry-level job positions are the first to disappear, and why even top experts are inadvertently cultivating their successors ("the coder's curse"). He also depicted three roles that could remain in transition: director, meaning creator, and accountability taker.

PANews has summarized the highlights of the conversation.

Host: I think many listeners might feel a panic about AI similar to mine. Why do you think people are worried about AI? Is their concern reasonable?

Christian: We all resonate with that. This is a rapid and transformative time, and the closer you are to code, the sooner you might witness this acceleration, which has become very real over the past few months. This technology has accomplished things that many thought would take longer to achieve; this feeling is something we are all trying to cope with. But I think the "apocalypse thesis" is wrong, and people often underestimate the potential these tools bring. Yes, there will be an extremely difficult transition period, and the pace of job transformation is unprecedented in history. Nevertheless, if you harness the most notable features of this technology and invest in it, it’s largely a positive thing in the long run, even though it will be bumpy along the way. Economics views work as a collection of tasks, some of which will be automated, which is good news, but the key is how you retrain yourself and stay at the forefront.

Host: Who do you think will be hit first?

Christian: This is an excellent question, and I have many different thoughts on it. First, when I say that those closest to the code are hit first, I mean they are the ones who will first experience how powerful this technology is. As the "Jevons Paradox" reveals, when something becomes efficient, we tend to consume more of it; for example, we will write more software. I think programming will become specialized, much like many other professions; we refer to this in the paper as the "disappearing junior loop." If you are a junior who has not yet acquired the "tacit knowledge" to distinguish excellent products from mediocre ones, then AI can replace you well across various fields.

Everyone can now easily access a rather good marketer, junior programmer, or a lawyer who can help you with most situations; you just need to consult a top lawyer at the end stage for final validation. On the other hand, even top experts, while integrating AI, are creating tags, information, and digital traces that will ultimately lead to the automation of their own jobs. Leading labs are hiring top talents from fields like finance to help create assessment standards, embedding that specialized knowledge into large models. So I believe no single job is 100% safe; even manual labor limited by robotic capabilities will see significant leaps in reward models in the coming years. Anything that happens on a screen can be tracked, replicated, and learned. For every profession, the key is to consider: if I delegate as much work as possible to AI, where else can I add value?

In reality, people have a lot of "self-soothing" about "taste" and "judgment." They are very vague. So, in the paper, we say: there is no such thing as taste or good-bad judgment, only the distinction between "measurable" and "immeasurable." If something has been measured, machines can replicate it. If something is still embedded in the weights of your brain, like a top designer who has accumulated thousands of hours of experience in deciding what should be released and what shouldn’t, that is what we call "validation." All validation is that final step: AI agents create the product, and you, as the decision-maker, judge whether it meets the standards for market release. As machines acquire better data, things will get automated; but when faced with unknowns or places where there is fundamentally no data, that part will still belong to humans in the coming years.

Host: This is a very profound insight. But I am also thinking, it’s natural for engineers to automate their own work. Are the impacts on every industry the same?

Christian: We have enough evidence to show that the changes will be uneven. You can think about it this way: Is this job just a "packaging" of something that society fundamentally does not need? For example, general consulting work, if it mainly re-packages, refines, and summarizes information that is already widely available, then there are obvious risks. But if it brings valuable domain expertise that is scarce or is needed for political reasons, those will survive. Ask yourself if this profession is profitable because it solves a complex problem or simply because there is an artificial bottleneck.

Host: What does validation really mean? I find it difficult to break down my day's work into which is cognitive work and which is validation work.

Christian: The agent has learned and measured everything from the web and books, and because they are cheaper and scalable, they will replace the measurable parts. But what the agent still does not know is the unique neural network weights in your brain. These are acquired through your own experiences and struggles that make you a top expert. For instance, early adopters of cryptocurrencies, many from Argentina, Venezuela, etc., who have lived through hyperinflation, react to assets completely differently. This intrinsic unique measurement is still a huge advantage.

What is validation? It is the difference between your own measuring standards of the world and the standards that the agent possesses. Like a top editor, who knows exactly what articles will resonate; or a top CTO, who knows which crucial edge parts of the massive codebase generated by AI must be checked by humans, as that part cannot yet be measured by machines.

Host: Let me give an example. If I see a video on X of Israel being bombed, but I find out it is AI-generated. I use my brain to identify the problem and may prompt to generate a better video; is this my "validation capacity"?

Christian: That’s a great example. Further, we might soon be in a world where, for most people, this video is indistinguishable from reality. The next step could be that military experts notice the dynamic of flames is off. The step after that might be that even military experts can’t tell at a glance and need AI to analyze physical principles and conduct simulation tests. Ultimately, it may become completely indistinguishable; by then, we will have to rely on cryptography-based infrastructure to confirm authenticity. The same goes for medicine, where edge cases will ultimately need top radiologists with 20 years of experience and an understanding of the patient’s specific context to override AI's judgment. That is the last thin "filter stage" we are focusing on. When we do this, we free up a lot of time. So, that’s the good side. We can do more with fewer resources. The costs of expensive things will decrease. Society as a whole will consume more of these things. I think that’s good news.

Host: But in your example, currently the expert is doing the validation, but soon he won't be able to validate, needing military commanders, and eventually even commanders can't validate and have to rely on AI. Doesn't this illustrate that "validation" initially has value, but will soon also be automated by AI? So even "validation" itself is not safe?

Christian: Precisely. We call it the "coder's curse" in our paper. The very rational act of validating is itself driving the development of cutting-edge technology and datafying experienced data. We can't stop, because all lawyers or practitioners are trying to use AI. Validation is indeed a diminishing frontier.

Host: The area of final validation work is shrinking. When can we stop feeling anxious?

Christian: First, some things are inherently unmeasurable, like the so-called “status games” or things that humans ascribe meaning to. These areas will not be invaded by machines because their characteristics involve coordination and consensus among humans. Cryptocurrencies are somewhat similar; what's important is the consensus among humans about what is valuable. As the area of measurable work shrinks, we will invent many methods to make immeasurable work meaningful.

Host: AI can build a website in 10 seconds but may not be able to write an attractive tweet for humans. Could this be one of the remaining validation tasks?

Christian: Captivating attention and telling a genuinely novel joke are extremely difficult creative tasks that attempt to break through what has never been measured. We have evolved a strong ability to cope with unknown environments over a long course of survival. People engaged in this work are called "meaning makers." In fields like art or culture, what is good depends on human consensus. Even when you use AI agents, you still must set "intent."

Host: Automation costs are decreasing exponentially; what about "verification costs"? Will they forever be constrained by human biological limits?

Christian: Currently, they are constrained by biology. So many companies have released large amounts of AI-generated code, but there simply aren’t enough people to read and verify it, which inevitably hides risks.

Host: Can't we use AI to validate AI?

Christian: If AI can validate correctly, then that part itself can be automated. After exhausting all AI verification, what remains is truly what cannot be validated by AI, which is the bottleneck for human intervention.

Host: If verification is a new scarce resource but continually recedes, how should one work and invest in this economy?

Christian: We created a 2x2 matrix based on "automation costs" and "verification costs." The lower left corner represents the replaced workers: automation is easy, verification is easy, and you absolutely don’t want to be here. The other three quadrants are:

Meaning Makers: Automation is difficult, verification is difficult. They are committed to social consensus, status games, and human connection. For example, those creating taste in the fashion industry, or crypto KOLs on Twitter, who create narratives and coordinate attention.

Accountability Takers: Automation is easy, verification is difficult. They are top experts in their fields, like elite lawyers, doctors, or venture capitalists. They leverage AI at scale but provide accountability and validation services for edge cases.

Directors: Automation is difficult, verification is easy. The core is "intent." They deal with "unknown unknowns," directing agents like entrepreneurs, setting direction, sensing deviations, and continuously correcting course.

Host: What should young graduates do when they want to enter the job market? On one side are entry-level jobs that have no value, and on the other are top experts that take ten years of industry honing to become; there is a huge gap between the two. AI can already handle entry-level tasks; how can young people grow to the other side?

Christian: The gap indeed exists. But the good news is you can compress learning time. You can skip traditional training steps. A junior engineer can now accomplish the work of a former team alone with tools. Although there may be mistakes at first, as newcomers, they can question traditions from an entirely new perspective, which is an advantage. They can achieve ideas in ways that we could never do when we were young. There are pros and cons.

The past path of "getting a degree, finding an internship, working hard to advance" truly no longer exists, which will create a huge cultural shock. This is very difficult for fresh graduates. If you're still in college, you still have time to clarify direction. If you find yourself in a bind, my suggestion is to use these tools to create something. Your ambition should be 100 times greater than what we had at that age.

Host: Will the disappearance of a large number of "button-pushing" jobs cause chaos in society in the short term?

Christian: Society will always recreate "button-pushing" jobs when needed to maintain stability. But many people in such jobs actually have the capability to do more; they were just constrained by the environment in the past. When physical labor is no longer necessary, we invented going to the gym; nowadays, facing the liberation of mental labor, people will develop various side gigs and creator economies to seek challenges. This is also why I think "universal basic income (UBI)" is entirely wrong; humans need meaning and the drive for self-realization. Moreover, even if a significant part of your work is automated, if you make good use of AI as a super tool, even a freshly minted entry-level employee can deliver outputs previously achievable by an entire team.

Host: Any advice for companies and investors?

Christian: For companies, invest in verification infrastructure, providing "responsibility as a service" (meaning not only providing agency services but also ensuring accountability for consequences). Additionally, grasp "exclusive fact sources," as AI is easily deceived, and companies that can provide exclusive real data or in-depth assessments akin to Bloomberg are highly valuable. For investors, besides investing in these areas, focus on "immeasurable" hardcore R&D. Previous ordinary network effects may fail; new network effects will be built on how you make your agents more reliable through better real feedback since what people truly want to buy is validated intelligence.

Host: Is cryptography useful in this verification process?

Christian: The underlying infrastructure established in the past decade in the cryptography field is crucial. When we need to ascertain the authenticity of identities and prevent account takeovers, on-chain technologies like "personality proof" can provide powerful validation. Additionally, data provenance and cryptographic regulatory chains need to ensure hard encryption guarantees concerning the generation of information and compliance of models.

Host: What should people do in the coming year? Are you optimistic about the future of humanity?

Christian: First, don’t panic. Experiment extensively, utilize tools to "淘汰" and automate your current self. Many hobby explorations of the future might be the most meaningful careers. At worst, you can identify the boundaries and shortcomings of the models. For many online creators, hobbies have transformed into careers; this will be the mainstream direction in the future. If you have children, uncover their talents and immerse them in their passions; that is the most important. There are no fixed professional templates; new AI tools can better assist you in finding that unique path that belongs to you.

Related reading: Night Read | Conversation with Silicon Valley VC Bill Gurley: Don't ask for stability, become a "AI-enabled" version of yourself

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。