Meaning & Morality of Work

It is a peaceful Sunday morning in the Zürich valley. I am waking up to the sound of the church bells, making breakfast from the few leftovers since the supermarket is closed, and I am drinking my morning tea from a package that some previous tenant left in this house. I came to the realization that two habits have stably improved the quality of my well-being over the last two years: connecting with people and writing my thoughts in a blog. Both have been a source of joy, but also a source of grounding. I have become way more open about my worries and I share them in conversations quite often (which I have to be careful with, since that requires a sufficient amount of trust) and when a narrative stays in my mind for a longer time I really like to put it on a piece of paper to help me deal with it. My conversations have been lately circling around just one topic: artificial intelligence.

As if it was not making headlines enough, right? I do believe that for many, their relation to AI has not changed throughout the last two years. It is easy to miss the developments if all you see is a chat window in your browser and it gives a slightly better output than before. Without going much into it, a lot has changed. AI went from answering simple questions in November 2022 to an absolute problem-solving beast by the start of 2026. Millennium problems, the most difficult mathematical problems, are now being solved thanks to them. Young people are usually among the early adopters, so I have seen people adopt the technology, and we are all stunned. From the press the impression goes that the researchers inside the companies are even surprised about these things. They themselves now use AI to create new AI, cause the models have just become so helpful. Repeated critiques of the models not being able to do a certain task, like ARC-AGI, have turned out to be just targets to be swept away. The amount of capital backing the whole AI development is unprecedented.

Speed of Improvement

So this is the state of affairs at the end of September 2026. The reason why I bring this up is very much personal. I have extended my studies by doing an internship and now I am starting my last semester and writing my Master’s thesis. I look online, and there is an insane explosion of science and materials to be read. One of the main AI conferences, ICLR, has received more than 60 000 submissions this year, around 6x compared to two years ago. The smart LLMs maybe do not drive the creation of the idea (yet), but the implementation and the writing can be greatly accelerated. Science is seeing great gains from the invention of LLMs (here note that LLMs are a product of science, back to that later).

Or losses?

I would not be the only one to argue that the developments are too fast to keep up with for the human brain. Traditionally, the education needed to be a software engineer would require 3+ years of tertiary education, while we have seen programming being taken over by large language models in the last 8 months. If the developments become even faster, the time frame for adaptation of humans might shrink to the level of absolute impossibility.

At this point in the blog, you might be thinking, Michal has completely lost his mind. And you would not be wrong. I did spend a lot of time talking about this topic to people, and there are some common counterarguments.

Counterargument 1: This is only coding, the models are specialized for it, they cannot do anything else.

Yes, but it seems like there does not exist a domain which you cannot express as code (as I try to argue here). You might say that the models are overfitted to software engineering, but actually they seem to be doing great at tasks like CAD design too (see CADBench). This is just one of many examples. The more data (or verifiable environments) there are in a domain, the better the performance will be.

Counterargument 2: The models still need to be steered, they cannot do anything without humans.

Yes, but the models are becoming more and more autonomous. If the charts of METR extrapolate even very conservatively, we are talking days or weeks of jobs with high success soon. If one means the question of alignment, then yes, aligning the models to human behavior might be quite difficult. My early professional experience seems to show me that an employee is more than just work being done, but also the full alignment with the goal of the organisation. That makes a difference between an employee and a freelancer.

Counterargument 3: This is only the virtual world, in the physical world the models cannot do anything.

Unless the response to counterargument 1 is true, and they will suddenly generalize to this domain. Astra, the latest model from OpenAI, seems to be doing quite well in the physical world. I recommend checking the collection of links on GitHub. It is quite surprising how well it works. Yes, it is slow, and it is costly, but if you extrapolate even very conservatively again, this technology has a massive potential.

If the speed of the developments in other areas outpaces human learning, we might have a problem… but I think there is a deeper issue even.

Moral Obligation to Work

The models are entering all areas of our life. What concerns me the most is how they will shape our incentives. Japanese ikigai, or reason for being, is a conceptualization of what it means to live a fulfilling life. It motivates one to search for occupation in the intersection of what the world needs, what you are good at, what you love and what you can get rewarded (paid) for.

The question is: what happens to the meaning if there is nothing a human can do that the world needs?

The developments are too fast to keep up with. We are now used to delaying the reward of contributing by studying for years, to be able to contribute to ever-more-complex society. What if the models can learn it faster than we do, leaving us nothing to contribute?

Recently, a guy on the street asked for a donation, cause he is travelling to Nepal to teach children English. I asked him how the journey was so far, and he recalled an interesting experience. He was just scolded by a woman on the street: she was older than him and she told him: “What are you doing with yourself, you are so young and healthy, why don’t you go and work and do something useful for the world.” For centuries, work was something that was needed for society, and there was a moral obligation to do something. We all had an internal calling to be productive members of society, and that was also rewarded. Older generations might be asking why young people want to be influencers and not some other traditional occupation. But what is the worth of work in a world where there is nothing to contribute? Or maybe, it takes 50+ years of study to create a marginal improvement?

Let’s face it, if you are going to be teaching English in Nepal, you are not doing it for the kids, you are doing it for yourself. It would be 1000x more efficient to send the money there, but you would not get the experience. That is not to criticize the traveller, it is to highlight that the value of the human experience might be more important than the work being done. In the same way, why should a human write code, if it costs 30 EUR/hour when a machine can do it at 0.05 EUR/hour?

(…)

By the same token, how does the old lady justify her statements now? Is there any moral justification for work if a machine can do it instead?

(…)

Or for studying? Is there any moral obligation to be intelligent?

(…)

Going further, what are our callings going to be like? In a society of such a speedup, how will humans align themselves to find a meaning in life? I mentioned before that AI is being used to create more AI as we speak (aka recursive self-improvement). This promises an automatic speed-up of creation of new technologies, but that also means that humans might not even have an idea how those technologies work, something we already see in mathematics…

Work as Art

I have been chatting with some PhD students here at ETHZ. If any one of them could have made the discovery in mathematics that ChatGPT has made, they would have made a career for life. Math will still be done, but the value might shift to understanding, rather than creating. Terence Tao, a leading mathematician and educator, writes in his blog how humbled he is by the current developments. Math has always been a domain for people that have been able to grasp new concepts fast and work hard to develop new intuitions. Now suddenly, there is something that is smarter than all humans, and can create material for humans to understand. (Same way I think interpretability of the AI models will be crucial, cause it is essentially understanding of AI.)

And if you happen to do mathematics, you will not be doing it cause of its utility. The math we need can be done by the machine. If you want to do mathematics, you do it to enrich yourself and others. Just like art. You share your deep internal experience with others, and enrich them by that. The role of a teacher might be to transmit powerful ideas, more than to teach hard skills.

The access to knowledge will be democratized. Right now you can hire Claude, the analyst/programmer/composer/designer/maintainer/manager/scheduler/(…) for 20 EUR/month (note: not 20 EUR/hour as for humans!). But just as Spotify did not remove artists from the world, but rather gave them a platform, this will not wipe out people who think. However, true human thinking will become a premium commodity not everyone can develop and have. Just like playing the violin and painting on canvas. There were days when each event needed someone who could play an instrument to create background music, now there are speakers that run in the background of every party. Maybe in the future thinking also becomes a backgrounded job that the computers will hum out for us. The stage will be given to the few that create new ideas, for computers to pick them up, re-mix them and play them back at higher decibels.

What a time to be alive?

The times we live in are anything but ordinary. I would have bet everything I own against any prophet that would have told me about this future. As always, I will leave this on a personal note of what I think will be valuable as we enter the post-human-intelligence era:

• Sharing human experience. Self-expression, connections and sharing with others. Making others feel what you feel, and creating alignment in society through politics and so on will be more important than ever.
• Research/entrepreneurship in overlooked areas. The gains from AI will be the biggest in areas which are now understaffed. The research directions that no-one has time to think about cause there is no funding.
• Arts & crafts. It is time to stop being a worker, and time to become an artist. Discipline is less valuable than creativity, and exploring new approaches brings more than making the old ones more efficient.
• Well-being. Taking good care of your physical body and mind. Not forgetting about oneself as we think about other things.

This might be the most remarkable time to be living. We will have to redefine ourselves as work stops having the same shape. Making sure that the value of human life is preserved and that the benefits are distributed to all of us will also be of great value.

Out of nostalgia, I am commiting and pushing this post by hand. Won’t let Claude do it this time, before I send this message out to the web.