OpenAI released hundreds of math proofs this week that sent tremors across the math community. These were not toy proofs cooked up for a benchmark but the solutions to long standing, career defining problems that have profound real world ramifications.
The reaction to this release has been a mixture of awe, shock, and loss. Math has long been a domain that belonged to humans and proofs were something derived from human intellect. The field now faces its Deep Blue moment as machines surpass the best humans. Mathematicians must now reckon with the future of their field and their roles within it.
This is the same reckoning that software engineers have been facing. There is today, almost no (economic) reason to write code by hand. In not even a year, software engineering went from mostly humans writing code with some AI assisstance to AI writing code with some human assisstance. It won’t be long before the little assistance humans still provide will be replaced by better models.
So how do we make sense of this?
In one view of the world, this can feel incredibly disempowering. Having the skills we’ve developed over the course of a lifetime be rendered “worthless” and infinitely reproducible is a hard thing to stomach. For many, this will land like that. It will be devastating.
In another view, this will be like a new renaissance. A blooming of superhuman capability now accessible to everyone. My friends that are amatour mathmaticians are now re-discovering the joys of math that would have been otherwise inaccessible to them. Non technical parents are vibe coding meal prep apps for their kids. Products like ChatGPT Health and Finance are making it possible for a billion people to have access to expertise and counsel that was previously limited to only those that could afford it.
So what comes next?
I think that the implementation of everything is going to be subsumed by AI. The implementation is the means to arrive at a particular end. The act of writing code. Of deriving proofs. Of reconciling financial records. Of optimizing drug molecules.
This pattern of using technology to automate the means of work is not new. Technology is largely this. The printing press obviated the need for scribes. The power loom the need for hand weavers. The digital computer the need for human computers. Each time that certain means were automated, humanity as a whole moved on to grander ends.
When all means of work are automated, the focus will shift to defining what problems are worth solving. The hallmark of a good researcher is not the means of how they do research but the good taste of choosing what problems to pursue in the first place. When building a business using software, the hard part is not the technology but making something that people want.
At the end of the day, what problems “matter” is still a deeply human question because we define what the evaluation criteria are. Technology has shifted human labor from physical work to knowledge work. AI is now making a further shift to redefine work from something that we do to something that we steer.
Our jobs in the coming years are to define what questions are worth asking. This is how AI will transform work and much of how we live. The answers we get will depend on the questions we ask.
Defining which questions are worth asking feels right to me. I'm a non-college guy learning to use AI, and the best results come when I bring a real problem from my own work. The steering part still takes someone who has lived the terrain.