August 23, 2026
The problem is in the name, Artificial Intelligence. We’re using a word “Intelligence” without a universally agreed upon definition. So how can we build something that we don’t understand or can’t define?
One way we’re currently measuring progress is with benchmarks. The problem with benchmarks is that they are just moveable goal posts. Once a benchmark has been satisfied, a new one will pop up and so will the goal posts. This is the problem with relying on benchmarks rather than a philosophy and theory of intelligence.
Another measure is with money. OpenAI and Microsoft have defined AGI as reaching $100,000,000,000 in profit. If that is to be believed, then several companies have achieved AGI. Obviously this is not a helpful definition but it shows you just how meaningless AGI is.
Finally, a biologically inspired project I check in on rather infrequently is OpenWorm. The project aims to simulate a small roundworm with 302 neurons called Caenorhabditis elegans or C. elegans for short. Despite only having 302 neurons and around 1,000 cells total, a full simulation of the roundworm remains elusive. This project showcases how much work is still ahead of us in terms of understanding life. I do consider C. elegans demonstrating intelligence.
I spent an hour today hoping to reach for this “AI” via “State of the Art” LLMs and prompted the LLM to translate a relatively obscure language to another obscure language. What I got was a confident answer with a resulting file, but when I ran it I got a compiler error. I prompted with the error and got back another file with more compiler errors. I prompted with the new errors and got back a compiling file with a crash. The LLM then outputted confidentally worded solutions to the crash. After the 10th try I gave up because I saw it had entered a non-progressing loop. This was in 2026, I’ve seen this exact same behavior in other scenarios in 2025, 2024, etc. To me this isn’t progress, this is a fundemental flaw with the technology that can’t be solved by throwing more money at the problem.
Last I checked, the over $1,000,000,000,000 spent on AI in 2026 wasn’t going towards coming up with a theory of intelligence. Maybe it doesn’t make a good headline, or it’s a self-defeating headline because that would admit that the money is going towards something that we don’t understand. My take is that we should look towards the philosophers and scientists to get their honest feedback. Maybe some sort of conference where philosophers and scientists can get together to speak openly and without consequence about our lack of understanding.
We don’t have a definition or theory of intelligence. This was true 20 years ago, and this is true today.