Not a general intelligence
The reason why an AI language model-based system like the AI Scientist cannot currently create meaningful novel research on demand is because LLMs' "reasoning" abilities are limited to what they have seen in their training data. LLMs can create novel permutations of existing ideas, but it currently takes a human to recognize them as being useful, which means an autonomous system like this (with no human in the loop to recognize and improve upon ideas or direct its efforts) doesn't work with current AI technology.
As Google AI researcher François Chollet recently put it on X while discussing an unrelated paper from 2023, LLMs are terrible generalists. "LLMs (used via prompting) cannot make sense of situations that substantially differ from the situations found in their training data," he wrote. "Which is to say, LLMs do not possess general intelligence to any meaningful degree."
To be fair, Sakana's researchers acknowledge some of these limitations themselves: "While the current iteration of The AI Scientist demonstrates a strong ability to innovate on top of well-established ideas... it is still an open question whether such systems can ultimately propose genuinely paradigm-shifting ideas."
The answer, right now, is most likely "no." That may change in the future, but such capabilities will be based on hypothetical technology that does not yet exist.
Promoted Comments