Google DeepMind AI, a pioneering entity in the field of artificial intelligence, has utilized a large language model (LLM) named FunSearch to solve an unsolved complex mathematical problem that has stumped humanity for years. This achievement could potentially usher in a transformative phase in the realm of AI.
The mathematical problem in question is the “cap set puzzle,” a longstanding enigma that essentially revolves around the maximum number of dots that can be placed on a page, connected by lines, without ever having three dots aligned in a straight line.
While this might sound perplexing, the key takeaway is that this problem has remained unsolved for years, with researchers only managing to find solutions for smaller dimensions. That was until DeepMind AI FunSearch came into the picture.
FunSearch astoundingly discovered new structures for large-cap sets, surpassing the best-known solutions to date. Although the LLM didn’t completely solve the cap set problem (contrary to some sensationalized news headlines), it did unearth facts previously unknown to science.
The researchers, in their paper published in Nature, stated, “To the best of our knowledge, this shows the first scientific discovery – a new piece of verifiable knowledge about a complex scientific problem — using a DeepMind AI LLM.”
In past experiments, large language models have been employed to solve mathematical problems with known solutions. FunSearch operates by integrating a pre-trained LLM, in this case, a version of Google’s PaLM 2, with an automated “evaluator.” This evaluator serves as a fact-checker, preventing the generation of false information.
LLMs have been known to frequently produce “hallucinations” — essentially fabricating information and presenting it as fact. This has understandably limited their utility in making verifiable scientific discoveries. However, the researchers at the London-based lab assert that the inclusion of an in-built fact-checker sets FunSearch apart.
DeepMind AI FunSearch operates through a continuous interaction between the LLM and the evaluator, transforming initial solutions into new knowledge. An additional promising feature of this tool for scientists is that it provides programs that elucidate how its solutions are constructed, not just the solutions themselves.
The researchers expressed their hope that “this can inspire further insights in the scientists who use FunSearch, driving a virtuous cycle of improvement and discovery.”
Significance of Google DeepMind AI
Scientific Discovery: FunSearch’s achievement marks the first instance of a Large Language Model (LLM) contributing to a scientific discovery. It has unearthed new facts about the cap set problem, a notorious mathematical problem, which were previously unknown to science.
Advancement in AI: This breakthrough showcases the potential of Artificial Intelligence in solving complex problems. It could herald a new era in AI development, where AI models are not just tools but collaborators in scientific research.
Methodology: FunSearch uses a unique approach of combining a pre-trained LLM with an automated evaluator. This methodology helps guard against the production of false information, a common issue with LLMs.
Transparency: Unlike traditional models that only provide solutions, FunSearch outputs programs that reveal how its solutions are constructed. This transparency can be extremely beneficial for scientists as it provides insights into the problem-solving process.
Potential for Future Research: The success of FunSearch can inspire further insights in scientists, driving a virtuous cycle of improvement and discovery. It opens up new possibilities for using AI in various fields of research.
DeepMind AI discovery signifies a major step forward in the integration of AI in scientific research and problem-solving. It demonstrates the potential of AI not just as a tool, but as a collaborator in the quest for knowledge.