![\"Albert](\"https:\/\/www.europesays.com\/ai\/wp-content\/uploads\/2026\/08\/1785981920_768_0x0.jpg\")
<\/p>\n<\/p>\n
Albert Einstein would intuitively appreciate the limits of AI<\/p>\n
getty<\/p>\n
There\u2019s a missing link in artificial intelligence. It is not capable of making that leap from number-crunching to actual original or novel thought, spurred by real-world circumstances.<\/p>\n
That\u2019s the view of Tom Zahavy<\/a>, discovery team co-lead at Google DeepMind, in a new paper<\/a> that points out that AI is incapable of making original discoveries on its own. While the paper is meant to discuss the shortfalls of AI for scientific discoveries, it has implications for AI\u2019s creative powers in business settings as well. <\/p>\n Zahavy dissected the process that went into Einstein\u2019s formulation of his General Relativity Theory, noting that the scientist\u2019s thinking evolved through three phases \u2013 deduction, induction, and finally, abduction. <\/p>\n AI has mastered induction, based on statistical pattern matching, and is rapidly conquering deduction, which is formal proof, Zahary illustrated. However, it doesn\u2019t clear a third of three hurdles, which is abduction, which is \u201cthe generation of novel explanatory hypotheses.\u201d<\/p>\n If scientific discovery were merely the sum of deduction and induction, “modern large language models (LLMs) should theoretically be capable of inventing theories like General Relativity given sufficient compute.\u201d<\/p>\n Einstein\u2019s discovery process was outlined in his letter to Maurice Solovine, where he \u201cconceptualized discovery as a cyclical process involving an intuitive \u2019jump\u2019 from sensory experience to axioms, followed by logical deduction,\u201d Zahavy explained.<\/p>\n Generative AI is capable of absorbing information, and then making logical deductions about the implications of that information. But it can\u2019t tell you why the information it has assembled for you is new or exciting.<\/p>\n The three stages of discovery include the following:<\/p>\n Deduction (rule + case = result): \u201cThe analytic application of a rule to a case to predict a result. It is the only mode that guarantees truth, e.g., executing code to verify output.\u201d Induction (case + result = rule): \u201cThe synthetic derivation of a rule from the accumulation of cases and results. It validates hypotheses through statistical frequency, e.g., generating a function to satisfy unit tests.\u201d\\Abduction (rule + result = case): “The inference of a case or a new rule to explain a surprising result.”<\/p>\n The bottom line is LLMs are great tools for assembling information, but it still takes humans to add new dimensions or perspectives to what the output is telling us. <\/p>\n Again, there is a connection to any original thinking that we may hope for from AI in non-scientific settings. A separate study<\/a> out of Duke University questions the creative output of LLMs in a different sense \u2013 the conclusions tend to be homogenous. \u201cThe creative outputs of commercial LLMs are more similar to each other than users might hope. When challenged with three standard tasks assessing creativity, answers from commercial LLMs are much more alike than their human counterparts.\u201d<\/p>\n