General Intelligence is Explainable Falsification

October 22, 2026 ·

Abstract: the idea that progress in research and development (R&D) happens via step-by-step, induction-based pattern-matching, where hypotheses are positively or negatively confirmed via empirical observation, has long been philosophically outdated; yet it remains widespread as the innate underlying assumption in scientific research, in the creative problem-solving methodology generally taught, and in many aspects of daily life, both at individual and institutional levels. A better epistemic approach to R&D-based problem-solving may be explainable falsification: finding deduction-based counter-examples to conjectures while protecting the error-correcting mechanisms around them. This subtle architectural change enables individuals and institutions to develop knowledge and real-world artifacts that gain from unpredictable and uncertain circumstances, as opposed to being fragile and vulnerable to them.

The problem of induction

Philosopher David Hume in the 1700s formulated what's commonly known as the problem of induction, in which he argued that no number of positive confirmations can justify a belief, yet it is this very mechanism through which we form our impressions. While discussing this problem of induction, logician Bertrand Russell in his 1912 book The Problems of Philosophy concluded that while we cannot prove the future will resemble the past, induction is an essential and unavoidable foundation for scientific reasoning. Karl Popper, in the mid-1900s, then proposed a breakthrough alternative: instead of trying to prove theories true via induction, science advances via deductive reasoning, what he called falsification. Popper claimed that the process of knowledge-creation and creative problem-solving works through a mechanism of conjectures and refutations, in which problems are solved creatively by making bold claims, called conjectures, that can be tested and refuted (falsified). Popper asserted that a theory can never be proven true, but that a potential theory, a conjecture, can be proven false by a counter-example refuting it.

Current status of the problem

In popular culture, high-impact observations violating past induction-based theories and models are referred to as black swan events, a reference to Nassim Taleb's 2007 book The Black Swan and to Hume's problem of induction. Taleb applies Popper's deductive problem-solving approach to risk management, aiming to capture the upside of real-world uncertainties that cannot be modeled upfront. Physicist and father of quantum computation David Deutsch is another key modern intellectual popularizing Popper's deduction-based approach, arguing that good explanations and error-correction mechanisms are the foundations on top of which infinite progress can be achieved, most importantly bypassing the problem of induction in the process.

Real-world example of refutation

The intelligence quotient (IQ) is a metric derived from standardized testing that aims to assess, measure and classify human intelligence. IQ is commonly cited as a basis for predictions on a variety of topics: social and economic status, mortality and nutrition, job and life performance, among other overarching aspects of life; all of which assume an underlying mechanism of induction: that past measurements and patterns are proof of future outcomes.

Nassim Taleb and David Deutsch, both building on Popper's deductive approach, disagree with this way of measuring and predicting intelligence. Taleb's refutation questions the assumption that IQ operates on a so-called bell curve, also called the Gaussian or normal distribution after the mathematician Carl Friedrich Gauss. Two underlying assumptions of a Gaussian distribution are that its samples must be independent of one another, and that they must have limited variation. Taleb, citing the highly interconnected and fat-tailed (extremely variable) nature of intelligence, refutes the conjecture that intelligence can be reliably measured and predicted by IQ metrics that assume a Gaussian distribution underneath. David Deutsch meanwhile takes a different approach and uses the idea of Turing completeness (the capacity for universal explanation) to refute IQ as a reliable measure of intelligence, claiming that memory and speed are the only limitations once humans are capable of general creativity, and thus rejecting the concept of intelligence having higher and higher levels.

Future potential and suggestions

Karl Popper's deduction- and falsification-based approach to knowledge-creation and creative problem-solving, through the mechanism of conjectures and refutations, is key to differentiating artificial intelligence from general (human-like) intelligence. While large language models are extremely useful at pattern-matching and outputting pre-existing, induction-based knowledge already available to us, the key to creating new and genuinely creative knowledge is deeper engagement with the idea of falsification. Popper's solution to the problem of induction can be thought of as the Socratic method for the information age: it offers a toolkit to distinguish artificial from general intelligence, a great potential upside for American R&D.