Matjaž Leonardis - Science, Identity and Probability

Critique of the "Scientist" Identity:

• Machos Leonardis argues that the social identity of "scientist" is a counterproductive early 19th-century invention that hinders natural inquiry.
• He contends that focusing on "what is science" creates a false belief in a single, rigid method rather than allowing inquiry to follow the logic of specific problems.
• Universities remain useful as institutions that aggregate resources, literature, and communities, but this function does not require a unified definition of "science."
• Leonardis compares the current fixation on "scientists" to a hypothetical fixation on "entrepreneurs," noting that while people build companies, they rarely debate who is truly "entrepreneuring."
• The interviewee suggests that self-consciousness regarding one's methodological role can lead to unproductive constraints on problem-solving.

The Popper-Miller Theorem and Induction:

• Leonardis and David Deutsch co-authored a paper re-examining the 1983 Popper-Miller theorem, which posits that probabilistic support is not inductive.
• The theorem argues that while evidence may increase a theory's probability, this increase is deductive rather than inductive, effectively negating the idea that evidence confers inductive support.
• The paper challenges the use of Bayesian reasoning in building AGI and other predictive models by exposing the limits of probabilistic support.
• Popper's view that probability measures "logical content" is critiqued, as people naturally prefer informative, explanatory theories over contentless tautologies that are mathematically more probable.
• Leonardis notes that the "conjunction fallacy" (e.g., the Linda problem) supports the idea that humans prioritize explanatory depth over statistical likelihood.

The Role of Universal Theories:

• Leonardis attributes the human drive for universal explanatory theories to a psychological need for regularity, though he remains skeptical about the exact mechanism of this need.
• These theories enable progress by forcing predictions that can be falsified, leading to knowledge accumulation that exceeds simple experience.
• He notes that "universals" allow truth to accumulate in ways that direct observation alone cannot, but he disputes Popper's psychological explanation for why this need exists.
• The interviewee highlights that probability in many languages translates to "believability," suggesting a semantic confusion where explanatory power is often mistaken for probability.

Advice for Aspiring Polymaths:

• Leonardis advises against the systematic, linear approach to learning (fundamentals to advanced) and argues these educational structures are often retrospective fictions.
• He suggests that the most effective learning occurs through osmosis and following one's curiosity rather than consciously trying to "become" a polymath.
• Textbooks are described as containing irrelevant information stripped of the original context that drove the knowledge creation.
• The core advice for young polymaths is to connect with people and groups actively working on problems they can contribute to, rather than trying to identify "unsolved problems" in isolation.
• Leonardis states there is no such thing as a universally defined "unsolved problem" that can be listed online; value is subjective to specific communities.
• He asserts that while the "chicken and egg" problem of finding mentors is real, the academic world is generally sympathetic to young people seeking contribution, making the challenge idiosyncratic rather than impossible.

Forward-Looking Statements and Future Content:

• The podcast host (Dwarkesh Patel) announced daily writing at dwarkesh.substack.com, with a new post titled "If Prediction Markets or Policies Were Legal" based on Robin Hanson's "futurearchy."
• Patel announced an upcoming podcast interview with Robin Hanson expected within the next week.
• The host acknowledged specific donations for equipment and books, including a $120 Raspberry Pi kit and a microphone purchased by Scott Hamilton.