TLDR
Evolutionary biologist David Krakauer joins Neil deGrasse Tyson to explore the nature of intelligence across biology and AI. They discuss how intelligence evolved from simple sensing to complex problem-solving, the distinction between capability and competence, and why current AI systems — despite impressive benchmarks — operate through inefficient, energy-intensive heuristics rather than true understanding. Krakauer argues that human intelligence is defined by making hard problems easy under constraints of time and energy, while LLMs solve problems by 'eating soup with a fork' — possible but wildly inefficient — and warns that outsourcing cognition to competitive cognitive artifacts like GPS erodes the intuitive competence that makes intelligence robust.
For you
PRO members can sign in and request more information — or their own version of this digest, tuned to what they care about.
Sign in to unlockSurprising Insights
Brains made of simple cells produce extraordinary thoughts, while human societies made of extraordinary individuals produce 'a total show in the aggregate' — collective intelligence can be inversely correlated with individual intelligence.
Current LLMs solve math problems using only language, without visual or geometric reasoning — they 'eat soup with a fork,' achieving results through massive energy expenditure rather than efficiency.
AI models learn all mathematical concepts simultaneously without a curriculum, so they develop heuristics that approximate answers but fail at scale (e.g., long division with many digits), unlike humans who learn exact algorithms stepwise.
The cultural deposit of human knowledge (books, texts) is being mined by LLMs like coal was mined in the Industrial Revolution, but unlike coal, knowledge can be fouled by the very process of extraction — AI generates conceptual entropy (misinformation) that pollutes the source.
Under universal Darwinism, alien intelligences evolving under scarcity would likely resemble human intelligence (efficient, data-sparse, fast) more than current AI (energy-hungry, data-greedy, slow), because evolution selects for survival under constraints.
Key takeaways
Intelligence is ubiquitous in life: any organism that senses its environment and acts adaptively to survive possesses some form of intelligence, making intelligence a continuum rather than a binary trait like life itself.
Human language enables 'infinite use of finite means' (Humboldt) through grammar, allowing arbitrarily complex thought, but also decouples us from immediate reality and enables both great intelligence and great stupidity.
Consciousness is about 'being and feeling' (Anil Seth) and is primordial — even single-celled organisms have a sense of self — while intelligence is about problem-solving, often occurring unconsciously through evolved heuristics.
Tools fall into two categories: complementary artifacts (abacus, map) that scaffold and internalize reasoning, and competitive artifacts (GPS, calculator) that replace reasoning and create dependence on external providers.
Capability (solving a benchmark) differs from competence (understanding how to solve it); AI systems demonstrate capability without competence, solving problems through statistical heuristics that fail at scale, like getting long division wrong with many digits.
LLMs are like steam engines liberating a 'cultural deposit' of all human text, but they generate entropy in both the physical environment (energy use) and conceptual space (misinformation, confabulation) because they lack a curriculum and learn everything at once.
Biological intelligence evolved under energy and time scarcity: brains evolved because genomic evolution becomes too slow in large, long-lived organisms, so brains pull information from the world in real time — a constraint AI ignores.
The future of human cognition lies not in calculation but in asking good questions; as tools automate technical problem-solving, the Santa Fe Institute cultivates 'amazing question askers' who can navigate the unknown.
Notable quotes
“intelligence for me is making hard problems easy across the animal every species even plants finding an efficient way of solving the problem not an inefficient way of solving”
▸ Notable quotes · 0:04“Human language is making infinite use of finite means”
▸ Notable quotes · 11:06“Intelligence is about problem solving and consciousness is about being and feeling”
▸ Notable quotes · 17:01“you could eat a bowl of soup with a fork you could slowly enough time enough energy you can solve math problem with language enough time enough energy”
▸ Notable quotes · 32:22“if you outsource completely you lose the ability to reason”
▸ Notable quotes · 39:19“the LLM is like the steam engine it's liberating that energy which is our literature which is our language which is our thoughts and cultures and everything that we are all in one depository”
▸ Notable quotes · 60:56Chapters
0:00 — Defining Intelligence Across Life: Krakauer defines intelligence as making hard problems easy through efficient solutions, present in all life forms from microbes to plants, contrasting with AI's inefficient approaches.
3:28 — Life vs Intelligence: Binary vs Continuum: Life is binary (alive or dead) but intelligence is a continuum; the theoretical challenge is how a binary system gives rise to graded intelligence.
6:00 — Collective Intelligence and Evolutionary Transitions: Discusses collective intelligence in bacteria, brains, and societies; notes the paradox that simple cells create complex thought while smart individuals create dumb mobs.
8:10 — Language, Signaling, and Human Uniqueness: Explores the distinction between signaling (indexical, one-to-one) and language (symbolic, generative), citing Humboldt's 'infinite use of finite means' and Peirce's semiotics.
16:10 — Consciousness, Unconscious Problem-Solving, and Dreams: Separates consciousness (feeling, selfhood) from intelligence (problem-solving); describes how unconscious processing solves complex problems (Kekulé's benzene dream) and communicates via symbols, not language.
28:30 — AI vs Human Intelligence: Capability vs Competence: Contrasts LLMs (language-first, no unconscious, collective training, energy-inefficient) with biological intelligence (embodied, curriculum-based, energy-constrained); introduces complementary vs competitive cognitive artifacts.
34:50 — Tools That Scaffold vs Tools That Replace: Uses abacus (complementary, internalizes skill) vs GPS (competitive, creates dependence) to illustrate how outsourcing cognition erodes competence and creates reliance on providers.
55:30 — LLMs as Steam Engines Mining Cultural Deposits: Draws analogy between Industrial Revolution (steam engines liberating fossil energy) and AI (transformers liberating cultural knowledge); both produce entropy — physical and conceptual.
Speakers
Neil deGrasse Tyson — Host
Chuck Nice — Co-host
Gary O'Reilly — Co-host/Producer
David Krakauer — Guest