✶Explainer01:30
Proof Doesn't Escape Fallibility — Maths Differs in Subject Matter, Not Method
The host asks whether proof sits on a different level to the rest of our conjectural knowledge — once something is proved, can't we carve it in stone? Deutsch says no. The difference between mathematics and other fields is not how we come to know things but what the subject matter is: mathematics is about necessary truth, so a mathematical discovery is a conjecture about what is necessary. Epistemologically, maths, computer science, psychology and physics are all the same.
- You cannot prove something about a thing you cannot define — hence no proofs about AGI
- A proof is always conditional: proved from what?
- Mathematics has necessary truth as its subject, not a special route to knowledge
- A mathematical discovery is a conjecture that something defined is a necessary truth
- These questions are settled by argument, not by proof
“this word proof is something we haven't talked about during our conversations but you do hear it deployed quite often”
“the subject matter of mathematics is necessary truth so when we make a discovery in mathematics we're making a conjecture about what is necessary truth”
#proof#mathematics#fallibilism#epistemology
✶Explainer09:00
Why Evolution Could Never Evolve a Way to Deflect Asteroids
Deutsch explains the two constraints that bind biological evolution: it can only make small conjectures in the form of mutations, and it can only improve incrementally, with every intermediate organism viable. Pterodactyl wings required limbs that could be changed step by step while the animal survived. Asteroid deflection has no such path — an impact every few million years exerts no selection pressure — which is why the capability appears only where explanatory thought exists.
- Mutations are small, blind conjectures — variation and selection, not recombination
- Every intermediate form has to be a viable organism, so evolution must stay alive all the way through
- Both thinking and evolution proceed in discrete steps; DNA is fully digital
- No incremental problem situation exists for asteroid deflection, so no lineage could ever get there
“it can only improve on things incrementally so you can only make you know pterodactyl Wings if you previously had limbs”
“that's why we can't expect biological evolution to my favorite example again to evolve a system for deflecting asteroids”
#evolution#creativity#search-space#asteroids
✶Explainer13:30
Knowledge Is the Most Resilient Thing in the Cosmos
From the observation that buildings resist erosion because knowledge keeps repairing them, the conversation reaches a stronger claim: knowledge-laden information is literally more resilient than any physical object. So long as we survive, what we know outlasts the longest-lived things in the universe. Shakespeare made the same claim for his sonnet, and Shelley's Ozymandias makes it for the artist's conception outliving the empire and the king.
- Everywhere rocks erode; in cities they are continually repaired by explanatory knowledge
- Knowledge-laden information outlasts physical objects — literally, not poetically
- Shakespeare's sonnet and Ozymandias state the same idea centuries earlier
- Prompts the question of whether spreading ideas beats having children
“this idea of knowledge as resilient information the very thing that will outlive even the Rocks”
“it's simply literally true that knowledge Laden information is more resilient than any physical object”
#knowledge#resilience#literature#legacy
✶Explainer15:30
Copying a Mind Isn't Free: Compute, Property and the One-Vote Problem
Asked whether uploaded minds mean near-infinite people and an explosion of creativity, Deutsch says both the opportunity and the problem are exaggerated. Duplicating an AGI either halves the memory and cycles available to each copy or requires buying more hardware — and if the AGI does not own its own computer, it is a slave. Copies do not propagate like memes. The governance question of whether a billion copies get one vote each is real but tractable: societies already extend human rights to people without the vote.
- Compute is not free; a duplicate costs real memory, cycles and hardware
- An AGI that does not own the computer it runs on is a slave
- AGI copies will not spread like a meme, because each one wants resources for its own problems
- One person one vote will need modification, but immigrants, children and temporary residents already show rights can be extended without it
“I think we mustn't think of compute as being free as the AI people call it”
“the original computer is presumably the property of the AGI because otherwise it's a slave if it doesn't even own its own body”
#agi#uploading#property-rights#governance
✶Explainer31:30
No Two People Speak the Same English
One reason it is impossible to speak so as not to be misunderstood is that everybody speaks a different English. Everyone means something slightly different by thought, freedom, idea, theory and creativity — and this holds even inside the exact sciences, where every physicist has a different conception of what a manifold is. The vocabularies overlap enough to communicate very well sometimes, never perfectly, and colleagues who want to work together often have to work at understanding what each other mean.
- People are constantly learning their own native language, especially if they care about communicating
- Key abstract words carry a different meaning for every speaker
- Physicists trained identically still differ on core concepts and problem situations
- Collaboration requires deliberate work at recovering each other's meanings
“no two people speak the same English”
“everybody speaks a different English everybody means a different thing by the word thought or freedom and and idea and theory and creativity”
#communication#language#misunderstanding#collaboration
✶Explainer50:30
How to Automate Mozart Properly: Model the Problem, Not the Style
Deutsch's second wild idea is that music generation would do much better if it stopped insisting the output resemble the existing works. If Mozart were alive, his next great work would fall outside the space an AI can synthesise from his catalogue. The right program would conjecture what Mozart's problem situation was — what he was trying to do, and why his tunes mesh with so many other considerations — rather than correcting its theory of what makes the existing works good.
- Interpolating a catalogue can only reach the space that catalogue already spans
- The target is the composer's problem situation, not the surface style
- Deutsch's own test: random notes on a piano quickly reach a place you cannot continue from
- A genuinely new work is recognisably Mozart and recognisably different
- Newton's anonymous solution was recognised by the claw — and was still a proof nobody had seen
“if Mozart was alive his next great work would not be within the space that an AI can synthesize from all his existing works”
“recognizably mozar but also recognizably different and I think that's creativity”
#ai#music#creativity#generative-models
✶Explainer54:30
Give Up on Ultimate Explanation — Including About Consciousness
Asked about the mystery of existence, consciousness and the enlightened teachers who claim to have seen through the separate self, Deutsch says it is important to give up on the idea of ultimate explanation: if you could know what you ultimately are, all further delights of understanding would be closed to you. He rejects the charge that science only looks at details — science has discovered what life is and will one day discover what consciousness is. It is the state-of-mind theorists who are focused on details and missing the context.
- An ultimate explanation would end inquiry, not complete it
- Science does the big picture — it discovered what life is
- Feynman's flower answer: the scientist appreciates everything the artist does, plus more
- Religions and mystical views can make some people happy and still hold only a sliver of truth
- The founding religions were hampered by not knowing what the sun is
“it's important to give up on the idea of ultimate explanation”
“science has discovered what is life one day science will discover what is consciousness”
#consciousness#science#religion#meaning