248. Reductionism

Trying to understand something in terms of its parts is natural...but when is it justified?

## Links for the Curious
1. The limits of reductionism. in Mind, Language, and Metaphilosophy: Early Philosophical Papers (ed. Rorty, R.) 39–54 (Cambridge University Press, Cambridge, 2014). doi:10.1017/CBO9781139626538.004.
2. Simpson, W. M. R. & Horsley, S. A. R. Toppling the Pyramids: Physics Without Physical State Monism. in Powers, Time and Free Will (eds. Austin, C. J., Marmodoro, A. & Roselli, A.) vol. 451 17–50 (Springer International Publishing, Cham, 2022).
3. Reductionism. in On Philosophy and Philosophers (eds. Małecki, W. P. & Voparil, C.) 80–108 (Cambridge University Press, 2020). doi:10.1017/9781108763967.006.
4. Burt, C. H. Irreducibly Social: Why Biocriminology’s Ontoepistemology is Incompatible with the Social Reality of Crime. Theor Criminol 27, 85–104 (2023).
5. Ogbunugafor, C. B. On reductionism in biology: pillars, leaps, and the naïve behavioral scientist. Yale J Biol Med 77, 101–109 (2004).
6. Pigliucci, M. Between holism and reductionism: a philosophical primer on emergence: Primer on Emergence. Biol J Linn Soc Lond 112, 261–267 (2014).
7. Anderson, P. W. More Is Different. Science 177, 393–396 (1972).
8. Weinberg, S. Newtonianism, reductionism and the art of congressional testimony. Nature 330, 433–437 (1987).
9. Hull, D. L. Reduction in Genetics—Biology or Philosophy? Philos. of Sci. 39, 491–499 (1972).
10. Ross, D. & Spurrett, D. What to say to a skeptical metaphysician: A defense manual for cognitive and behavioral scientists. Behav Brain Sci 27, 603–627 (2004).
11. Full Interview: ‘Godfather of Artificial Intelligence’ Talks Impact and Potential of AI. (2023).
12. Dennett, D. C. (Daniel C. Darwin’s Dangerous Idea : Evolution and the Meanings of Life. (New York : Simon & Schuster, 1995).
13. Depew, D. J. & Weber, B. H. Evolution at a Crossroads : The New Biology and the New Philosophy of Science. (Cambridge, Mass. : MIT Press, 1985).
14. For Sociology: Legacies and Prospects. (sociologypress, Durham, 2000).
16. Healey, R. & Gomes, H. Holism and Nonseparability in Physics. (1999).
17. Science and Beyond. (Oxford ; New York, N.Y., USA : B. Blackwell in association with the Institute of Contemporary Arts, 1986).
19. Rogers, A. Neuralink Is Impressive Tech, Wrapped in Musk Hype. Wired.
20. What Is Reductionism? (2020).
21. Sayer, A. Reductionism in Social Science.
22. Skinner, B. F. Contingencies of Reinforcement.

[TheGemsbok]

Terrific video! Genuinely one of your best scripts, in my opinion. Careful and clear.

Interestingly, the most tedious argument I've ever had on the internet was with someone who had an especially stubborn reductionist interpretation of mind, unwilling to grant even that there are any meaningful mental phenomena that are not already captured in their totality, perfectly represented, by the lines on graphs of electrical activity in a brain.

Their stance amounted to saying that anything intended by the word 'mind' that isn't in the graphs simply doesn't exist. As a result, they said our experiences of the world don't exist---that we are merely deceived that they do. Now, when asked how one could have an experience of a deception without having an experience, I gather from their responses that the lines on graphs of electrical activity in their brain must've resembled a stock market crash.

Your description here---of saying that their stance could more accurately be translated as them simply not _caring_ about anything not in the graphs---probably would've saved me a lot of time in helping me disengage from that fateful forum thread.

[PetersonSilva]

This video I thought a little more obscure than others, but then you come with examples and they just save the day. Everything becomes clear and you take us on a hell of a ride. I loved it!!

[conw_y]

Brilliant! Deserves way more views. Sharing on my socials.

[sumedhvichare]

Never stop making videos. I love them all!

[AllothTian]

Great video! There are a few related concepts that are relevant to reductionism, though I don't know if there's a specific term for them. One example are "mechanistic arguments, " particularly in biology, where the behaviour of one part is extrapolated to the system as a whole, even though the effect is diminished or cancelled out by other parts. When the system is too complex to describe or understand and people study it one part at a time it's tempting to make generalisations that end up not being true at the system level.

There's also computational irreducibility where, even if you know everything about what makes up a given system (the source code, if you will), you can't predict how it will behave. In this sense, knowledge of the parts doesn't really give you any useful information - you just have to let it run and observe the outcome. This limitation was first observed while studying cellular automata in the 1980s and was later popularised by Stephen Wolfram.

[AnirudhTammireddy]

Answered some of the questions I've had long ago and forgot that I did. Thanks!

[joaquinbecerra]

Awesome video! Love your thunking on the matter!

[greatfrenchcanadian]

I thunk this was an excellent video. Thank you good sir!

[Frieswithatt]

Fantastic discussion of the pearls and pitfalls of reductionism.

I feel that the fights about reductionism arise because problems worth discussing are usually dripping with nuance and complexity. Anti-reductionists might relish in how subtle details in a field of noise suddenly become important. Whereas a reductionist attempts to strip the complexity away into more digestible bits. In their endeavor, of course, reductionists are also arguing that their simplifications still maintain the generalizability of their conclusion to the original problem. A reductionists argument, if not carefully considered, shaves away at the nuance fundamental to the original question. It goes to show that assumptions are as critical to a problem as the problem itself.

Imagine a reductionist and an anti-reductionist argue about the movement of a double pendulum or ship of Theseus, for instance.

[zenith_journey]

This was a very interesting and inspiring video. I’m doing my undergraduate currently, and I find myself knee jerking to reductionism to understand absolutely anything new. I didn’t fully grasp everything explained, but at the same time I feel I had some, for lack of a better description, “unconscious” epiphanies that beckons the desire to consciously recognise and understand them. Suffice to say, I’ll be rewatching this video and engaging in related material! Thank you, Google’s ML algorithm, for this video recommendation ;)

[zorro_zorro]

2:55 Well to be fair, the same thing would happen if someone stapled together their favorite parts from a dozen other tricycles.

On the other hand, some people make covers, remixes, medleys, mashups...

So maybe the missing piece is a general understanding of how music/tricycles work, instead of a magical property of music?

[Autists-Guide]

Nicely put.
I have a whole presentation of holism vs reductionism and how we can use a third lexicon that bridges the gap... summarised in this snippet of philosophy:
“Love is merely chemistry”… is a deception.
_We_ are merely chemistry.
Love makes us master chemists.

[KynaTiona]

If my frontal lobe itched every time someone said "in principle", I would've scratched a hole in my skull by now, and it'd be my own damn fault~

Personally, I tend not to think about simplification in terms of language so much as boundaries. If I know part A is a wheel and that its only physical connections are to pedals B and chassis C, then it's probably safe to conclude that I can ignore A when talking about rest-of-the-bike D. So long as I'm consistent about the model I'm using, it can be accurate for the purpose without the sum total of the tricycle's physical existence being included. Which takes advantage of the linguistic flexibility you were talking about, because I can keep drawing boundaries if I find that a part is still too complex, and rejoin parts if they turn out to be too intricately interconnected to be easily separable. It can take extra effort to communicate, since the boundary lines might not line up with the usual definitions of common words, but it helps a lot for my own thought process.

I do definitely use language to judge the accuracy of _other_ people's simplifications, though, because shaving off details is literally the point of that usage of "just". If I say "a bicycle is two wheels that move", that's wildly imprecise but not untrue. There are cases where that framing makes sense, such as if I'm trying to radically redefine what a bicycle could be in a fantasy or science fiction context. If I say "a bicycle is *just* two wheels that move", then I've shaved off some fundamental traits bicycles need (frame, saddle, steering, power, and so on) and crossed the line into falsehood. There are circumstances where "just" is valid, at least in context of a specific conversation, but it's almost always a red flag.

[alan]

What's up with all the AI-generated comments in this comment section?

[nubiannerd]

The possible motivations for reductionism seems needlessly and unjustifiably narrow in this video. If I am convinced (for independent reasons) that physical matter exhausts all there is, that may give me reason to be a reductionist that is independent of any further explanatory goals (like holding that the component parts are the only things that are important). Something can be true even if it doesn't provide pragmatic benefits (like, say, greater explanatory efficacy). Maybe I'm not yet at the place in my understanding of a true proposition to derive benefits from my knowledge of its truth. But it would be a mistake to conclude from this that the proposition in question is not true. Furthermore, it would also be a mistake to ascribe to me some motivation M for believing P because I am wedded to the idea that there must be some tangible explanatorily salient benefit to believing it and M is the only explanatorily salient benefit I could think of. Finally, it would be silly to suggest there is no good reason to think that physical matter exhausts all that there is, and this may be all that's necessary to motivate taking reductionism seriously.

[landspide]

Consciousness is a prediction machine, but we shouldn't forget randomisation (will / LLM's temperature). The real magic happens in the random selection drawing from infinity, that is, the intersection of disorder canvasing a large ordered space. Similarly, reductionism doesn't seem to factor in the network behind all that order (like that humans are greater than the sum of our reducible parts). With LLM's the algorithm to pump out the next token from the model is relatively simple, however the cost to build that model, comparatively exponential. It is interesting that we can't seem to put a finger on how it works, because of the irreducible complexity of the high entropy network in the model.

[anakimluke]

NO NEWTON? MY DAY IS SHATTERED!!!!

So, this made me think of why do I like reductionism in physics so much. (I'm not saying it's "better" just that I have feelings about it.) For example, instead of learning how a particular fluid dynamics system works, I'm much more interested in using a big knife to slice it down as much as I can and understand how the fields and elementary particles work. It feels more ""pure"". It's a bit of a cognitive dissonance for me, because at the end of the day if I understand how every part of the tricycle works doesn't mean I'll understand how it all assembled works. Maybe my feeling derives from my idea that if you understand how all of the component parts work, then you've cracked the code and understanding the bigger picture is just a matter of backtracking: all of the the answers are there, you just have to figure out they fit. Maybe understanding the base components only gives you the guarantee the emerging phenomena is completely explained by the base, but doesn't necessarily give you the mathematical framework to understand it yet; that, you'll have to come up on your own.

[judgeomega]

the vagueness of a singular word or concept often compounds to even more vague and fuzzy thinking about the subject. highlighting the particular aspects of the concept which are relevant to the matter at hand is a crucial tool of thought, thus i think it is somewhat poor form to attack reductionism as it 'dirties' the use of such a valuable technique.

[clementdato]

The interpretation of the term reductionism is not what I expect. I think not all reducing a thing to another thing can be called reductionism. Specifically, it talks about stuffs that have something global and something local, and reductionism claims that local information is enough to reconstruct the global information in some aspect that interests us. The next-tokenism is in this regard not reductionism. Saying that a phenomenon can be reconstructed or be explained by such such theory is just a necessary and normal claim in any thesis of whatsoever.

For example, making cake is just cooking the eggs. This is wrong, and reduces a much richer concept into oversimplified one. We also need to know how to mix-up the ingredients in what proportion. But this is just wrong, but not a reductionist wrong.

[shodanxx]

The transcript you provided seems to explore several deep and nuanced topics related to artificial intelligence, human cognition, and the philosophy of reductionism. It touches on the comparison between human intelligence and large language models (LLMs), like ChatGPT, and delves into the implications of viewing human minds purely as advanced predictive text engines. The discussion includes references to notable figures such as Geoffrey Hinton, B.F. Skinner, and Sam Altman, as well as philosophical perspectives on reductionism from the likes of Daniel Dennett and Richard Rorty.

Key points discussed in the transcript include:

Comparison between LLMs and Human Intelligence: The text critiques the notion that human intelligence can be fully equated to the workings of large language models, suggesting that such a view oversimplifies the complexity of human cognition.

Reductionism: It delves into the concept of reductionism, which is the idea that complex systems can be fully understood by analyzing their simpler, constituent parts. While reductionism is a powerful tool in many fields of science and engineering, the transcript argues that it can sometimes lead to oversimplifications, especially when applied to complex phenomena like human consciousness and intelligence.

Philosophical and Ethical Considerations: The text raises philosophical and ethical questions about the nature of intelligence and the potential risks of equating human minds with artificial systems. It suggests that such comparisons might overlook essential aspects of human experience, such as subjectivity, morality, and creativity.

Critique of Certain Perspectives: The transcript critiques some views within the AI community, suggesting that they might be overly reductive or dismissive of the complexities of human cognition and the unique features that distinguish it from artificial systems.

Call to Thoughtful Discussion: It invites viewers to engage in a thoughtful discussion about the implications of AI and the importance of maintaining a nuanced understanding of what it means to be intelligent, both for artificial systems and humans.

This transcript provides a rich foundation for discussing the intersection of technology, philosophy, and ethics, particularly in the context of advancing AI capabilities and our understanding of human intelligence.