---
title: 'Scientists identify what it takes for AI to become conscious'
source: 'https://youtube.com/watch?v=Q-tcml1wzM0'
video_id: 'Q-tcml1wzM0'
date: 2026-11-11
duration_sec: 378
channel: 'Sabine Hossenfelder'
---

# Scientists identify what it takes for AI to become conscious

> Source: [Scientists identify what it takes for AI to become conscious](https://youtube.com/watch?v=Q-tcml1wzM0)

## Summary

The video discusses a new scientific paper arguing that current AI systems are unlikely to replicate consciousness as it arises in biology, proposing a middle ground called 'biological computationalism.' The creator, who supports computational functionalism, finds the paper's arguments valuable and outlines three key functional differences between the brain and AI.

### Key Points

- **Paper's Core Claim** [00:15] — A team of scientists lists reasons why current AI cannot be conscious, with headlines like 'Consciousness may depend on the physics of the brain.'
- **Debate Context** [01:08] — Two sides: computational functionalism (consciousness from information processing, not physical means) vs. biological naturalism (consciousness requires specific biological processes).
- **Biological Computationalism** [02:06] — The paper proposes a middle ground, focusing on functional differences between the brain and AI rather than the biological basis.
- **Difference 1: Multiscale System** [02:37] — The brain operates at multiple scales from molecules to oscillations, all necessary for consciousness; anaesthesia disrupts large-scale patterns while lower levels remain intact.
- **Difference 2: Temporal Span** [03:12] — Brain processes span many orders of magnitude in time, from fast blips to slow modulations, unlike current AIs.
- **Difference 3: Metabolic Limits** [03:41] — The brain's energy efficiency constraints may have driven the evolutionary path to consciousness, not just better thinking.
- **Paper's Clarification** [04:14] — Authors explicitly state they do not claim only biological tissue can instantiate consciousness, but current AIs are unlikely to replicate conscious processing.
- **Path Forward** [04:49] — Looking for similarities to human brains is the obvious way to proceed, despite not knowing what consciousness is.

### Conclusion

The paper offers a nuanced perspective that current AI is unlikely to achieve consciousness in ways we recognize, emphasizing functional differences. The creator finds this approach valuable and suggests it points toward a practical research direction.

## Transcript

Can artificial intelligence be conscious? This&nbsp; is maybe the most controversial topic today,&nbsp;&nbsp; at least among the people who I&nbsp; interact with. Though this tells&nbsp;&nbsp; you more about my social circle than&nbsp; about what’s controversial. In any case,&nbsp;&nbsp;
today I have a team of scientists who list some&nbsp; good reasons for why artificial intelligence,&nbsp;&nbsp; the way that we currently pursue it, can’t&nbsp; be conscious. “New research argues that&nbsp;&nbsp;
consciousness depends on biological computation,”&nbsp; one headline put it. Or “Consciousness may depend&nbsp;&nbsp; on the physics of the brain.” That totally&nbsp; disagrees with my opinion. So I’ve had a look.
Most of what we call artificial intelligence&nbsp; today are large language models. These have&nbsp;&nbsp; become extremely good at producing text and&nbsp; increasingly also audio. When it comes to&nbsp;&nbsp; language generation, these AIs are now&nbsp; experts at mimicking humans, including&nbsp;&nbsp;
our most distinctive trait, being confidently&nbsp; wrong. And the more human-like they appear,&nbsp;&nbsp; the more pressing the question becomes whether&nbsp; they can become conscious. Anthropic reportedly&nbsp;&nbsp;
keeps backups of earlier Claude versions just in&nbsp; case deleting them will one day count as murder. There are two sides of this debate,&nbsp; the one, which I am solidly on,&nbsp;&nbsp; is called computational functionalism. It has&nbsp; it that consciousness comes from the way that&nbsp;&nbsp;
the human brain processes information. That&nbsp; it’s nothing to do with the physical means by&nbsp;&nbsp; which this information is being processed. In&nbsp; particular, one doesn’t need actual neurons,&nbsp;&nbsp;
one can very well simulate the&nbsp; action of a neuron on a computer. The other side of the debate is called biological&nbsp; naturalism, and it has it that consciousness needs&nbsp;&nbsp; the specific biological processes going&nbsp; on in the human brain, you can’t replace&nbsp;&nbsp;
them with something else. Why not? I don’t know,&nbsp; that’s why I am on the other side of the debate. The new paper now comes from two biologists,&nbsp; and you’d expect me to hate this. But I think&nbsp;&nbsp;
the paper actually makes an important contribution&nbsp; that might become influential. The authors argue&nbsp;&nbsp; for a middle ground that they call biological&nbsp; computationalism. They say basically, look,&nbsp;&nbsp; let’s leave aside the question of whether&nbsp; we need the computational basis to be truly&nbsp;&nbsp;
biological or not, let’s instead look&nbsp; at the ways that the human brain is&nbsp;&nbsp; functionally different from the current&nbsp; AIs, and see what we can learn from that. They say there are three key differences. The&nbsp; first is that the brain is a multiscale system.&nbsp;&nbsp;
The relevant processes don’t start with&nbsp; neurons, they begin at a lower level,&nbsp;&nbsp; with molecules, chemical reactions, then cells&nbsp; and connections between them, electrical signals&nbsp;&nbsp; along those connections and activity oscillations&nbsp; throughout the entire system. All of those are&nbsp;&nbsp;
necessary for consciousness. We know this for&nbsp; example because anaesthesia interrupts some&nbsp;&nbsp; large-scale patterns while the lower levels remain&nbsp; intact and consciousness disappears with that. layers at which they operate, from single&nbsp; tokens to overarching plans and back down,&nbsp;&nbsp;
but the complexity is nowhere near&nbsp; comparable to that of the human brain. The second key difference is that brain&nbsp; processes span many orders of magnitude&nbsp;&nbsp; not just in size but also in time, from&nbsp; incredibly fast blips to slow modulations.&nbsp;&nbsp;
It’s like the brain interpolates&nbsp; from digital to analog but isn’t&nbsp;&nbsp; really either. Again there is nothing&nbsp; comparable going on in the current AIs. And the third difference is probably the most&nbsp; obvious one, that the brain has metabolic limits&nbsp;&nbsp;
that force it to be energy efficient. If you look&nbsp; at it from a computational perspective, the brain&nbsp;&nbsp; is optimizing more difficult constraints than&nbsp; large language models. So it’s possible that the&nbsp;&nbsp; evolutionary path to consciousness wasn’t better&nbsp; thinking per se, but more efficient thinking.
Now, maybe I am misunderstanding the paper,&nbsp; but I don’t find this objectionable. To me&nbsp;&nbsp; the paper doesn’t say that we actually need&nbsp; biological tissue to get consciousness. In&nbsp;&nbsp; fact the authors write explicitly “We do&nbsp; not claim that only biological tissue can&nbsp;&nbsp;
instantiate consciousness”. Rather their point&nbsp; is that current artificial intelligence systems&nbsp;&nbsp; are “unlikely to replicate conscious&nbsp; processing as it arises in biology”. This can mean two things. One is that the&nbsp; current AIs are unlikely to become conscious,&nbsp;&nbsp;
period. Or that they are unlikely to become&nbsp; conscious in ways that we would recognize. Now you might say but this doesn’t really&nbsp; solve any problem because we still don’t&nbsp;&nbsp; know what consciousness is. And that is true, but&nbsp; I think that looking for similarities to human&nbsp;&nbsp;
brains is the obvious way to proceed.&nbsp; So why isn’t there more work on this?&nbsp;&nbsp; I suspect it’s because from the economic&nbsp; perspective, it’d be bad news. I am&nbsp;&nbsp; pretty sure the European Commission is&nbsp; already working on a consciousness tax.
As you can tell, I think a lot about the future&nbsp; that science and technology will bring. And if&nbsp;&nbsp; there's one conclusion I've arrived at, it's&nbsp; that today's school education doesn't prepare&nbsp;&nbsp; children for what's to come. Luckily, today's&nbsp; sponsor, Brilliant, helps my kids to get ready&nbsp;&nbsp;
for the future, and they can help yours, too.&nbsp; Brilliant just recently introduced a new AI&nbsp;&nbsp; tutor that'll walk children through problem&nbsp; sets in their many courses on science, maths,&nbsp;&nbsp;
and programming. The tutor doesn't give away the&nbsp; answers, but helps children when they're stuck,&nbsp;&nbsp; reminds them of what they already know, and&nbsp; explains each step to the solution. And the&nbsp;&nbsp; more you use it, the better it understands your&nbsp; personal struggles on how to get you unstuck.&nbsp;&nbsp;
My daughter loves it because in contrast to&nbsp; me, Brilliant's Tutor is infinitely patient&nbsp;&nbsp; and always has time. You can get started&nbsp; with Brilliant's Tutor for free and if you&nbsp;&nbsp; use my link brilliant.org/Sabine or scan&nbsp; the QR code, you can save 20% on an annual&nbsp;&nbsp;
subscription. So go and check this out.&nbsp; Thanks for watching. See you tomorrow.
