AI Consciousness: Biological Constraints and Current Limitations

Clip title: Scientists identify what it takes for AI to become conscious Author / channel: Sabine Hossenfelder URL: https://www.youtube.com/watch?v=Q-tcml1wzM0

Summary

Sabine Hossenfelder’s video addresses the highly debated question of whether artificial intelligence (AI) can truly become conscious, arguing that current AI systems, specifically large language models (LLMs), cannot achieve consciousness as we understand it. She frames this discussion by outlining two prevalent philosophical viewpoints: Computational Functionalism, which posits that consciousness arises from information processing regardless of its physical substrate (e.g., silicon chips vs. neurons), and Biological Naturalism, which insists that consciousness is fundamentally dependent on the specific biological processes and structures of living systems, like the human brain. Hossenfelder aligns herself more with the latter, finding the idea of “biological computationalism” a promising middle ground.

The video then delves into a new research paper that advocates for “Biological Computationalism,” suggesting that conscious experience is intrinsically linked to the concrete physical processes of living systems. The authors of this paper identify three critical differences between how the human brain functions and how current AI models operate, making it unlikely for today’s AI to replicate biological consciousness. Firstly, the brain is a multi-scale system, with processes integrated from molecular levels through chemical reactions, cells, and large-scale network oscillations. Anesthesia, for example, disrupts these large-scale patterns while lower-level functions remain intact, causing consciousness to disappear. LLMs, despite having layers, lack this profound multi-scale integration and complexity.

Secondly, the brain’s processes span a vast range of temporal scales, from extremely fast neural blips to slow modulations, effectively operating as a hybrid analog-digital computer. This dynamic temporal integration is crucial for consciousness but is absent in current AI systems. Thirdly, and perhaps most obviously, the brain operates under significant metabolic constraints, forcing it to be highly energy-efficient. This evolutionary pressure has optimized the brain for efficiency, leading to a complex form of “efficient thinking” rather than simply “better thinking.” These metabolic limits and the resulting optimizations are not present in the design of current AI, which prioritize computational power over energy efficiency.

Based on these functional differences, the paper concludes that present-day AI systems are either unlikely to become conscious at all or, if they do, their consciousness would be vastly different from what humans could recognize. Hossenfelder supports this view, emphasizing that understanding these fundamental biological differences in computation is key to progressing AI research toward potentially mimicking consciousness, rather than merely focusing on information processing alone. She muses that perhaps the lack of focus on these biological constraints in AI development is partly due to economic factors, highlighting a broader concern that current educational systems might not adequately prepare future generations for the scientific and technological advancements on the horizon.

Video Description & Links

Description

One of the biggest debates in tech at the moment is the question of whether current AI models can become conscious. Whatever “conscious” means. According to a group of biologists, current AI models are far from it – but it’s not because consciousness can only appear in biological systems. They identify which key properties an artificial system would have to fulfill to become more like the human brain and hence have a chance at developing consciousness. Let’s take a look.

Paper: DOI: 10.1016/j.neubiorev.2025.106524

sciencenews AI consciousness neuroscience

👕T-shirts, mugs, posters and more: ➜ https://sabines-store.dashery.com/ 💌 Support me on Donorbox ➜ https://donorbox.org/swtg 📝 Transcripts and written news on Substack ➜ https://sciencewtg.substack.com/

URLs

Sabine Hossenfelder — Wikipedia
Large Language Models — Wikipedia

NemoClaw Knowledge Wiki

Explorer

AI Consciousness: Biological Constraints and Current Limitations