Neurotechnology, neural democratisation and AI safety
8 July 2026 · 11:00 am–11:05 am · Cullen
A recent paper introduces the "neural democratisation of AI" (Bain & McCay, 2024), suggesting that advanced neurotechnology will increase data and help neuroscientists build better brain models. AI researchers could leverage these insights to create more sample- and energy-efficient, human-like models, lowering the barrier to frontier AI development. However, this democratisation presents economic, regulatory, and geopolitical risks. If a broader range of companies and countries can suddenly develop frontier models, regulation becomes significantly harder, potentially compromising safety. While it is open whether neurotechnology will fundamentally advance AI, emerging approaches already integrate user brain states with multimodal input. For example, EEG wave patterns can improve transformer performance by reweighting attention during token-processing (Short et al., 2026). This lightning talk will outline the hypothesis and its safety implications. Will neural democratisation undermine safety by multiplying regulatory targets, or might brain-inspired AI yield safety gains as Mineault et al. (2025) have suggested? References "The neural democratisation of AI" by M. Bain, A. McCay. AI & Soc 39, 2589–2591 (2024). "Steering Transformer Attention with Human EEG" by C. Short, S. Basart, S. Erisken; Proc. First Workshop on NeuroAI Multimodal Intelligence, PMLR 308:199-204 (2026) “NeuroAI for AI Safety” Preprint https://arxiv.org/abs/2411.18526 (2025), P. Mineault, N Zanichelli, J Z Peng, A Arkhipov, E. Bingham, J Jara-Ettinger, E Mackevicius, A Marblestone, M Mattar, A Payne, S Sanborn, K Schroeder, Z Tavares, A Tolias
Recording
Speakers
Allan McCay
Co-director of The Sydney Institute of Criminology, Academic Fellow at the University of Sydney Law School and President of the Centre for Neurotechnology and Law, University of Sydney
As well as consulting with international bodies such as OECD, UNICEF, Allan is a member of the Expert Advisory Group of the UNODC and INTERPOL Collaboration on Neurotechnology in Law Enforcement and Criminal Justice + member of UNESCO Expert Group for the Implementation of the Recommendation on the Ethics of Neurotechnology. He is also member of the Australian Human Rights Commission Expert Reference Group on Human Rights and Neurotechnology.
Avinash Singh
Senior Lecturer and Director, Human Augmentation Lab, University Of Technology Sydney
Dr Avinash Singh is Director of the Human Augmentation Lab at the University of Technology Sydney (UTS). His current research explores cognitive function modelling, brain dynamics, adaptive human–AI interaction, and next-generation brain–computer interface technologies for neuroadaptive systems, cognitive enhancement and human augmentation.
Michael Bain
School of Computer Science and Engineering, UNSW
Researcher in Machine Learning and AI.
Audience Q&A
Ask a question or upvote others.
Loading questions…
Transcript
Avinash Singh
This talk might be interestingly another succession. What you're hearing so far in AI safety. And we're talking a little bit further, which is also another safety risk, with the help of neurotech. The brain obviously continuously perceives everything around us. There's no doubt in this. And when it perceives all that information, the whole idea behind that processing is just to adapt, to learn, to make decisions and of course also making different choices. That's what the brain does all the time. But the best part of this, especially from the eyes of neurotech, is that all this higher-order processing happening in your brain — when you're processing information, making decisions — happens on the surface of the brain, which is called the cortex, or neocortex in some cases.
And the best part of this area is you can measure it. You can use some neuroimaging technique to extract that information happening in your brain. And one of such work from my lab — I did just want to share one of the examples, because that's what the topic of this talk is as well. I could use this cortex area to measure, using EEG, if there's some degree of agree to disagree between when I present some information to a user. Essentially, what I'm trying to say is I can use that neocortex processing information from the brain signal and can tell you within a short period of time — up to 200 milliseconds, which is one fifth of a second, very fraction of second — to extract and tell you if someone is agree, disagree, or something in the middle somewhere on that axis, if I put this on the line.
And that's incredible. Incredible as in, it's essentially telling you a lot of different things are possible just by measuring this brain signal, and potentially using it for something more, some application and so on and so on. And over to Mike.
Michael Bain
The neural democratisation of AI is a hypothesis. And the original idea in the paper was two things would lead to AI improvement. One is reduced sample need for data and the other is reduced energy requirements. In the ensuing periods, there is some evidence that neuro AI is potentially moving in that direction. At the top of the slide, this is from a paper where they showed that brain data could be used to improve attention, which is the core mechanism behind most current AI LLMs, for example. But there is also the question of safety.
Human natural intelligence is the best current route that we know of regulating somewhat safe intelligence. And there is also the possibility of using brain data, for example, to align better with semantic representations in the brain, which can help potentially train AI under decision-making. The upshot of this is that there might be some safety benefits to a kind of neuro approach to AI, but there might also be some risk, and it might be the case that more powerful AI can be created more cheaply with less data, less compute, and less energy.
Which sounds good, but it might be the case that the neural democratisation shrinks the moat of the big AI companies, which have a big moat at the moment because of their deep pockets. Hopefully AI might become open to many more entrants into the market that can compete with big companies at the frontier, and that might lead to a regulatory risk, which would be a kind of safety risk. And so if you think about things from the perspective of regulators, regulators would rather regulate a small number of large companies than a large number of small companies, as noted by Berg in another paper.
And as the AI ecosystem, according to our hypothesis, starts to evolve to have a larger number of smaller players, including companies or even countries, then it seems like the risk from inadequately regulated AI might increase. And so the upshot of all of this is the democratisation hypothesis might force a sort of change in the economics and geopolitics of AI safety, which might require a reconsideration of AI safety in light of a larger number of entrants into the frontier of AI development. And finally, if you're interested, come and chat to us, we're around.
But on Saturday the 5th of December at UTS, just after NeurIPS. Oh, sorry. Just before NeurIPS we're planning to organise an event. And if you're interested, please scan the QR code and we'll let you know, and we'll get in touch near the time. Thank you very much.