Is biological computing the future of AI?
A recent paper advocates for a new field of "organoid intelligence" based on interconnected and interfaced brain organoids. It could be a game-changer for AI but it's also fraught with challenges
Imagine the scenario: An AI that’s smarter than ChatGPT, that feels comfortingly human while being incredibly knowledgeable, that masterfully solves complex problems, and that uses only a fraction of the energy consumed by current AI systems. Now imagine that, when you start digging into the tech that lies this AI’s innocuous-seeming interface in your browser or on your phone, you find is an array of miniature interconnected and interfaced human brains — a chimeric mashup between inorganic tech and living systems.
This all sounds very science fiction, and probably will remain so — for a few years at least. But earlier this year a group of scientists proposed researching and developing brain organoids — small three dimensional clusters of human brain cells that demonstrate brain-like behavior — as the basis for advanced computing systems.
Their paper — Organoid intelligence (OI): the new frontier in biocomputing and intelligence-in-a-dish — was published in the journal Frontiers in Science, and lays out a pathway to developing brain organoids as a substrate for biological computers that “could be faster, more efficient, and more powerful than silicon-based computing and AI, and only require a fraction of the energy.” These, the authors claim, could compliment and extend what is possible using current compute substrates for AI.
The idea is certainly an intriguing one. Unlike current iterations of AI — including foundation models like GPT4 and LaMDA — the human brain is capable of inductive reasoning based on limited training, while using a fraction of the energy of digital systems with similar compute power. If brain-like hardware and software architectures could be developed, it would revolutionize artificial intelligence.
But what if, instead of emulating the brain with digital systems, we could simply grow a brain do do all of this instead?
The bad news (or good news, depending on your appetite for straying into ethically nebulous territory) is that growing a biological brain-based supercomputer isn’t that easy. Brain organoids are currently exceptionally small — only a few hundred micrometers or so in diameter — and are size-limited by the challenges of keeping them fed with oxygen and nutrients while removing heat and other waste products. And until recently it wasn’t possible to create the 3D structure necessary to support true brain-like behavior.
But the field is moving fast, and the possibility of using brain organoids as a compute substrate no longer looks as fantastical as it might once have done.
Using current technologies, it’s possible to develop 3D scaffolds that allow human brain cells derived from pluripotent stem cells to grow and self-assemble. And while small, these offer tantalizing possibilities as a compute substrate.
For starters, these clusters of cells naturally begin to develop structures and behaviors that mimic those found in the human brain. And as they grow and develop, it’s possible in principle to embed sensors and conductors that will allow the them to receive data input, generate data output, and network with other organoids.
This is already beginning to sound like the architecture of a deep learning neural net, but with some important differences. Because of the nature of biological neuron networks, they are able to process information in ways that massively extend their ability to learn, adapt, and problem solve — all while being exceptionally energy-efficient.
I’m particularly intrigued here in the potential ability to integrate read-write capabilities into the very structure of brain organoids. One of the challenges of the human brain is that, once formed, it’s pretty much impossible to embed brain-machine interfaces that connect with every neuron and every synapse. But what if these “neural nets” grew organically with the brain.
This is clearly not possible in humans, but it may be with brain organoids. We’re not there yet, and current (and proposed) organoid interfaces involve inserting and attaching super-fine conductors. But if we could fully integrate an embedded read/write network into brain organoids, it would lead to fascinating possibilities — including revolutionize the study of brain behavior at the cellular and sub-cellular level.
But the potential innovations don’t stop there. Consider the possibility of incorporating “sensory modules” into the brain organoid matrix — biological modules that can respond to optical signals maybe, and even translate images into processable data.
This is not as far fetched as it might seem. As the paper’s authors write, “A sensory organ, such as a retinal organoid, could then be connected with a brain organoid. Eventually, networks of organoids will be interconnected to implement more complex [organoid intelligence]. The organoid will be interfaced with electrical and fluidic-sensing and simple outputs controlling machines through biofeedback on the cellular level; i.e. giving the brain organoid control by feeding back the results of its induced actions.”
In other words, the future of organoid intelligence will likely lie in the integration of many different types of organoids, all with their own specialisms and functions … much like in the human central nervous system.
Scale is still a challenge though. Networking together multiple organoids allows some degree of scalability, but ideally the biological computers being envisaged would be based on organoid brains that are tens of millimeters to tens of centimeters in diameter.
For this to be possible, new approaches would be needed to supply oxygen and nutrients, and remove waste products like heat and metabolic byproducts. In other words, the ideal brains behind organoid intelligence would need the equivalent of a built-in vasculature.
This, though, is not beyond the realms of possibility. Interestingly, it was precisely the challenge of heat extraction and embedded vascular-like systems that led to me speculating about the possibility of 3D printed artificial brains some years ago in the journal Nature Nanotechnology. It’s interesting to see 3D printing and brains once again converging — but this time with the brain component being very much a biological one.
It’s these possibilities and more that led to the authors of the Frontiers in Science paper advocating for the new field of organoid intelligence. Such technologies do not currently exist. But they might so with sufficient funding and focus. And they could become a game changer for advances in AI as well as in understanding the human brain as they emerge and mature.
Of course, there will be more than a few challenges along the way, including those at the intersection of what’s possible and what is responsible. And the paper’s authors are well aware of many of these. Despite the intellectual appeal of growing brains in vats to act as superintelligent supercomputers, there is a societal side to this research that suggests the pathway forward won’t be an easy one.
Even if the reality of organoid intelligence doesn’t track with some of the more extreme science fiction narratives, there are too many sci-fi tropes where brains grown in vats do not turn out well to avoid uncomfortable territory here. There is also a very real possibility of a “yuck factor” response here — a term coined by bioethecist Art Caplan to describe an instinctive abhorrence toward some technologies and their uses.
Beyond the sci-fi and a possible instinctive public backlash, the challenges of experimenting responsibly and ethically with organoid intelligence are complex.
There are questions for instance around whether brain organoids could feel pain, or develop self-awareness, or exhibit consciousness or other attributes that would indicate they should be conferred with certain rights. These are questions that the paper’s authors are at pains to stress are critically important, and that demand integrated approaches to ethical research and development of organoid intelligence.
When the paper was published back in February, I suspect that many of the authors’ articulations of these concerns seemed reasonable. However, watching the flurry of questions emerging around how large language models and other foundation models are shaking up underlying assumptions of mind and consciousness, these concerns — and their resolution — may need revisiting.
I also suspect that the ethics of organoid intelligence won’t get us all the way to where we need to be if the science and technology are to be placed on a responsible footing. As with more conventional AI, there are likely to be complex risks associated with developing biological computing-based AI systems. And when these systems are based on lab-grown brains that demonstrate human-like behaviors, these risks (including moral risks) are likely to get a whole lot more complex.
At this point I have no idea how organoid intelligence is going to play out. I would suggest, however, that now is the time to be thinking about socially responsible research and development around organoid intelligence, including how the field is governed, how democratic decisions are made around what is done and is not, and even where the boundaries of appropriate, inappropriate, and definitely not appropriate, research lie.
That said, the possibilities here are intriguing, and I would strongly recommend reading the paper — if only to have a heads up on an advanced technology transition that may be transformative in positive ways, but may also turn out to be another bumpy one.