In the latest wave of extraordinary computer development, tiny ‘brains’ are being grown in dishes and connected to digital interfaces to perform tasks. In Cosmos Print March 20205, Ken Eastwood took us to the lab to meet the brains behind the research.
In season 2 of Star Trek, released in 1966, on a fictional planet called Treskelion, a fresh-faced William Shatner comes across a large bowl-like structure with 3 alive, connected brains encased inside it. Almost 60 years later, Professor Thomas Hartung of Johns Hopkins University in the USA points to this in an online forum on organoid computing, as an example of how teams like his are making the unbelievable come true.
“Captain Kirk encountered in 1966 a computer built on brains, which was actually biological computing, which at the time was pure science fiction and now we try to make this a reality,” says Hartung.
Around the world, progress is rapidly being made in organoid or biological computing – the meshing of specially grown brain cells with digital platforms.
In 2022, Australian researchers at Cortical Labs successfully got their ‘Dishbrain’ to play the simple computer game of Pong.
Last year, Chinese researchers revealed that their open-source biocomputer MetaBOC is using human brain cells to control robots. Companies such as the Swiss Final Spark and Cortical Labs are now developing commercial hardware – plug-in brain computer systems – that you can buy or use via the cloud.
And, at Johns Hopkins University in the US, and other institutions around the world, researchers are rapidly developing everything from ethical frameworks for experimenting on these new biocomputers through to new methods to engage with and ‘teach’ them.
Proponents say the practical uses for organoid computers could be legion. Imagine, for example, if you were testing a new drug, say for Alzheimer’s, epilepsy or dementia, and being able to directly ask an organoid computer how the drug was affecting it.
Researchers are excited by other advantages too: potential computing speed, processing and energy efficiency. Human brains may be slower than machines at processing simple information, such as a maths problem, but they far surpass machines in processing complex information, particularly when there is uncertain data.
Unlike silicon constructions, humans can perform parallel processing (running separate parts of an overall task at the same time). It takes supercomputers many minutes to model even a fraction of a human brain, and they use massive amounts of power to do so – roughly a million times more energy. It would take the construction of huge new power plants to digitally replicate the activity of even one human brain.
“The potential for [organoid computing] is massive and there’s nothing really we can compare it to,” says scientific futurist, tech influencer and author Dr Catherine Ball. “The human neuron always beats the silicon creation – it’s always the best way. So, using the best kind of cognition – the human brain – in a tiny form just seems logical. It makes so much sense.”
The DishBrain chip, containing human cells. Credit: Courtesy of Cortical Labs.
Organic development
Professor David Gracias, Johns Hopkins University. Credit: Courtesy of Professor David Gracias.
Professor David Gracias, from the Department of Chemical and Biomolecular Engineering, is one of 20–30 lead researchers at Johns Hopkins working directly or indirectly on biocomputing.
He points out organic computing hasn’t just developed in isolation, but has progressed at this time because a whole suite of technologies have come together simultaneously – the ability to keep cells alive in a dish for months at a time, the power of AI to help run multiple experiments and interpret the masses of data they produce, and micro wiring at the digital interface to connect to the organoids.
Gracias says that the Johns Hopkins researchers began working on organic computing a few years ago. “If humans are trying to emulate the brain in a silicon chip in modern day neuromorphic computing, why not just emulate the brain with the brain itself, right, instead of trying to mimic it?” he asks.
The organoids he works on have about 10,000 human brain cells – which together make up a tiny ball up to about 0.5mm across. Elsewhere in the world smaller or larger organoids are being used. But even the process of keeping the organoids alive is time-consuming and thought-provoking.
“If you don’t have blood vessels, you cannot grow things that are bigger than a few hundred microns across,” Gracias says. “If they get bigger then the centre starts to die. So, we and others are trying very hard and we are developing instrumentation to see how to make blood vessels in them [to keep them alive].”
In addition, the organoids are ‘wetware’ – they need to be kept in a liquid medium to survive, which makes connecting digital computers with electricity to them a challenge. “Water is the enemy of computers because it shorts the circuit,” he says.
“This is going to be a paradigm shift in computers, which is to learn how to build computers which have water in them, which can keep the media ‘perfusing’ [or flowing].”
Another challenge that Gracias has been specifically working on is how best to digitally connect to the organoids. The connections are used both to stimulate the organoid (e.g., putting a current into a neuron) and to receive any electrical response.
In simple organic computing models, the organoids are kept as small, 2D structures, making it relatively easy to connect wires to neurons. However, to better create the multitude of activity that can occur in a brain as multiple neurons interact with each other, 3D structures are required. Placing one of these little brain balls of activity on a 2D interface limits the digital contact with the structure to just a couple of points – not an ideal outcome when you are trying to monitor what is going on throughout the structure.
“We’re going after the tough Holy Grail problems,” Gracias says. “One is to stimulate and monitor every cell in the organoid in real time.” Taking inspiration from an EEG skull cap used to measure brain activity, Gracias worked on a tiny shell that could wrap around the miniscule 3D brain organoids, recording activity at many points.
“One of the problems with the mini EEG cap is there are a lot of wires on them,” he says. “So, we’re now also thinking of electronic tattoos or non-contact or wireless, so we don’t need all the wires… Maybe we can use light or we can use electromagnetic fields.”
Of course, merely sending and receiving signals from the organoids doesn’t mean we necessarily know what those signals mean, so many current studies involve sending millions of signals into the organoids in order to understand how they respond, and how those responses change over time, i.e., how they ‘learn’ to respond to something like a game of Pong.
“You and I speak in English, but the brain speaks in a language of chemical and electrical signals,” Gracias says. “We have to learn the language of the brain, and that is not a trivial thing. We can stimulate and we can record, but then we don’t know what to stimulate and we don’t know how to interpret the recording. So, we are learning this language. There’s a lot of data that comes out of it, which is why we need data scientists, because brain organoids fire on the timescale of milliseconds, so you get a lot of data very fast.”
Gracias says teaching the organoid computers a task is a bit like teaching a small child, with rewards and punishment. “Basically, we have to find out what the brain organoid likes. And you give it a pattern and if it repeats the pattern, you give it a reward of some kind. It could be an electrical spike or it could be a chemical. Then if it does something you don’t want it to repeat, you kind of give it something it doesn’t like.”
Some of the current research Gracias is involved in includes giving the organoids neurochemicals such as dopamine. “So, we are trying to create a multi-modal shell which can stimulate and record both chemical and electrical stimuli. We’ve made a lot of progress there.”
Measuring activity in organoids: A tiny shell can be wrapped around organoids to detect their activity (in microvolts, μV). Wrapping 3D organoids with a shell allows researchers to detect activity at multiple points at once.
Source: Gracias Lab, JHU Image reprinted by Huange et al. Sci Adv 2022 (top)
Source: Image reprinted from Smirnova et al. Frontiers in Science, 2023
Scaling up
Dr Brett Kagan, Cortical Labs. Credit: Courtesy of Cortical Labs.
After making a splash with their Pong-playing ‘Dishbrain’, Dr Brett Kagan and the team at Cortical Labs pivoted the company’s goals to extend beyond pure research, and instead, provide a viable, commercial organoid computing product that others could use.
Looking like a computer mainframe or server, the product that is being prepared for commercial release has 40 biocomputing units in racks, all joined together digitally with 60 contact points per culture. At this stage, the individual cell cultures are expected to survive for 6 months or so before needing replacing. In the next year or so, large institutions will be able to buy either single units or whole server racks for their own research programs.
Even now, institutions around the world are using the existing Cortical Labs units for research via the cloud. “We’ve really expanded, predominantly over the past 12 months or so,” Kagan says. The company now has 16–17 people on the team. “I suspect we’re still at the very starting stages of the snowball.”
He says that while Cortical Labs is still very much involved in its own research with organoid computing, the founders of the company realised that many more scientists needed to be able to access working organoid units to progress research in this field.
“It’s been really reassuring to see a lot of other people sort of start to circle back around and engage in this in a far more serious and deeper way since early work established the viability of this approach,” Kagan says.
“We alone can’t answer all the basic science questions, and if we’re committed to making this industry work, what we have to do is actually enable others to be able to ask and answer their own questions.
We’re very collaborative – whether it’s with other companies or with labs – our goal is, how do we get this technology in the hands of as many people as possible?”
Kagan says that in historic gold rushes, the people guaranteed to make money were those who sold picks and shovels. “And that’s been our approach. We don’t have to be the ones to find the gold if we can actually just help enable access.”
Part of the impetus for the shift in the company’s direction was as a result of struggling using off-the-shelf hardware and software that wasn’t specifically designed for working with biocomputers. “We needed to take a step back and really understand what we needed,” he says. “We had to build our own hardware from the ground up that we could actually use to interact with this work.”
It took 12–18 months for a talented developer to build up the Pong environment that an organoid could successfully respond to, but Cortical Labs can now help researchers develop similar projects “10 times, even some cases 100 times quicker if needed,” he says. “It takes maybe a week or two to build that code that previously took a year because we’ve just abstracted away so much of the technical complexity.”
Kagan says part of the joy of the research over the past few years has been understanding how the brain cells react to stimuli. “We’ve been building up these patterns of behaviour, but every time we think we know a thing that’s going to happen, we find out well, actually sometimes and sometimes not. So, it’s just this continual process of discovery. It’s really interesting just continually learning, which is the beauty of science, isn’t it? If we knew all the answers, then we wouldn’t be doing anything novel.”
Research currently using the Cortical Labs platform includes breaking up the organoids’ response signals into different sorts of spikes – not just a binary on or off, but a sharp curve or shallow curve etc. “We’re looking at not just the beats, but the notes,” he says.
Kagan emphasises that a unique approach of Cortical Labs is that it is exploring both 2D and 3D organoid computing. In particular, the 2D approach allows researchers to better understand individual components, rather than having to wrestle with the more complex interactions going on in more brain-like 3D structures. “We don’t need the whole brain for biology to be useful. Using bioengineering techniques, we can construct discrete neural circuits for bespoke purposes.”
Other advantages for 2D bioengineered computing, he says, are that it provides more scalability and control, and ethically it’s less likely to raise some of the issues people are concerned with, such as organoids developing consciousness, or feeling pain. “Why would we seek to create human intelligence, when we don’t have bee or worm intelligence yet – and you could do a lot with that.”
Brain cells like these are being grown by Cortical Labs for use in computing. Credit: Courtesy of Cortical Labs.
Wired and ready
Organic computing is “here to stay and it’s going to change the world,” says Dr Fred Jordan, whose company Final Spark is also allowing online access to its banks of 3D organoids. You can even watch experiments happening live on screen on the company’s website. “The neurons that we have can be accessed remotely by researchers and we are sharing those neurons with 9 universities,” Jordan says. “Now we are using a booking system where all the universities can book some hours, including us. So, they are used 24/7.”
Jordan says that part of Final Spark’s unique approach is that their organoids are in an “air-liquid interface. That means the organoid is not immersed in a liquid medium. Most of it is actually out of the liquid in contact with air, but there is still a very thin layer, 100 micrometres, of a liquid medium at the surface and it’s sitting on the porous membrane where the electrodes are.”
Each Final Spark organoid is connected by 8 electrodes, which he says is plenty for current research. “We’re not going to wire millions of axons and dendrites by ourselves… What we have to do is to find the right set of stimulations, whether electrical or chemical, in order that the whole network reconfigures itself in order to provide the correct answer for a specific input, which is the definition of learning, right?”
He says part of the challenge and beauty of organoid computing is that the clusters of cells change over time. “If you stimulate simultaneously all the electrodes, the answer will depend on several things, on the organoids and on the time of the stimulation. If you do this experiment on the first day or on day 100, for instance, the reaction will not be the same.”
Do dish brains sleep?
Gracias at Johns Hopkins says a common question he gets asked about the organoids is whether they need to sleep. “I don’t know. I mean, what is sleep?” he says. “These things create very profound questions, like even what is consciousness?”
Jordan says Final Spark began looking at the issue of sleep for their organoids about 6 months ago and now all their organoids have a rest in every 24-hour period. “We were considering this like a machine, but it’s probably a mistake,” he says. “This is a living organism and we have to take care that it’s in a good situation to perform tasks. We are not forcing sleep, but with some molecules we may induce a state similar to sleep during some periods. I think it’s really appropriate to ask ourselves these kinds of questions.”
Hartung and the Johns Hopkins team led an international forum that created and published in 2023 the Baltimore declaration toward the exploration of organoid intelligence – an ethical framework around moving forward in this relatively new industry. It notes that “we also need to anticipate… and address the significant and largely unexplored ethical challenges associated with this research. We must be alert to any possibility that organoids could develop forms or aspects of consciousness.”
Other ethical concerns centre around the minefield of when an organism develops sentience or consciousness, and whether the organoids can feel pain, particularly as external ‘feeling’ devices, such as light or chemical sensors, are rigged up to them.
Kagan at Cortical Labs says all the researchers who worked on the Baltimore declaration are working alongside ethicists every step of the way. “We’re not waiting until outrage happens. We’re figuring out how do we build this in concert with not just the ethics community, but society at large.”
The worst fears about organic computing fed by dystopian science fiction include a deep-seated horror that these organic computers could gain sentience and control. Even in the 1966 Star Trek episode, the 3 brains in a dish that Captain Kirk encountered were actually advanced, evolved life forms that had no need for bodies, and were now using other species – including humans – in lethal gladiatorial contests for entertainment.
Jordan says we needn’t worry about that too much at this stage. “The thing is that at this point we have few thousand neurons. So, we are more in the realm of the mosquito than that of a human being. But it’s still a good question to ask for the future because the scalability is there.”
When asked about time frames, and when organic computers will be able to be used for more complex tasks, Gracias smiles. “I think there’s a joke that academics always say, ‘5 years or something’, but you know, it’s very nonlinear. There’s a quote attributed to Yogi Berra which says, ‘It’s difficult to make predictions, especially about the future’. So, we really don’t know.”
