I read an excellent article on Wired the other day on the topic of organoids and thought I’d share a bit about it here. Even the basic concepts of this topic feel more like science fiction than reality, and that is right in my wheelhouse.
I’ll try not to directly repeat too much of what the author of the article said: for details refer to that larger article. Instead I’ll be summarizing then adding some of my own thoughts on the topic at the end.
What are organoids?
Simplistically, organoids are any collection of cells from an organ grown under laboratory conditions. These cell bundles act as artificial analogues to full organs in most senses: they react just like the real thing to stimuli like chemicals and radiation. This makes them excellent for the purpose of drug testing as they can be made from actual human cells making them superior to animal testing.
These little organs are living in the same sense that the larger organ they are related to, and must be provided with nutrients and waste removal to remain living. They are limited in size, however, due to the lack of a functioning circulatory system for these very needs. Only a few layers of cells can be kept living as a result, although there are investigations underway into biomimetic techniques to provide a circulatory system for organoids of various types.
In the case of the article the focus is on organoids made up of neural tissue. The neural tissue is produced by taking common cells from the body like skin or blood and inducing conditions that render them pluripotent i.e.: capable of becoming any kind of cell. Then the cells are further encouraged to become neurons. If created from human cells, the collection is called a human brain organoid, and it can do some fairly interesting things.
The primary purpose today for human brain organoids is to use them for medical testing. Things like drug testing are constrained with animal test subjects because the human brain differs so greatly from most animals. Testing on animals that are similar to us such as primates often has implications regarding animal cruelty. The organoids provide a living human brain sample, albeit on a very small scale, that can be used for fairly sophisticated levels of medical testing.
Organic computer intelligence
Human brain organoids possess some unique characteristics that lend themselves to more complex forms of experimentation. Notably neural tissue is very strongly ‘motivated’ to interconnect with other cells, so a group of such cells rapidly forms a more complex interconnected whole just like the human brain. These neural organoids also react to multiple forms of stimuli in complex ways: key to computing applications, they react to electrical stimulation.
As noted in the previous section, there is currently a hard limit on the maximum size of organoids in general due to lack of a circulatory system. For human brain organoids this limit means a maximum of about 2.5 million cells or a little blob about half a centimetre in size. To date this is the largest neural organoid that can be experimentally tested for possible computing applications.
Tests have been done with the current largest size of human brain organoid that demonstrate nascent processing capabilities in the tissue. Experiments whereby stimuli were encoded electrically resulted in an organoid that could play a passable game of Pong. Some tests have expanded this capability to encourage an organoid to play a rough game of Doom, navigating Doomguy through a level and shooting monsters.
The idea of having an organoid play a video game seems on the surface to be ridiculous, but it does have a secondary purpose. The drug testing mentioned earlier can piggyback on such experiments: a drug-addled organoid might begin to play Pong less accurately. Interestingly some testing has been done with potential Alzheimer treatments to see if they affect human brain organoid game related memories.
All of this is very early experimentation. It is unknown whether larger human brain organoids might be more capable, perhaps performing tasks like complex image recognition or solving Captchas. It is also unknown whether the techniques to increase size limits for organoids in general will ever bear fruit, or if a different technique such as some sort of network processing could increase organoid capabilities. Regardless, the developments with organoids are coming quickly and new challenges, both ethical and technological, are a near certainty.
My thoughts
Human brain organoids are a remarkable development. Just the technology that induces human blood or skin cells into a pluripotent state to allow their creation is amazing to me: it overcomes the need for embryonic cells and eliminates that source of moral outrage. But I am uncertain of their future capabilities.
What we have now is tiny clumps of grey smaller than a pencil eraser that will quite literally die if conditions such as temperature and chemical balance aren’t carefully maintained. You can’t run such a thing in your home: in fact there is a company called Cortical Cloud in Melbourne that provides access to individual organoids as a service, quite similar to current quantum computing services.
Human brain organoids themselves probably must be larger to be effective for large scale computing , which requires solving a rather sticky problem with keeping the little things alive. Artificial blood supplies may be created, but how much more complex will that make maintaining the health of the individual organoids? And will a larger organoid actually have more capabilities?
So there are a lot of interesting questions to solve just in terms of the mechanics of creating and keeping organoids alive. But as they get larger we probably start encountering difficult ethical questions. The current organoids with around 2-3 million neurons have about the neural capacity of a bee, which feels safe to experiment upon. What happens when they get to mouse size? Will keeping a small “brain” of that size without outside stimulus in itself be considered too cruel? Or is it just a machine, a disconnected organ with no feelings or mental processes due to its disconnected state?
I can’t answer the ethical questions, but they do give me pause. The development of human brain organoids seems primed for several leaps forward and it is probably important to at least consider such questions. Setting them aside for a moment, however: do I think a human brain organoid could outperform traditional large language model AI? For convenience, let’s agree to call human brain organoid computing “Organoid Intelligence” or OI.
Yes, I think OI could outpace current AI, but it would require an entirely new form of codification of problems to ‘educate’ the bundles of cells in the problem space we want solved. Vast data stores with trillions of documents won’t work to train a few million or billion neural cells in a performant way. Maybe the OI system could sit in a supervisory role over an AI system, providing encoded value judgements on results based on training data for truth and ethics. I feel like OI would be more capable of making that kind of judgement than AI is.
Certain dedicated areas like image recognition, pattern recognition (e.g.: captchas), and possibly creative thinking (i.e.: artificial general intelligence) seem like areas where OI could excel. I am basing this assessment on the assumption that human brain organoids will continue to possess characteristics of the human brain as they are scaled up, but that may be an entirely false assumption. OI could just be “AI but slower” if we try to do things the same way with organoids as we do with RAM and GPUs.
The header image for this post was AI generated using ChatGPT