In a world first, Australian firm Cortical Labs has produced a functional biological computer, It uses stem cell grown neurons directly connected to processors. This design is configured for commercial applications.
The top link is a real-time link to the current coverage of Cortical Labs.
Far more useful is the Cortical Labs website, which includes links to published papers and much more.
First, some background and the current state of play:
This organic link to computers is one of the holy grails of science fiction since forever.
In 2022 Cortical Labs created DishBrain, an assembly of neurons that learned how to play Pong. The real-time play indicates the fast learning and processing efficiency of the neurons. DishBrain apparently uses the same volume of neurons as the new biocomputer.
Cortical Labs has gone from DishBrain to this in two years.
The goal of the biocomputer is “organic AI” which Cortical Labs calls “Actual Intelligence”.
Having done far too much coverage of AI and its inexcusable hype myself, I can appreciate the distinction.
Think about the processing requirements and range of roles of AI for about a quarter of a second, and you can see how ambitious this is.
Science media had serious difficulty describing the whole idea of DishBrain. Thankfully, coverage of the biocomputer is more articulate and actually useful.
The current iteration of the biocomputer sells for $35,000. That price tag tells you everything you need to know about commercial viability.
It uses about 1,000 watts of power, far less than AI, blockchain, and the other horse-drawn standard processing technologies.
Those are just the basics. Take your time checking out those links. If you’re in any way technically inclined, it’s “ongoing learning”.
Meanwhile:
The power usage parameters are critical for future science. Neurons work on microvolts. Current tech’s power usage is relatively ridiculously inefficient and not at all competitive.
The learning capacity of the neurons is well established. How would a biocomputer do with a large language model? It works differently, but who’s going to argue with so much faster learning capability?
The hardware is extremely straightforward and therefore very efficient. Neurons are simply overlaid on the chips. You can see pictures on the links above. There’s no need to reinvent anything and everything and add layers of tech beyond the neurons.
If you’re somehow getting the impression that biocomputers could be the critical workaround for the spectacularly unimpressive clunkiness of current generation AI, yes. Even this very first model is a Get Out of Jail for researchers.
This is probably the equivalent of the Altair microcomputer, the world’s first PC, launched in 1975.
The only problem I see so far is in the descriptions. “Body in a coffin” isn’t what you want to tell computer users. It’s too easy for them to relate.
“BrainBox” is common usage, and pretty apt.
________________________________________________________
Disclaimer
The opinions expressed in this Op-Ed are those of the author. They do not purport to reflect the opinions or views of the Digital Journal or its members.
