- The CL1 ‘biological computer’ has been shown playing Doom
- This is essentially 200,000 human neurons plopped on a microchip
- This brain soup has apparently learned to find and shoot enemies in the game, although skeptics are arguing about the definition of ‘playing’ here
Living human brain cells are capable of playing Doom, and by that, I mean a bunch of cells not inside a brain (of course, that wouldn’t be much of a revelation), but in a lab — though there’s still some debate about what the definition of actually ‘playing’ a game is here.
As posted on Reddit, this is the work of Cortical Labs, a firm that used a “multi-electrode array” to achieve this feat, which is essentially 200,000 human neurons plopped on a microchip.
This is otherwise known as the CL1 ‘biological computer’, which is a “high-performance closed-loop system where real neurons interact with software in real time”. In this case, Cortical Labs worked with an independent researcher called Sean Cole to get the CL1 running Doom — or rather, playing Doom. Well, sort of, and therein lies some of the controversy.
Article continues below
These brain cells aren’t playing the game via some sort of controller, but rather, they are very directly engaged in navigating Doom’s maps and dealing with its grumpy demonic inhabitants.
What happens is that the video feed of the game is directly mapped to the neurons using patterns of electrical stimulation. As Cortical Labs explains in the YouTube video below, demonstrating this, when a monster appears on the left side of the screen, certain electrodes stimulate the left side of the sensory area of the neuron soup in the CL1. The neurons then react to the stimulation, and that cell activity is interpreted as motor commands, meaning controlling the Doom guy (moving and shooting).
Four years ago, Cortical Labs ran a similar experiment with Pong, although Doom is obviously a lot more complicated.
Analysis: learning skills
While this is fascinating, it’s also a very abstract way of playing Doom. Theoretically, the brain cells are learning to play the game, albeit using a very direct interface as noted.
As Cortical Labs explains in the video, the CL1 is capable of finding enemies and shooting them in Doom, or spinning around if getting hit from behind — but it plays the game badly. If you watch the footage, there’s quite a lot of bumbling around and shooting at walls. As Cortical Labs puts it: “The cells play a lot like a beginner who’s never seen a computer,” adding, “and in all fairness, they haven’t.”
That makes it all look pretty random, which is why some of the detractors on Reddit aren’t so impressed with the demo, essentially saying that we have no proof that the neural soup is actually honing its skills, or gaining an ‘understanding’ of some sort about the game. Sure, it’s reacting, they argue — but is it actually reacting, or learning, in any meaningful way? Or is it just waving a pistol around, blasting here and there, and getting lucky some of the time?
Still, whatever arguments there are about intent here, or what’s actually happening under the surface of the neural soup, Cortical Labs concedes that, as far as Doom goes, the learning skills of the brain cells need to be improved, but it’s confident this can be done.
For now, this is an eye-opening demo (albeit not the only one of its kind) — certainly more so than the Pong effort — and we shall see where Cortical Labs takes it from here.
As we saw last year, the company’s grand vision is to allow cloud access to its CL1 biological computers. You can rent a CL1 for $300 per week via Cortical Labs’ wetware-as-a-service offering, or buy one for $35,000.
The best laptops for all budgets
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds. Make sure to click the Follow button!
And of course, you can also follow TechRadar on YouTube and TikTok for news, reviews, unboxings in video form, and get regular updates from us on WhatsApp too.
https://cdn.mos.cms.futurecdn.net/Fe3R6YN2fFdoWkYtBQUAq4-1920-80.jpg
Source link
