Last year, Monash University scientists created the "DishBrain" – a semi-biological computer chip with some 800,000 human and mouse brain cells lab-grown into its electrodes. Demonstrating something like sentience, it learned to play Pong within five minutes.
Computer chip with built-in human brain tissue gets military funding::undefined
Overall, I think this research was well-conducted and meaningful, although use of the words “intelligence” and “sentience” are deliberately sensationalist. In the context of the paper, they mean that it is responsive to its environment in the same way a sea sponge or Tomagotchi is.
The cells are neurons cultivated from dissolved rat brains (after they were gently euthenized) and reprogrammed forebrain neurons cultured from cells collected from a baby’s foreskin after circumcision (which isn’t unusual even though it’s weird if you don’t work in this field).
Then, the cells were transferred to a commercially available plate with thousands of little electrodes on it that can read the random electrical pulses healthy neurons in a dish routinely shoot off, and also stimulate these. The team then used a program to take the input value from a bunch of the readings and map one area of the plate’s activity to moving the paddle up and one section to down, and then had another region that was given stimulation of random noise except when the ball hit the paddle. Then they tried to quantify if this loop incentivized the neurons to coordinate to try to increase the number of ball strikes. I don’t know if it could read the ball position or just got a “reward” when it collided.
Really, they expanded an existing set of tools elegantly in a way that can improve our ability to study neurons in a dish. Which is great. Culturing neurons well is harder than people like to admit. BUT: nothing about this currently advances computer science in a specific way. It might in ten years, but so might superconductors. There’s nothing about a biological network that is inherently superior to other informational networks, except that they’re currently the underlying component of the most successful informational networks we know of. And, of course that they get you hot, hot, HOT press, which is certainly nice when you’re trying to get money to do research.
Yes, except there’s also a line where the doc says, ‘will you attest to the institutional review board that you won’t use this to make a Kaiju/clone Hitler/travel through time and kill your grandfather, etc?’
Those rules only exist because somebody couldn’t help but travel through time to kill their grandfather Hitler who may or may have not have been a clone. It happens every time they don’t enforce it.
The plate is a consumable dish that is designed to fit into this guy: Maestro Pro.
It’s kind of the iMac Pro of reading electrode signals from cells. It’s a high end piece of lab equipment, and popular for this kind of work. Which is great: it means that this work is more replicatable, as I’m sure lots of people are going to want to try variations on this. There’s a good chance my old boss is at least considering it right now.
Reward in this context means a nice, soothing syncronized electrical pulses. I’m curious whether more informed experts would validate that this constitutes a desirable effect, but that’s what they went with. The alternative was random noise.
For you, this would be like, bad: four seconds of audio static. Good: four seconds of metronome.
Having worked in neuroscience and synthetic biology I want to give some context.
First, here is the actual paper: https://www.cell.com/neuron/fulltext/S0896-6273(22)00806-6
Second: is this bullshit?
Overall, I think this research was well-conducted and meaningful, although use of the words “intelligence” and “sentience” are deliberately sensationalist. In the context of the paper, they mean that it is responsive to its environment in the same way a sea sponge or Tomagotchi is.
The cells are neurons cultivated from dissolved rat brains (after they were gently euthenized) and reprogrammed forebrain neurons cultured from cells collected from a baby’s foreskin after circumcision (which isn’t unusual even though it’s weird if you don’t work in this field).
Then, the cells were transferred to a commercially available plate with thousands of little electrodes on it that can read the random electrical pulses healthy neurons in a dish routinely shoot off, and also stimulate these. The team then used a program to take the input value from a bunch of the readings and map one area of the plate’s activity to moving the paddle up and one section to down, and then had another region that was given stimulation of random noise except when the ball hit the paddle. Then they tried to quantify if this loop incentivized the neurons to coordinate to try to increase the number of ball strikes. I don’t know if it could read the ball position or just got a “reward” when it collided.
Really, they expanded an existing set of tools elegantly in a way that can improve our ability to study neurons in a dish. Which is great. Culturing neurons well is harder than people like to admit. BUT: nothing about this currently advances computer science in a specific way. It might in ten years, but so might superconductors. There’s nothing about a biological network that is inherently superior to other informational networks, except that they’re currently the underlying component of the most successful informational networks we know of. And, of course that they get you hot, hot, HOT press, which is certainly nice when you’re trying to get money to do research.
Scientist: Hey Doc, you gonna use that foreskin you just cut?
Doc: No, go right ahead!
Scientist: thanks.
This is how I imagine this conversation goes down.
Yes, except there’s also a line where the doc says, ‘will you attest to the institutional review board that you won’t use this to make a Kaiju/clone Hitler/travel through time and kill your grandfather, etc?’
And then they give it to them.
Those rules only exist because somebody couldn’t help but travel through time to kill their grandfather Hitler who may or may have not have been a clone. It happens every time they don’t enforce it.
What does “commercially available plate” and “reward” mean in those respective contexts?
The plate is a consumable dish that is designed to fit into this guy: Maestro Pro.
It’s kind of the iMac Pro of reading electrode signals from cells. It’s a high end piece of lab equipment, and popular for this kind of work. Which is great: it means that this work is more replicatable, as I’m sure lots of people are going to want to try variations on this. There’s a good chance my old boss is at least considering it right now.
Reward in this context means a nice, soothing syncronized electrical pulses. I’m curious whether more informed experts would validate that this constitutes a desirable effect, but that’s what they went with. The alternative was random noise.
For you, this would be like, bad: four seconds of audio static. Good: four seconds of metronome.
What’s the raspberry pi of the industry?