Original article was published on Artificial Intelligence on Medium
The dawn of Artificial Bio-Intelligence
Believe me, it is not some sinister plan to win the world championship of criminal masterminds. Since that place is already taken by Gru and his minions. This thing is real and it is happening… like.. right now. This topic is quite old, with one very interesting work where “living neural networks” learned to control an aircraft in a game. The neurons from a rat embryo were grown on 60 electrodes. It was then electrically stimulated to learn how to keep the aircraft steady in the game.
In 2010, Professor Kevin Warwick grew rat brain cells on 100 electrodes. It could control a real-life robot over Bluetooth. Where the sensors on the robot feed into the biological neural network (let’s call it BioNN, shall we) and the output from the brain cells drives the robot over Bluetooth. One important thing came up in this study, that BioNN’s can scale pretty fast. In just a few days we could get 100,000 neurons from just a few neurons. Work is in progress for a larger few million neurons. Just imagine a small ant growing up to the size of an elephant in a few weeks, cool right??
In 2015, another interesting study could do blind source separation using living neurons. Blind source separation is a very tough job. It is like when you go to a party where there are a lot of people talking loudly, but you can still hear your friends just fine. Your brain is subtracting the background noise for you, without any training or funny math (signal processing) and also without overheating. how cool is that!
Well, we can do a lot of crazy fun stuff in labs but what about real-life business applications?
The short answer is that a lot of potentials is there. Like an Australian company, Cortical Labs is making really tiny brains and putting them in computer chips. These guys build from the rat brain(old fashioned way) and also by turning human skin into stem cells and then into neurons. It is possible guys, check out this paper about a technique that uses Yamanaka factors.
Another company called Koniku based out of San Rafael, California is developing a 128-neuron chip. Built using biological neurons, that can sense certain chemicals. The companies want to use the chips for military, surveillance and agricultural applications.
This begs the question, why do we need slimy biological neurons when shinny semiconductor chips are already there??
Well, you see(and I am oversimplifying) deep learning-based artificial intelligence has tried to mimic the brain. By modelling the neurons (called perceptrons, cool name huh) and then trying to run large quantities of these perceptrons in layers, as fast as possible. Computers for AI like CPUs, GPUs, ASICs, Neuromorphic are mostly stuck in a compromise between going crazy fast (aka speed and throughput) and the electricity bill (aka power and cooling cost). In 2016, AlphaGo consumed nearly one megawatt of power while playing with the world champion of Go. Our brain sniffs just 20 watts which is a whole lot less. So you see the benefits? The electricity bill goes down a lot.
If you want to know how much power consumption and carbon footprint your deep learning AI has, you should check out this slick web-calculator and the paper. Data is the new oil so it causes global warming too.
The other important factor is the inherent ability of biological cells to grow in growth mediums. It is impossible to make your computers grow in numbers just by throwing some food in front of it. So this makes biological neural network quite easy to scale.
Thirdly, again simplifying but no matter which type of computer we talk about, it fundamentally works to process batches after batches of perceptrons really really fast. But in the case of biological neurons, every neuron involved is running in parallel. So it can be fast while keeping your electricity bills in check or you don’t have to put a big battery for it.
So when are we getting these shinny…ahem.. sorry slimy chips in our smartphones?
Well, let’s just say it will take a while. There are few tactical challenges which still needs some work. Unlike their silicon cousins, for biological neurons, we can’t just feed them electricity. They need glucose, oxygen and the right temperature etc. So let’s keep our fingers crossed for a while because really exciting times are ahead of us…. right Gru?