The human genome sits at roughly 700 MB of data if converted from base pairs to binary data, or, to put another way you have the entire instruction set for the creation of the most adaptable creature we have ever encountered contained in the same space as a CD. This sort of sticks out in the filed of developing AI. Current methods seem fixed on the idea of having enormous databases instructing what the AI should say and do. Is this correct? If the goal of creating AI is to simulate a human like mind, then how can the answer possibly be to use terabytes of information to create it?
It appears that evolution has come up with a clever way of creating intelligence that doesn’t require much information on the actual world the creature is being put into. Our brains can adapt to such a wide range of environments, languages, talents, and situations. Nature has programmed the base information for this incredible machine with a fraction of the 7o0 MB defining the human. Nature and lazy coders like to copy as much as it possibly can in order to be more efficient. It is not by chance that a majority of things in nature have structures defined by recursive functions. A trees branch is self similar to the rest of the tree. Copy and paste. By defining recursive and fractal like functions nature has defined complex structure and solutions with elegant and simple lines of code requiring little storage.
Those familiar with machine learning will be quick to point out the similarities of neural networks and the human brain. In my opinion, no matter how well we define a neural network in code it cannot achieve the complexities of reality. Reality, despite our best efforts, has a lot of chaos that we normally throw out as noise. But when a computer becomes physical it can utilize the chaotic nature of reality. For instance, using genetic algorithms to configure a 100 Logic Gate FPGA to identify a 1 kHz and 10 kHz tone yielded a chip that worked with some completely bizarre settings. The chip had circuits not even connected to the actual circuit, but when those were changed the chip no longer worked. That’s like your car not working because your neighbor is out of gas… Sort of. Similarly, when the configuration was loaded onto another FPGA it failed. This seems to indicate the randomly evolved circuit was using the EMF interactions of the circuit at a level no one would ever intentionally design for.
We seem obsessed with the idea that we can create a simulation of the brain if we simply scan in the geometry of the human mind and click go. Referring back to chaos, our brain does not exist in a vacuum. Our brain swims in a bath of chemicals and happens to be attached to a rather busy body, but so what? It seems that beyond the obvious chemical interactions of exercise and food even that bacteria in our gut has an effect on our brains. Accurately simulating a human mind would require taking into account all the interactions of the environment, the chemicals in our mind, bacteria, social interactions, physical cues, and those are simply the ones we know about.
Being a lazy programmer myself, I wouldn’t want to program an AI using this approach. I want to know how nature does it. I wish I could just copy the code from DNA, but that will be covered in a future post as being near impossible. A more in depth look into the possible AI revelations based on DNA will come in the next few weeks. I just felt compelled to delve into the fascinatingly limited amount of data the worlds best intelligence is derived from.