The Brain Search Engine
The brain does not use a language to think. I can think of something without using the English language. For example, I can listen to music and enjoy it without repeating the statement, “I like this music” throughout the duration of the track.
We’ve used artificial neural networks to replicate the way the brain thinks. I think AI is more of a problem of deciding how the brain stores information rather than how it operates on it. We’ve always tried to create AI by simulating optimization and statistical regression to solve problems. I do not agree with that.
Think about this, what was the name of your class teacher in the 1st grade. Got it? What did you have for dinner yesterday? How much time did you take to answer both these questions? Somewhere around 5 seconds for the first if you remember it at all and the same or somewhere less time than that for the second question.
Hold up, if you ever thought of your memories as a table ordered by time, you’re wrong! The access time to both these memories was completely independent of the memories beyond the exact two occasions. You did not have to think about the name of your teacher in the 5th grade to remember the name of your teacher in the 1st grade right?
Doesn’t this make you feel that you are dealing with a more horizontal database? Is that why we resorted to a graph. That basically means you had a graph in your head where you could traverse from your current chain of thought to the section of your memories holding the name of your class teacher at almost the same speed as accessing yesterday’s dinner. So great it’s a graph! The study of neurons indicates its physically a graph. Cool! But wait, say one of these memories took more time to be retrieved than the other. Does that mean you had to travel a longer path to reach the memory for one? Or could it mean you were more confident about one memory than the other? Could be both right? Alright, so now we know that access time to memory depends on something about how confident we are about it too.
Ok, what was the name of your class teacher in the first grade? Got it quicker this time? So wait, whats different? Did your brain just create a shorter path to that memory? Unlikely, because, dendrites and synapses take time to form. That memory was indexed by charges flowing through the neurons. But again, how can a charge or a pattern of charges represent a path for traversal? Is your brain using something like a cache? Memories you fetch from your head which you haven’t thought about for a long time become super fast to access. There definitely is something like a cache at work.
Back to neural networks, we have numerical weights to store the actual logic. But there has to be an analogue equivalent in the brain right? The synapse length(gap between neurons)? But that is creepy right? So even basic motion can mess with that? I tilt my head, I should’ve lost all memories. But I don’t. Great, so the weights are definitely not represented by synapses. Awesome, so another option could be binary, viz. like connected or not connected. Definitely, all information can be represented in binary. Computers are a proof of that fact. But wait, what about confidence? Maybe that would be about the strength of signals flowing through the synapses. We’ve got tons of neurons, binary information? Completely doable!
I hope we figure it out someday!