WHAT IS THE MASTER ALGORITHM
If the master algorithm existed then what would it look like? What form would it take? When the real answer to these two questions is found we will be that much closer to actually inventing the it. Over the years there have been many theories as to what the universal learner would look like and how it would operate, in this post we will attempt to isolate a couple of the possible candidates for the master algorithm and break them down.
THE HUMAN MIND
The first thought that usually comes to mind when thinking about universal learners is the fact that a human brain seems to closely resemble a biological version of exactly what we are looking for. In more ways than one our minds are our own master algorithms, if you examine closely at how our brains are wired it appears our brains use the same “algorithm” for virtually every cognitive task they perform.
Evidence for this has been observed in blind people whose visual cortex tends to take over other brain functions in order to seamlessly adapt to ever changing circumstances. There have been reports of blind people learning to “see” by using various techniques like clicking their tongues which is similar to how bats see in the dark. This means that areas of the brain that are designed for each specific sense are only distinguishable by the different inputs they are connected to and each of these regions are connected to executive areas of the brain that repeat a similar wiring pattern.
SIMILARITIES BETWEEN OUR MINDS AND ALGORITHMS
Brains of humans have been examined under microscopes to confirm that the same wiring pattern is repeated throughout the entire cortex. It is my understanding that the cortex is organized into columns and layers with wiring and feedback loops running to other brain structures with recurring patterns of various connections. These connections aren’t all exactly the same but when you really think about it these similarities represent more than one variation of the same algorithm as opposed to different algorithms. This supports the theory that our minds are programmed with a universal “algorithm”.
The cortex is also known to take over the function of another part of the brain that becomes damaged, this has been observed in cases where people have suffered strokes or injuries involving blunt force trauma. There are other aspects of the brain that seem to have similar patterns throughout including the rate of firing between neurons which is largely responsible for memory formation and learning. These patterns have not just been observed in humans but animals as well, Our brains are obviously larger and more complex but ours are designed with many of the same principles as other animals.
CREATING A UNIVERSAL LEARNER
What does this all mean? Well in order to create a universal learner we could simply figure out a way to map out the human mind and create mechanical copies with the master algorithm powering all the cognitive processes. Unfortunately it’s not that simple. Even if we could accomplish something like that such an algorithm would be limited to only what the human mind is capable of learning and since there are many things we already have innately learned through evolution and many things that we can’t learn since we don’t know they exist and the fact that what exactly the human brain is capable of learning is still being debated among scientists and the truth of the matter probably won’t be discovered any time soon.
Then there is the fact that the brain is extremely complex and when it comes to understanding how our minds work we have only just scratched the surface, however, mapping out the mind is not the only approach to creating the master algorithm, there are many more theories and possibilities that exist.
THE MEMORIZATION PROBLEM
This next theory is more of an incomplete piece of the puzzle because alone it does not solve the problem of the master algorithm but when combined with the foundation of machine learning research over the last few decades it may still prove useful in the future. A seemingly obvious idea is that if you were able to flawlessly memorize everything you’d ever seen and memorize every bit of information that was available then at some point you would know everything there was to know, designing a computer program around this model could be
a useful first step. However this comes with a myriad of problems like the fact that drawing upon information from the past doesn’t truly provide you with enough information to understand or prepare for the future.
Even if you existed since the beginning of time and memorized absolutely everything from then up to this point that knowledge would pale in comparison to vast possible futures that await you. If an algorithm was only designed to memorize it would have severe limitations but what if we could combine an idea like this with a capability to use human-like intuition to predict future outcomes from a database of past knowledge as opposed to just looking for patterns in the past where patterns may not have even existed? Then, it might be doable but most experts are moving away from theories like this in favor of ones that are more realistic.
MACHINE LEARNING POSSIBILITIES
There are many possibilities when it comes to candidates for the master algorithm and it’s very fortunate that at this point we have many decades of research to build upon. Many experts have dedicated their entire lives to writing algorithms. With each new generation building upon the work of previous ones we can be certain that the ultimate discovery could be just around the corner. In this post we have taken a look at two of the many possibilities of the future master algorithm, throughout the course of this series we will attempt to completely break down the concept of the master algorithm and figure out what complete singularity would mean for all of our lives. In the next post we will cover some more theories and also examine some more evidence that suggests that a master learner is indeed possible to discover.