I like solving problems. My goal in my life is to come up with an algorithm that solves the Artificial General Intelligence problem (there is a bit of obscurity in the actual definition of this term. Here I refer to the actual Human-like intelligence). I believe that AGI can be solved (without requiring extreme computation power. I.e., with the average computation power we have today) by searching for the solution a little different than it is usually done.

I believe, the best way to know About-me is to look at my code:

My github links: https://github.com/Aplokodika, https://github.com/sreramk

The biggest problem I face when thinking about AGI

The hardest part about trying solve AGI is, thinking about the human brain's functioning is a bit too frustrating, and exploring any potential end-point makes us feel that there could be no answer to this problem. I would describe the human brain as a "super recursive machine" that let's us think about thinking, and above all, every part of it is too interconnected and dynamic that one isolated piece of information in one isolated area could be responsible for altering another piece of information in a completely different area/domain. Also, it is a complicated equilibrium system, where every part of it is responsible for maintaining a persistent experience.

My say on AGI:

Most methods formulated today are highly optimized for pattern matching, at an abstract scale. Any complex corpus of information can be classified by taking their most common pattern, as a representation of a particular class. Therefore, every piece of data that contains that pattern is referred to by a class name. These "unsupervised learning" mechanisms which fine-tune it's pattern classification comes furthest in solving the problem, when compared to all the other known algorithms.

Eg., Works like Alpha Go, rely on "rebuilding" their assertion to different situations by relying both on a unsupervised algorithm that stores associated rewards to each situations (which are represented as abstract patterns in a DNN), and use that knowledge to explore the solution space more efficiently. This mechanism is also Turing complete.

I believe, all these methods are still only optimized for solving problems that involve pattern matching. But AGI is further than just pattern matching.

I'm still in college, doing my BE computer science degree. I have started implementing my algorithm, and I'm putting in all my effort (free time) in trying to finish it. I'm building this algorithm from scratch, in C++, and later I would use Cython to bring it into the Python world. I hope it works, if it doesn't, oh well... I'll cook up a newer hypothesis to try and implement that!.