Abhijit Annaldas

For the love of data and machines that can learn!

GitHub | Kaggle | Twitter | LinkedIn | Feed

Pondering about Artificial General Intelligence


Researchers in the field of Machine Learning and Artificial Intelligence are relentlessly working in a quest to solve intelligence, build Artificial General Intelligence aka Strong Artificial Intelligence. And we see progress every once in a while as we witness jaw dropping feats accomplished by intelligent systems built at various industrial and academic research labs. Two such feats are, an AI system which mastered PAC MAN and the Alpha Go beating a human expert at the game of Go.

Some recent feats in AI systems are making great progress towards solving intelligence. If we would have thought about such breakthroughs a few years ago, probably we would have considered such AI systems to be an Artificial General Intelligence back then. Every breakthrough brings a lot of learnings and opens new possibilities. When researchers find answers, they also find new questions. The solution to such great problems is usually built on the mathematical and scientific principles/concepts that have been around for quite some time. Once such breakthroughs/feats are achieved, new knowledge/expertise is within reach and it isn’t seen as Artificial General Intelligence because we know we aren’t there yet. But, we don’t know how far we are from solving Intelligence. Hence solving intelligence appears to be a target that is moving farther.

I’ve been thinking of what would qualify as an Artificial General Intelligence and will it come from the fundamental sciences that are yet to be invented or discovered? There must be field of study that we are yet to discover which spans across mathematics, computer science, neurosciences, cognitive science, statistics, probability, linguistics, etc. It has been proven that humans, even in toddlers stage, tend to demonstrate qualities that are studied in depth across these subjects. There has been a lot of work put into understanding how a babies’ brain works, how they learn and grow. Children are the best manifestation of acquiring intelligence, which is exactly what Artificial General Intelligence is supposed to do. Children demonstrate the best way of learning. Period. Nothing else beats the evolutionary growth. This brings me to a thought that probably intelligence can be solved by understanding about babies brain development without constraining our self to one field of research. We should see understanding human intelligence and reproducing it as an interdisciplinary research, not as research area in AI, neurosciences, statistics, probability, etc. individually.

Here are few references/talks about how children think. Be ready to be awe inspired if you haven’t explored this area before…

Abhijit Annaldas
avannaldas [at] hotmail [dot] com