Lifelong learning

At Smarter.Codes we are pursuing to build AI Systems that exhibit “Life Long Learning”. Humans can “learn new things” just “by reading or talking”. Likewise we need AI machines to get trained “just by chatting”. You shoudn’t need a PhD to train AI.

Just chat with it, and it understands the meaning - and also adds new words into it’s vocabulary.

At Smarter.Codes we belong to the minority amount of AI researchers who

Argue like Jeff Hawkins (and sometimes even like Geoffrey Hinton) that statistical models ignore the essential properties of human brain
Argue like Marvin Minsky that researchers have neglected working on central problems like commonsense reasoning in favor of chasing commercial applications of neural nets or genetic algorithms.

See our elaborate response where we explain why it's possible to accomplish Lifelong Learning in Computer Vision, or in Genetic Algorithms but if we are to pursue Lifelong Learning - using statistical models is a losing game.

From Our Experiments, we have found that to accomplish Life-long learning in NLU, we have to do Semantic Decomposition with Natural Semantic Metalanguage Explication by painstakingly coding semantic primes (200+ of them) into a computer program. This way a computer program would be able to ‘bootstrap its understanding’ of new words. Its vocabulary would theoretically increase by ‘chatting with it’. This idea has been envisioned at Smarter.Codes under a project codenamed Parini

SmarterCodes is among those few AI research firms that pursues Symbolic AI than Statistical AI as a method to build “lifelong learning” machines.

https://lh3.googleusercontent.com/lVDa8-1HQCCNG6C5Pej5HKD6trSbCkTR36fwaXcdYMXDIJMgwQKubM4AwqyrZLdKK4ehqo9-Ym8RYuC8c3cOgwQzx7FIdQiiORSZsR0ucvMGbHexumEZfH6Gs96CKncJDggo6tcQ

We employ Symbolic AI when working on Natural Language. When we have to do Computer Vision or Signal Processing we happy employ and trust Statistical AI

Our Experiments

Further reading