The question is, what is the path forward?
It’s free, every week, in your inbox.
Hinton has compared hybrid AI tocombining electric motors and internal combustion engines.
Bengio has also shunned the idea of hybrid artificial intelligence on several occasions.
Whats missing in deep neural networks?
The limits of deep learninghave been comprehensively discussed.
Whilehuman-level AIis at least decades away, a nearer goal is robust artificial intelligence.
Those are key features missing from current deep learning systems.
However, theyre very bad at generalizing their skills.
A deep learning algorithm that plays championship-level StarCraft cant play a similar game.
It wont even be able to maintain its level of gameplay if the controls are changed the slightest bit.
When it comes to dealing with language, the limits of neural networks become even more evident.
Results are invariably too pointillistic and spotty to be reliable, Marcus writes.
In his paper, Marcus discusses this topic in detail.
Why do we need to combine symbolic AI and neural networks?
Connectionists believe that approaches based on pure neural web connection structures will eventually lead to robust or general AI.
We already have proof that symbolic systems work.
Its everywhere around us.
Our web browsers, operating systems, applications, games, etc.
are based on rule-based programs.
And that is why symbol-manipulation should be a vital component of any robust AI system.
Marcus iterates several examples where hybrid AI systems are silently solving vital problems.
One example is theNeuro-Symbolic Concept Learner, a hybrid AI system developed by researchers at MIT and IBM.
The researchers showed that NCSL was able to solve the VQA datasetCLEVRwith impressive accuracy.
But we cant tell unless we look, Marcus writes.
It will be interesting to see where Marcus quest for creating robust, hybrid AI systems will lead to.
