From compositionality to understanding: how deep learning is making AI a serious intelligence engine

I must confess that I am one of the optimists about Artificial Intelligence (AI) one day achieving general intelligence similar or even surpassing Human Intelligence. We should reckon though that we are still some way off to really achieve it by some decades or so. Some prominent thinkers, like the great physicist Sir Roger Penrose, are still completely skeptical as to a machine to be able to think, see and behave generally like a Human Being. But the question might not be if a machine will be able to think like a Human but more of when will we see the day when an entity that we regard as not really conscious is able to perform tasks that are on the border of real consciousness. There is nothing within current, and future laws of Physics, that completely forbids that such thing to happen. Actually I am of the opinion that what current machine intelligence is doing is beginning to resemble more and more a primitive form of conscious activity – the hard part of validating such a claim is not even possible for standard human or animal consciousness, with our current knowledge and science -; so to me Sir Roger Penrose’s skepticism is a bit contradictory. If and when we do find an appropriate theoretical framework for consciousness, will we be able to discard completely some form of consciousness beyond organic biological material or, for instance, some hybrid of  organic and inorganic form of consciousness…?

This long first paragraph serves me well to introduce the two videos I would like to share with viewers and followers of The Intelligence of Information. For in them we might accept why the current research developments within the increasingly exciting and closely watched field of Deep Learning, might lead us into the right path for an AI that will one day be a serious intelligence engine, capable of really understanding what itself is doing and do it with increased fluidity, rendered by the compositionality of enhanced cognition. Indeed in the first video below we recover a talk delivered by Standford’s Dr. Richard Socher (he is also the founder of the company MetaMind) at a Re.Work conference from 2015, where he presents how recursive deep neural networks  advanced to the point of being able to model compositional and grounded meaning between recognising words and reproducing their meanings. The name of talk is inspired by a paper with a different name that might be a worthy read after watching it: Recursive Deep Models for Semantic Compositionality Over a Sentiment Treebank :

Figure 5: A single layer of the Recursive Neural Tensor Network. Each dashed box represents one of d-many slices and can capture a type of influence a child can have on its parent.

The second video I would like to present here today follows tha arguments above. It is also from a Re.Work talk form 2015 and features  Bart Peintner PhD,  form Loop AI Labs. It is concerned with how modern deep learning frameworks for natural language processing are moving machine understanding of text to new heights. For example in the talk we can learn that machines are able to discriminate context to infer the right meaning from text, all accomplished by the increasingly sophisticated unsupervised learning pipelines being developed. The technology platform that Bart presents is a cognitive computing platform similar to IBM’s Watson. It harbours capabilities popular with businesses for market research or marketing activities, such recommendation tasks or collaborative filtering. One interesting take away from this talk is the basic reasoning capabilities of humans being actually emulated by an artificial entity (though hard this is and contentious as to if the machine is really reasoning…), another sign of AI being a real serious intelligence endeavour, if we were still in doubt…


Loop AI Labs – pressroom

Nevertheless we are still far from Artificial General Intelligence (AGI) being a reality. But what we should accept is that something will yield a path to it sooner or later. We just need to be reminded of advancements with Quantum Computing and Quantum Information, and their most probable integration with AI developments, to be with some intuition that AGI won’t be a complete impossibility, of the kind Sir Roger Penrose lingers on. The yielding path is rightly uncertain as of today though…


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s