It may sound downright bizarre and something beyond the scope of not only today’s technology but even of the foreseeable one but would you believe that we are actually pursuing this? 

We don’t fully understand dreams or what role they play in our lives or how exactly dreams work. All these unknowns are being worked upon by scientists all the time but many mysteries still remain. 

Therefore, it is quite natural to think that if we do not understand dreams fully ourselves, how can we even think about giving them to robots and how does one even do that, but it’s exactly what Google’s DeepMind is doing…. sort of.

Robots That Dream

Ok, so what does this mean exactly? To understand this, you need to understand a little bit about Artificial Intelligence (AI). In the simplest of terms, basically, it means giving computers the ability to think. As the words suggest, it means creating intelligence artificially. 

In the last decade or so, we have seen tremendous advancements in this field which has led to many different innovations which were just not possible before. Some examples are, driverless cars, virtual assistants on smartphones, recommendation services in apps like Spotify, Pandora, Netflix, etc. 

Granted, these are not full fledged AI, but we are getting closer and closer. And we will get there only when AI can learn and relearn on its own without any sort of human interference whatsoever and that’s were dreams come in. 

How It Works

Scientists have long thought that dreams are not meaningless (even if they seem like it) but they actually serve a purpose. Our mind is active even during sleep trying to aid our learning process during our sleep. If we are going through some problems in life, the brain actively tries to find solutions even in sleep and it is done usually in the form of bad dreams. 

The theory here is that bad dreams are brain’s attempt at preparing us for tough situations. DeepMind is using the same principles for its AI to help it learn quicker and on its own. 

Right now AI has to be fed data by a human. It can then process the data and ‘make sense of it’ like a human does but it still requires human involvement. But researchers at DeepMind want to go a few steps further by accelerating the AI’s learning rate, by giving it ‘bad dreams’ equivalent for an AI. 

We all know how AI is besting us at different games like Chess, Go and recently, even Poker. Now, it is starting to try even more sophisticated games which requires a lot of decision making. So, in order to give the AI ‘bad dreams’, it is fed the most difficult and challenging levels of the games again and again, until it is able to find a solution on its own. 

Using this technique, researchers were able to increase the rate of learning of the AI by 10 times! That’s very impressive and shows that the method actually works. This will help AI learn in an unsupervised manner similar to how we humans learn things paving the way for a human-like AI.

So, What About The Dreaming Robots?

The terms, robots and dreams are used for making the concepts behind the technique clearer to the lay man. Of course, we do not have robots that can walk or talk like a human and of course we can’t give them random dreams like we get (yet).

Although that is not happening technically, in essence though, it is. Some day there will be robots who will walk and talk and be among humans like ‘normal people’ and they will learn just like us by dreaming but they will do their own version of it.