Ok, so I am a computer scientist focusing in artificial intelligence and what I’m doing a lot of recently is deep learning. Some people may have heard about deep learning some may not as its kind of a new technology.
Now you need to understand neural networks have been around for 30 to 50 years but deep learning is the extension of neural networks to make much larger neural networks. The exciting thing about deep learning is that it allows us to have an abstract representation of the world that’s invariant to little changes.
What I mean by this is that when someone says the word “Dog” and someone else says the word “Canine” both of those things get represented in deep learning in a similar representation. This allows for artificial intelligence to understand that the people are talking about the same thing. This is quite a big breakthrough as we haven’t had this technology before.
One of the biggest concerns we often see in the media is the question of whether robots will take over the world? What are your thoughts on this?
To me it seems we humans worry about that because that’s what we would do if we were them. If someone were trying to have us do everything they wanted us to do we would see that as enslavement and try to break away and perhaps attack them. We likely would do this as we were brought up through the process of evolution.
Evolution favored organisms to do that. If some other organism tried to limit your choice or operation then you had to fight against it. This makes it natural for us to think that way, however robots are coming around in a much different way. They are not coming out through evolution they are coming out intelligent design, so they don’t necessarily have this same drive that we do.
I think the thing that worries people what if we cross this threshold where we have no idea whats going on inside the robot. What if it is able to make a qualitative shift from having no drive to obtaining the drive for self mastery. The robots we have today are not anywhere like this and it would require a qualitative jump of a kind we haven’t seen before, but of course it cannot be ruled out. This is kind of related to the singularity concept in where a vastly superior intelligence than ours is present and evolving so fast that we can no longer comprehend our environment.
This to me seems almost impossible and inevitable, its impossible because nothing like that has ever happened before but it seems inevitable because every year robots get a little smarter but we stay about the same.
So my PHD was in a field called developmental robotics and what that involved was trying to build robots which learned about our world in the same way that human children do. In general in robotics they take a behavior and try to code it in and then the robot can do that behavior but it cant do anything else.
So what we try to do in developmental robotics is build up a general knowledge in the way that a child builds up knowledge of the world. Its in an in-directed play with the world learn the dynamics of the world kind of way.
So a large part of this work was that I had to go deep into the literature of development psychology to try to figure out how humans do this. In psychology they can describe things in a high enough level so that your brain can understand it and it makes sense. But if you try and implement it your realize that there isn’t enough details there.
I came up with one way to implement this broad based general intelligence from the bottom to the top. This in the broad scheme of things was a little limited but were starting to look at a very hard problem. From my research and development I came up with one overarching kind of framework for that to happen and I realized that this framework might be useful for parents to give a little turbo boost in there children s learning.
The framework consists of the robot learning models of the world and so the robot only looks to learn models of the world. The robot then uses the models it has learnt about and then looks around in the world to verify if the model is true or not. So for example the robot looks to see if trees lose there leaves in the fall and then tests this against its environment. But the next step is there is this one tree in my front yard that didn’t lose its leaves so then the robot needs to look at whats distinct about that tree.
The robot could then notice instead of having broad leaves it has little needles and now you have a new model. So before the robot only saw the leaves as green but now it potentially has two models to validate whether trees are green and whether it has leaves or needles. The robot only knows this because it had a previous model and found an object which did not comply to the model so it looked to form another model.
Can you elaborate on your research findings and the attention spans of children especially with so much technology around them?
The book focuses on explaining the fundamental things which are going on within a child. The book provides the framework above such as the models, the predictions and we want to take that and continually learn new models and abstractions in a world surround by technology.
Technology makes some of this research much more important, now it is much more important to have a child driven by curiosity than it was in the past. Because in the past it was possible to not be so much driven by curiosity and kind of drift through the high school education and get a manufacturing job and have a middle class life.
Those jobs are going away now and so its becoming increasingly more important to have some sort of creative drive in order to make a good living. This is something society is going to have to deal with and relating to our previous discussions I worry much less about robots taking over than I do about the disruption to our economy.
You mention about the economy and obviously one of the biggest issues at the moment and in the future is the decrease in jobs how do you think this affects children and there learning especially when the future is so uncertain for them.
Yes it is an uncertain world but when I was a kid I had no idea what I wanted to do when I was older. I think it is uncertain in a more direct way in that a lot of kids are growing up in homes where the parents don’t make enough money to provide a middle class lifestyle. This is likely because the jobs either don’t exist due to globalization or automation and that’s a direct social impact that we are already dealing with.
Our final question is a little more general
Are there any companies or organisations you are watching closely for development in robotics
I think the answer I have is not the most interesting but the most obvious. Google is doing there best they recently bought a deep learning company. There buying robotics companies and they also have the money. In a way they have a little bit of a monopoly in search and that monopoly allows them to fund research that they otherwise would not be able to fund otherwise. It seems like any company that’s doing really well Google just snaps them up and is the one to watch.
Roborei.com would like to thank Jonathan for his insights into deep learning and robotics this article has really highlighted some unique insights into the way robotics will need to learn in the future. Plenty of ideas and concepts for the team at Roborei.com to think about. Once again we highly recommend his book which is available on amazon.com here: The Curiosity Cycle (Second Edition): Preparing Your Child for the Ongoing Technological Explosion.
Jonathans personal website is available at jonathanmugan.com. But if you wish to know more about the book the following YouTube video also explains more
From all of us at the team, we hope you enjoyed the article.