Deep learning in action with robot learning to walk
In an article on recent advances in artificial intelligence, we shared some thoughts on how AI is going to revolutionize all kinds of business and everyday life, including ecommerce. However, it may be hard to imagine what AI actually is and what it can do. Especially if you don’t have some background in computer technology. So here is a perfect example to share. And it may be more fun to read about it than about some complicated data analytic magic. A robot learning to walk the way humans do.
Nvidia, one of the main developers of innovative hardware and software solutions for the area of AI called Deep learning, posted in their blog about a robot called Darwin, capable of learning to walk the way humans do. This is huge, but has nothing to do with anything supernatural. People get sometimes confused by the term Artificial intelligence. Everything that AI does was developed from the technical and scientific knowledge that humans developed and documented, while on the other hand, natural intelligence with consciousness, human psyche, brain functioning, the life itself, if you will, is full of black wholes in scientific understanding.
Let’s get back to the robot Darwin, and see whats so special about it. After all, robots have been used in industry for long time. What Darwin makes different from classical industrial robots is it’s ability to learn, implemented with the help of Machine learning, an area of AI, or rather Deep learning, which is the name for more advanced type of Machine learning used in project Darwin.Ordinary industrial robots, perform tasks exactly according to their programs. Suppose such kind of a robot is programmed to walk from the point A to point B. Let’s name this hypothetical robot Hutton. Hutton is programmed to avoid obstacles, on its way from A to B by always turning right when there is an obstacle on it’s path. Turn right, pass the obstacle, turn back left, return to normal path. Something like that. But what if turning right would get Hutton stuck to the wall, while turning left instead of right would be the right decision? Hutton would never have figured it out to try to turn left instead of right. The only way to do it would be to have it’s program been improved by the programmer.
When Darwin would fail to avoid the obstacle by turning right, it would try other solutions by itself. Turning right doesn’t work, so let’s try something new, let’s turn left! It works, so let’s save this to memory. The next time Darwin would recognize the same type of obstacle that turning right didn’t work with, it would turn left in it’s first attempt, without first try to turn right. That is machine learning.
Machine learning in eCommerce
Same approach as with robot learning to walk can be applied to software systems for ecommerce. No robots legs and wheels needed their, except maybe for some warehouse and delivery robots, but still same computer technique of machine learning.
Suppose that Christmas shopping season is approaching and that the weather is extremely mild. How will this fact impact people’s decisions to buy winter scarves for Christmas presents? You don’t need machine learning to predict that people will buy winter scarves less than usual. It is just the matter of common sense.
Computers don’t know anything about common sense in the human way. But they are very good at processing large amounts of data. In the real life of ecommerce you have to predict sells for thousands of different products. You operate with many other pieces of information besides just if the weather is cold or warm. What if we have 10.000 records of data, where the fact that the winter is mild is just one record of data and we have 10.000 products to predicts Christmas sells for? That’s the case for machine learning. If we would have access to historical data for few years backwards, we could feed the machine learning program with that historical data. Among other facts, program could learn that mild winter does not predict strong sells for winter scarves.
The sky is the the limit when it comes to possible use cases of using artificial intelligence in eCommerce.