Let's now consider some examples of really existing AI which I'm going to call machine learning. So machine learning is the core technique of how we are today deploying AI systems and I want to start here with a side definition and then some examples. So to define this what I'm going to say is it's the use of statistics to tune or specialise algorithms-, to tune, train, another word here, or specialise algorithms to problems. But let's come through-, let's go through some of the, sort of, domains of application. So applications and examples. 

Now, I think we should start at the most sophisticated form of artificial intelligence in existence today. And what do I mean by that? I mean, consider having an almost infinite amount of money, an almost infinite amount of data, an almost infinite amount of expertise. With that, what is the best possible system you could build? And here is my example. Alexa. But what is this system? What is this system? This system is one of processing natural language. So it's a natural language processing system which tries to take audio signals obtained with a microphone and looking at only the pattern of the variation of volume of those signals over time. Correspond that to actions, really. So, you know, you're taking this volume and time and you have all of these, sort of, interesting little variations in volume and you say, 'Well, perhaps this means 'turn the light off'.' And really, that is really about it. 

Now let's say-, let's just say what that isn't. That isn't understanding that knowledge. In other words, if I say to you, 'find me the light switch in the room'. You're able to navigate and try out various things and perhaps the light switch has an odd shape and it's in an odd place but you can keep going. You know what you're looking for because you have all this prior knowledge you've gained from being alive for a very long time and you know what kind of thing would meet that knowledge. You will turn the lights on and off and you can keep going until you find it. Now this machine has none of that background knowledge, has only this audio signal and the audio signals of other people speaking of course, and then there is some system which corresponds pieces of these signals to actions the machine can take. 

So it's a very limited system but it is really and genuinely the best we have in the world. Some other, kind of, examples include those around prediction and detection. If you can think of something to predict from something you know, you may be able to use a machine learning system to solve that problem. And we can use lots of informationals, demographic information and other kinds to engage marketing campaign, to target ads to you based on what we have seen has been successful previously. So there are applications across finance, politics for electoral campaigns, medicine, retail, marketing and so on.