[Reshare] Artificial Intelligence Has 3 Big Technology Drivers - by Chris Thomas, McKinsey & Co

[Transcript] Technically three things that matter around the technology drivers of artificial intelligence.

One is the core computing technology and that's basically how quickly you can just process and learn.

You guys heard of Moore's Law in semiconductors? So actually right now in the natural image processing, it's actually amazing. We're basically doing a double speed on Moore's Law.

So if you actually look at Nvidia roadmap, if you look at Intel's Nervana road map, basically flops per second are more than doubling per year which is really impressive.

The problem with all these things and the big problem with all these things is that these things are way too power draining. So if you open up the back of a fully autonomous vehicle, there's like - you can't put anything in the back of the trunk. Because it's all just rows and rows of power-hungry GPUs. Which means you get no battery life. So that's actually not a performance issue and it's a power issue. This has got to get solved.

The second thing is around the algorithms - the programming language.

Now this is something that's really fascinating because this is the first time that sort of a major technology transition is happening almost fully on open source.

So open source is exactly that - you put the source code on the Internet and anybody can use it which is as opposed to Microsoft Windows - which you can see this part and you can make your APIs and you can write on Microsoft Windows but you can't open the Kernel.

Now what's interesting about this is that people don't think this will persist because there's no stack.

So if you're making applications for a smartphone, you know you can either write to the Android stack or you could write to the iPhone stack but you know where to write it. If you're writing for the PC - write to Windows.

And this of course makes the people control Android or iOS or Windows really really rich. But it creates enormous benefits because you get all of the innovation. You don't have to have innovation at the raw programming level, you can have it at the application and software level - it makes life a lot easier.

And people are expecting this to happen. And at the war that's going on out there and it's just fascinating and I have no idea how it'll end.

But you basically have about seven or eight different semiconductor companies plus about 12 major open-source platforms on AI. And then you also have all these Internet companies of their own closed source and everyone is thinking how do I create Wintel but in the artificial intelligence world.

This means that when you are a company down the line in ten years and you want to build robots on your factory, you want those to be smart robots replace human beings, you actually just like the same way today - you take an iPhone out of a box and it just works. Somebody is going to provide that stack that just works, and then you're just going to customize that for your specific factory.

And that is necessary because if every single company that wants to use thinking robots has to figure out all the programming to make thinking robots work, this business will never scale. Because there just aren't enough programmers in the world has every company (decided to) recreate (the) stack. The problem is that everybody wants to create the stack and they don't want their stack to work on somebody else's stack. So it's going to be an interesting battle.

The other thing is the amount of data that's getting collected. Because you cannot learn anything or do anything in artificial intelligence without data.

And the two big things that are happening of course is that due to chat and due to social media, there's an enormous amount of digital data that's being created.

But the second thing is you guys have heard of Internet of Things - IOT, facial video camera... There's now an enormous amount of real-world physical data that can now get integrated to the digital data and that's all getting the ability to get put together.

So that you know lot of data is basically you know... Then I have the numbers in here somewhere... but it's basically going up 5x every year.

I think that more data has been created in the last two years than in the previous - you know ten millennia of human history.

Now most of the data is pretty useless. But it is a lot of data. As someone whose wife takes about eight hundred photos a day I know that most of that data is pretty useless. So I have to get some jokes in here as I'm getting tired up here.

So and then there's a lot of venture capital. So a venture capital into AI has - it depends on what you call AI. There's a lot of companies which are just sort of random normal companies with a Website which are now calling themselves like Fintech or artificial (intelligence) so they can get a higher valuation, right? So it's difficult to say…

So China is approaching artificial intelligence from sort of a different perspective. Like most technical evolutions that push on productivity the upside in China of applying these sorts of technologies probably higher than any other market. [/Transcript]

Additional Reference: The Rise of the Machines:How Chinese Executives Think about Developments in Artificial Intelligence.