Actually… It’s more accurate to say they’ve accepted our invite and we’re not sure how fashionably late they will be to the AI hype party. But we’re excited – apparently they’re excellent dinner guests.
This is very much the idea we got from Re:Work’s Deep Learning Summit in London recently. It was a very technical dive into the latest developments in artificial intelligence and machine learning, with talks from industry leaders DeepMind, all the way to startups cognifying machines in sectors ranging from automotive, to entertainment, and 3D printing.
‘Cognition’ is the most appropriate term when speaking about a lot of AI right now. In the excellent view of the major technological innovations advancing our world, Kevin Kelly’s The Inevitable makes the clarifying point that AI is largely about cognifying machines so that they can take up simple activity, allowing people the freedom to perform higher order tasks. We need to live alongside AI, not in opposition to it. Our role then becomes to create new tasks which we eventually ‘hand down’ to cognified machines. This is similar to what happened in the industrial revolution as electricity, steam and coal gave way to machines that removed the need for much manual labour, giving people the gift of time to focus on other pursuits eg. arts.
The Summit’s opening speaker, Neil Lawrence, a Professor of Machine Learning and Computational Biology at the University of Sheffield set the scene with a correlation to the industrial revolution. He said that while Thomas Newcomen’s invention of the steam engine was significant, the tin mines in the south of England for which it was invented didn’t have much use for it. But coal mines up north did – they had the fuel to feed the machine. Even then, it was James Watt’s application of a condenser some 70 years later that made the machine work more efficiently. And thus the industrial revolution began.
We now have better computing power and more data to feed the beasts than ever before. However as Lawrence said, AI needs its ‘condenser moment’.
We’re starting off small. Mastering of simple tasks was very apparent during the Summit’s sessions. Whole teams are dedicated to teaching a machine the difference between cats and dogs, helping a robotic arm follow a tracking ball, or challenging a programme to master 80s classics on the Atari.
Once you dive into the complexity of the algorithms behind the movement you really appreciate what it is these geniuses (those in this field are nothing short of this descriptor) are trying to achieve. To master Pong requires a huge funnel of data, supported by a long line of code, which only then results in the task being completed. Even then, the machine has only learned that one task; and to build upon that requires a whole new set of information and instructions. Which then leads us into the exciting realm of progressive machine learning, which can be equated to learning to walk, then applying that knowledge to running; then training for a marathon.
Taking Deep Mind’s Atari example as the benchmark, we can surmise that AI is currently like a ten year old playing video games. But really, really well. And not only is there no telling what it can do, but we are learning about learning along the way. The applications of that in itself is hugely exciting.
One of the constant themes during this event was actually the problem facing the Deep Learning community: data. There just isn’t enough of it, or rather, not enough of it outside the dominant platforms of Google, Amazon and Facebook. In a world where more data is created in the last second than was created in the previous minute, this comes as a shock. These algorithms are hungry for, and their creators starved of, data.
There’s a lot of great work taking place. But the truth is, AI is not yet mature enough to have a drink with us. But as we all know, kids grow up quickly. There’s a lot of hugely exciting work being done, and putting aside self-driving cars and other machines for the moment, there are some exciting developments in healthcare where companies like Benevolent.ai are using AI-enabled research to identify effective drugs more quickly, or new ways of using existing treatments. Or beyond this, apps like Babylon Health even aiming to reduce the strain on the healthcare system by helping people accurately diagnose and treat some ailments themselves.