Training AI is something journalists touch upon or brush past but never really dig into unless somehow it has failed.
We touched on the failings of the mismanagement of parts of the pandemic. That failing came about because the data the experts were using in training AI were not fit for purpose. Different datasets arrived in different formats from around the world, and there was little time to standardise them to be effective.
Now start-ups are being formed and funded to automate the data labelling process and, therefore, the AI training process.
One such is Snorkel AI, funded by Blackrock, Lightspeed and others, which has just reached Unicorn status and boasts Apple, Google, Intel, IBM, YouTube, the U.S. Department of Justice, and Stanford Medicine among its customers.
The idea is that training AI becomes much faster and more efficient, and therefore companies can deploy AI applications over 10 times faster, sometimes much more. And the AI enterprise market is one of the fastest-growing markets in the world.
Yet faster is not necessarily better. And using digital processes to make things so is far from foolproof.
Training AI is no different. It is at the core of the debate about AI itself.
Now, too, AI is training itself to become better at the tasks it is dreaming up for itself.
Deepmind is at the heart of AI, and with its latest ‘game,’ it is training AI bots to multi-task across 700,000 games in 4,000 different worlds. The bots seem to randomly throw items around until they identify something useful, such as throwing a tile across the game, and it becomes a ramp to another level.
It is, according to the MIT Technology Review, a small step towards Artificial General Intelligence.
It is bad enough that some people are already training AI to do bad. If they can harness the power of self-improving AI, which can be deployed at a huge speed by training AI with platforms such as Snorkel, the alarm bells will be ringing loudly.
Technology, and AI, in particular, is moving and ‘improving’ at ever faster speeds, and that should make us all a little nervous. Perhaps a lot nervous.
Training AI is critical to the success of this game-changing technology. Let us hope we manage to keep a balance and avoid losing control of the vehicle.
It may not be too late; AI can be for Good as well.