Title: 10 Breakthrough Technologies 2017: Reinforcement Learning
Why has reinforcement learning recently become so formidable? The key is combining it with deep learning, a technique that involves using a very large simulated neural network to recognize patterns in data
My Perspective on this: This is a good article not only for what it covers but a few things that I inferred from it:
- Why is it that the ultimates test for any AI technique, tool, or program is competing with humans? AlphaGo did not beat a human because of its superiority at playing Go – it beats humans because they are not used to playing machines (same with the poker players that lost to the machines earlier this year – how do you play poker without cues and body language?) The competition with humans only reinforces what we already know – computers are better at managing large series or combinations than humans — which is why I call that style of AI advanced analytics. Fast processing is why we build computers – but still lacking many of the components that would threaten us.
- Of course, this does not take away from the power of reinforcement learning as an AI technique, but you have to consider it as one more tool – which is what the article infers. The quote above tells you more about how combining techniques we see the value of the approach grow dramatically. We tried for a while to use cognitive approaches, but without deep learning (marketing term for ML) in the mix, we could’ve never done it. And this is done because of the availability of cheap, fast, easy-to-access computing power.
- If you add this technique combo to Generative Adversarial Networks (GAN) – another hot technique being used right now – you will start to see what are some of the ways (very advanced) in which we can train computers to do better. This is the key point, IMO, from this article — takes a village to train a computer to pretend to be human.
One more thing. This article is part of a series that goes back some 15 years. Take some time, or come back with time, and read some of the past years’ technologies and you will quickly see how we got to where we are — and maybe peek into where we are going.
Thoughts? Starting to go too deep? I can go much further 🙂 or just pull up a little… this is the last week I cover AI, moving on to platforms and ecosystems next.
follow-up reading: https://medium.com/@Synced/why-alphago-is-not-ai-eeeacbc5918c
(Cross-posted @ thinkJar)