All About AI – Part 1: What is AI, Anyway?

Some say that AI is everything a computer can do – others say it’s everything it can’t do (yet). Clearly, AI is an umbrella term that means many things to different people. To keep it simple, let’s just say that AI is anything a machine does that appears intelligent to you.

Sure, Siri, Alexa, Google Assistant and co just pretend to understand us. Still, the answers they provide can seem quite smart. Why? Because the developers behind these systems anticipate what we’re likely to ask and come up with great solutions. But step out of the areas a voice assistant has been trained in and you will be disappointed or bemused at best.  

Book Summary

More Than Just Weather and Music

Alexa skill developers are inventing new voice-based ways for people to interact with technology.

Bradley Metrock Score Publishing

Read Summary

For voice assistants (or chatbots) to get smart on their own, machines will need to be able to comprehend language. We may see a breakthrough in Natural Language Processing (NLP) within the next 10 years. Let’s hope that controlling our smart homes will keep us busy until then!  

Playing with AI  

Before we look ahead, it’s worth glancing back. It’s hard to believe, but developers had already created incredible AI algorithms in the 1960s and ’70s – just imagine computers diagnosing certain diseases at the level of human physicians! Unfortunately, after that initial boom, it took some time for exciting new AI technology to surface. 

Two developments kick-started the “age of the algorithm”: In 2010, Siri appeared on the iPhone and in 2011, IBM’s Watson computer won against the human Jeopardy game show champion. Now we see new, fascinating applications every day. Most of the systems do one type of job – and more precisely, faster and better than humans – but the future might bring general AI that is capable of living up (pun intended) to many different challenges. Self-learning machines will no doubt play a role in that.  

Take AlphaGo, Deepmind’s software that learned to play the Chinese game of Go (which is even more complicated than chess!): It learned by “watching” humans play the game. In this so-called supervised machine learning approach, human experts teach the machine. So, AlphaGo learned the strategies of the experts – and soon beat them. For humans, this was both a victory and a loss, depending on how you see it. And for the machine? Did it celebrate? Is it even possible for machines to have “feelings”? Well, let’s not get ahead of ourselves…  

Article Summary

The Superhero of Artificial Intelligence

Demis Hassabis, co-founder of the artificial intelligence empire DeepMind, hopes his technology can help solve the world’s greatest problems.

Clemency Burton-Hill The Guardian

Read Summary

AlphaGo Zero, AlphaGo’s successor, was a different breed entirely. Trained through “unsupervised” machine learning, it didn’t need to watch humans at all (who needs humans, anyway?). All AlphaGo Zero needed was to know the rules of the game, sit down (!) and play against itself. Good old learning by trial and error – AI style.

What would have taken half a lifetime for humans to learn, took AlphaGo Zero only a few days. And its successor, AlphaZero, could learn other games as well (chess within a few hours). Now, apparently, board games are too boring, so AI has turned to StarCraft II, which allows it to do stuff simultaneously instead of waiting for the opponent to hit that chess clock. Wait, didn’t it play against itself? Never mind.  

Next, let’s talk about machine learning. Stay tuned for Part 2 of “All About AI!”