Latest self-taught AI defeats human counterpart

At times, you wonder “What if a robot could do all my work!” Well, something better is happening. Artificial Intelligence or AI has been around for a long time. It’s a super-smart computer that imitates human intelligence. Say, for example, thinking or logical reason under simulation.

However, in a breakthrough, AI has defeated the human counterpart. Moreover, a recent upgrade makes it functional WITHOUT human data!

Yes, you head that right. AlphaGo Zero, developed by Deepmind, is taking over the internet by a storm. Excited? Me too! Let’s get to the story now.

 artificial intelligence

Tracing  sensational feats of Artificial Intelligence

To begin with, there’s this popular game called Go. It is China’s most favorite game. You need to arrange black and white colored stones on a board to establish your territory. The incredible difficulty lies with the vast number of possible moves. I mean, it’s as high as 10^171. Yikes!

In 1997, IBM’S artificial intelligence Deep Blue defeated Garry Kasparov at Go. This marked the first sign that machines will take over the world someday.

Then, it was time for Lee Sedol to lose to a machine. He lost to AlphaGo, that too by a devastating 4-1 score. Since then,  AlphaGO defeated hundreds of anonymous human opponents. Without any human help.

Deepmind chief executive Demis Hassabis was indeed excited delivering this news. He said and I quote, “We’re quite excited because we think this is now good enough to make some real progress on some real problems.”


There is a considerable difference in approach made by humans and machines. While you try to focus, AI understands the game at a much higher level.

  • AlphaGo intricately goes through the games played by human counterparts. This is exactly what you mean by supervised learning.
  • Indeed, this self-tutoring helps AI understand Go from a humanistic point of view.
  • However, supervised learning relies on available data.
  • The only downside here is that these data are pretty costly and challenging to maintain.
  • So, here comes the big news. The latest update called AlphaGo Zero DOESN’T NEED HELP! Have a little patience, will you? We’ll get to this shortly. I promise.

Welcome the bigger and the better- The all-new AlphaGo Zero

Indeed, AlphaGo Zero takes a massive leap from AlphaGo. It has a much stronger and efficient algorithm. It discards the need for storing valuable human data. Instead, it tutors itself to perfection. Here’s a brief idea for you:

  • AlphaGo Zero begins right from scratch. Initially, it starts playing against itself. Hello, dedicated enough?
  • As opposed to supervised learning, this is called reinforcement learning.
  • After a “rigorous” training period of 40 days and over 30 million games, AlphaGo Zero was ready. It easily beat the previous version – AlphaGo Master.
  • AlphaGo Zero uses something very similar to the human brain structure- a deep neural network. Indeed, this helps it learn several abstract “ideas” and “strategies.”

Incredible? Yes. Perfect? Almost.

To be honest, this latest self-teaching AI has just one downside. This process of reinforcement learning still exhausts resources.

Is the future bright or bitter? Or both?

While a highly updated AI is a tremendous technical happiness, where does this leave us? I mean, have you seen the Terminator series?

Jokes aside, this self-tutoring ability is astonishing. Indeed, we are heading towards a future where AI will possibly deal with every task. Deepmind is coming up with more soon.

I’d love to see other advances. For example, AI dealing with complex tasks like “DNA fingerprinting” or “Protein sequencing” without help.

However, I hope AI doesn’t replace our need completely. Afterall, we made them and not the other way round.