In March 2016, an AI-based computer program defeated an 18-time Go world champion – Lee Sedol. In the five-game Go match held in Seoul, the AI-based computer program defeated Sedol not once, not twice, not thrice, but four times. It did not just win the match; it conquered it. This brilliant AI-based computer program that plays the Go game is Google’s DeepMind – AlphaGo. AlphaGo donated the winning prize of $1 million to charities. Korea Baduk Association honored AlphaGo with the highest Go grandmaster rank – an honorary 9 dan.
To put simply, DeepMind is a machine learning system that builds its algorithms and makes complex decisions based on massive datasets. It has currently come to a stage where it can understand and perceive the spatial environment. In other words, it can see like humans do and take a call concerning the next course of action. Of course, such an intricate and elaborative algorithm design cannot be accomplished overnight. It takes thousands of labeled pictures for DeepMind to absorb information and interpret a scene.
What does this mean for the human world?
It simply means that humans are no longer responsible for the outcome of DeepMind‘s decision-making process, as it is now up to the machine what it makes of the data it is fed. DeepMind‘s ability to make a decision on-the-spot from hundreds and thousands of inputs from past data is risky. Imagine allowing DeepMind to pick you up from a point A and drop you off at a point B through hostile traffic. If DeepMind is not given such opportunities, it will never evolve and understand how to go about driving around the city without any hiccups.
While the core future of DeepMind is played close to the chest, Google has released TensorFlow, an end-to-end open source machine learning program that allows businesses and communities to train and develop their machine learning models to solve challenging, real-world problems. DeepMind is not designed to only play games and win a million dollars. Google is building DeepMind to detect fraud, speech, handwriting, streets, language, species, and provide help in translation and search.
Unlike other AI programs, DeepMind learns from the principles of an action, rather than from the principles of an action by a human. This makes it powerful, self-sufficient and resistant to human errors, as its knowledge is first-hand and not handed down by a human.