Skip to content Skip to footer

OpenAI: The Path to Safe AI


What is OpenAI?

The idea of a self-learning AI that gains sentience has been a popular topic in many forms of fiction, from the film The Terminator, to the video games Mass Effect and System Shock. These fictions tend to revolve around a helpful AI turned malevolent, bent on the destruction of mankind or a similar disaster.

Due to Elon Musk’s venturing, this grim future is a bit closer than some might think, depending on your view. OpenAI is a non-profit organization founded by Elon Musk and Sam Altman. In 2015, $1 billion USD was put toward the project in the hopes of creating a free and safe, self-teaching artificial intelligence. Research continues quite smoothly today.

What did it start with?

Part of this research was done in the past, with skeletal systems simulated and created to walk. This AI in question was able to take any skeletal structure, then learn how to run through self-simulation and trial-and-error. Such structures include from an orb on four, one-jointed legs to a dog-like figure with only two legs.

These situations are run under multiple conditions, primarily without any modifiers and under a hail of fire from orbs intending to knock it over. This process is known as Proximal Policy Optimization and it’s useful for developing actual robotics too.

What was the next step?

The second step of this research might seem a bit unusual, but it’s quite the clever method.

Dota 2 is a free-to-play multiplayer online battle arena with a very high skill ceiling and over a million players. The game is highly competitive, with many tournaments with large cash prizes held every year. It can take multiple years to learn and master.

This AI was set to play matches against itself for two weeks, then it was pitted against highly skilled professional players. The bot won, then did so against every player it faced. Only one player prevailed: Dendi, a former world-champion, and in a single match out of three.

This was all with only a month under its cyber-belt and continues to train still. The chances aren’t good, but it forces players to train and develop new, innovative techniques.

Game-capability aside, there is a lot of hope for what this free AI can do for society if properly utilized.

(Image Source: iCLIPART)