DeepMind experiments on AI shows aggressive behavior

https://www.weforum.org/agenda/2017/02/ai-learned-to-betray-others-heres-why-thats-okay?utm_content=buffer2d2c2&utm_medium=social&utm_source=facebook.com&utm_campaign=buffer


This article explores the AI's tactics at either cooperation or aggression. The video at the end of the article is very interesting, and I recommend everyone reading to watch it.

The study was done by DeepMind to see whether AIs would use guns to shoot their opponent to get the apples. Or would they work together?

What was found was both startling and some might say predictable.

Given the instructions to collect as many apples as you can on the board and having the ability to shoot your opponent and stun them were simply it.

Discovered was that given many apples on the board, the AIs didn't shoot or m much at all. Given few apples, the shooting spree began. more interesting, was that more advanced AIs would shoot more often even with plenty of apples on the board.

The definitely showed a preference to logic and strategy to compassion and care.

But again, some might say this is no surprise. Given the parameters, that is what they will do.
The AIs are responding to rules humans gave them, human rules. They've been given a gun, and an objective and no limits.

They don't shoot for fun, or out of fear or excitement or nervousness. They shoot because it will allow them to win. Pure strategy. We didn't teach them anything else.

So, as humans in a world that is coming closer and closer to merging with machines...we will one day be walking along them on the streets and if we want to stay safe, we are going to have to get better at teaching them what they need to know and not giving them the ability to shoot;)



From author M.Black
EXOITQA, THIRTY, and SPHERE are part of EXOTIQA WORLD series, a 3 part book series that explores the blurred lines between human and machine, and what happens when machines start becoming conscious.