!rant

This is more of a thought-related post. In the morning I stumbled across an article about artificial intelligence and the research from Facebook. I couldn't get around the thought of Elon Musk warning the people about uncontrolled developing of AI. The article was written about the experiment of Facebook, where two bots (Bob and Alice) were told to communicate with each other. As the developers "forgot" to implement a reward for using the English language, the bots started to change the grammar and spelling. They invented their own english-styled language, removing words that were too complex in their opinion. As soon as this happened, the researchers stopped the experiment, stating that they "couldn't follow what the bots were saying".

I wouldn't call myself a neural network expert, but I can understand why the bots could have behaved like that. But: Imagine that we invent an artificial intelligence with greater responsibility and just "forget" the reward for a specific task. If the AI will then try to increase it's own efficiency, I believe that we will be in alot of trouble.

Any thoughts on this are highly appreciated, as I think that this is a topic we should all look into (especially on a platform for developers).

Original article (german): http://gamestar.de/artikel/...