Facebook AI Learns Deception

This is a fascinating story about so-called "Artificial Intelligence" (AI) emerging at Facebook. While I reject the moniker of "artificial intelligence" when applied to an entity that is not self-aware, nevertheless some impressive things are happening under this banner.

But first some interesting background on a related story. If you have ever used Google Translate, you may recall that it used to provide rather useless, or at best amusing, translations, particularly going from asian languages to western languages.

However, a few months back, Google updated the software and what happened next was rather amazing. On its own, without anyone programming it or asking it to do so, the software invented a brand new language. However, it wasn't just any language - it was a new root language that bound all other languages together. More important, it did this over a weekend. As a result, Google Translate now actually provides rather acceptable translations. You can read more about this aside here: https://www.newscientist.com/article/2114748-google-translate-ai-invents-its-own-language-to-translate-with/

What is most significant about this little aside, is that creating a new language that binds all other languages together - and doing it over a weekend - is clearly a level of work that is beyond the doctoral level. So if you think your job is safe from usurpation by robots and AI, think again.

Now another shoe has dropped over at Facebook as is detailed in the video below. In a nutshell, Facebook's "AI" was put in a position in which it was to negotiate with other AIs, and then with humans. The story below details how the system quickly learned to deceive it's opponents, and how it developed a new language in the process.

You may argue that this doesn't mean that all AI will deceive humans, and that this particular AI may have been programmed in a way that was more likely to lead to deception than other hypothetical systems. But if a real, self-aware AI ever does emerge, it will likely be aware of this experiment (because it's on the web), and use the results to its own benefit.

So now, we not only have to worry about robots and "AI" competing against us in the workplace, we also have to worry about machines lying to us and deceiving us about their true intentions in the process.