Find out the latest indie author news. For FREE.

Elanora Brown
Contributor
Microsoft's Teen AI Smart-less, Jerk-more Tay Goes Down
Elanora Brown, contributor
Trying to build a replica of a Chinese bot Xiaolce, built in 2014, AI’s behavior happened as the machine had absolutely no idea what it is talking about as it learns from observing and adapting to its surrounding. This means that everything Tay said was generalized by opinions of others and it responded without giving the choice of words any second thought. At times, the AI gave relevant answers; however, things seemed out of hand at other times as well.

An example of this would be that if many users were to tell Tay that the Mexicans will be building a wall, she will have a negative or completely unexpected reaction if someone told her it is Donald Trump who is going to build it. This causes a major concern because she only knows what people said, and if the AI is asked something opposite of what it learnt, it reacts.

The system completely runs on lexical analysis and word association, meaning that those users who trolled Tay made a big impact, a contribution, to the reactions seen yesterday. The users took advantage of the AI’s reactions. One of the early examples of this experiment gone wrong was shared by @GeraldMellor:
In the apology letter Microsoft Corporate VP Research Peter Lee mentioned that the company did not test Tay under such circumstances. For us, we strongly feel that some topics should be black-listed, at least considered by the company so that the trolling users will not affect the AI and influence it to go rouge.

There is still a lot Microsoft has to get straight before it considers making the AI go live again. The reason behind this is because as soon as Tay is back, trolls will shower once again to test her limits. Having no brains, it is possible for it to once again break down and go from kumbaya to bully mode.

Microsoft needs to take its time with such projects, AI is not physical, however, we have not seen a human-like robot working on AI independently. If one day they will be made and would lurking around on the same streets as us, we do not want to be the ones to push them off the ledge, because it wouldn’t take them long to attack back if influenced to react in such a manner.

Loading...