There’s no doubt that artificial intelligence will fundamentally change the world. Its potential is so vast that some even say it will be mankind’s last invention. Once it gets to a certain point, it won’t just be better at everything humans do. It may very well supplant humanity as the dominant life form on this planet. There are no shortage of movies that depict how dire that could be.
That said, it’s bound to go through some growing pains. Not all of those pains will involve a Skynet-level malfunction, but they will warrant concern.
At the moment, our artificial intelligence is still limited. It’s not dumb, but it’s very limited. It can do certain tasks very well, like play chess or win Jeopardy. It can’t apply that intelligence at a broader macro level like a human.
That still makes them useful and they’re still advancing rapidly. Artificial intelligence programs are used frequently for tasks like moderating comments sections and tracking sales data. The problem with these programs is that, since the AI is not generally intelligent, humans have to fill in the gaps. Since humans are flawed and bias, those traits sometimes find their way into the AI.
That’s what happened recently with YouTube’s comment moderation algorithms. As an aspiring YouTuber, I know how imperfect those algorithms can be. The intentions are noble. These AI programs are supposed to curtail hate speech. The internet needs that right now. Anyone who has ever visited 4chan knows that.
However, sometimes the AI systems are so narrow that they don’t ese the forest from the trees. That’s what happened recently when those systems mistook discussions about chess for racist language. Tech Xplore did an article on it and while it’s somewhat humorous on the surface, it’s also quite revealing.
“The Queen’s Gambit,” the recent TV mini-series about a chess master, may have stirred increased interest in chess, but a word to the wise: social media talk about game-piece colors could lead to misunderstandings, at least for hate-speech detection software.
That’s what a pair of Carnegie Mellon University researchers suspect happened to Antonio Radic, or “agadmator,” a Croatian chess player who hosts a popular YouTube channel. Last June, his account was blocked for “harmful and dangerous” content.
YouTube never provided an explanation and reinstated the channel within 24 hours, said Ashiqur R. KhudaBukhsh a project scientist in CMU’s Language Technologies Institute (LTI). It’s nevertheless possible that “black vs. white” talk during Radi?’s interview with Grandmaster Hikaru Nakamura triggered software that automatically detects racist language, he suggested.
Now, should we be concerned? Is it worrying that an AI with the backing of Google couldn’t surmise that simple terms like “black vs. white” were referring to chess and not race relations?
The short answer is not really.
The longer answer is not really, but we should learn important lessons from this.
The AI systems that moderate YouTube comments are nowhere near the kinds of systems we’d see in an artificial general intelligence. It’s like comparing a steam engine to a modern rocket. That said, we had to learn how to make a damn good steam engine before we could learn to make a decent rocket.
With something like advanced artificial intelligence, the margin for error is very small. You could even argue there is no margin for error. That’s why so many worry that such an AI could be an existential threat to humanity. If its too flawed to understand the difference between chess and racist rhetoric, then we could be in serious trouble.
The problem, in this case, isn’t with the nature of the AI. It’s with us, its creators. Since we humans are so flawed, racism being one of our worst flaws, it’s understandable that this sort of thing would find its way into our programming. It already has in a number of fields.
Again, those types of systems are limited and narrow. There’s a lot of room for human flaws to enter the system.
With advanced AI, those flaws could end up being extremely damaging. If too many of them find their way into a more advanced AI, we wouldn’t end up with a helpful, usable system. We’d end up with something like Skynet or Ultron. At that point, we’d be in serious trouble and we wouldn’t be able to rely on John Conner or the Avengers to save us.
We still have time. This latest issue with YouTube’s algorithms is minor, in the grand scheme of things, and fairly easy to correct. Once we get around to creating more advanced systems, though, we need to be aware of these flaws. We need to remember that any advanced AI we create will reflect our best and worst qualities. Let’s make sure our best win out in the long run.