On a whim, and perhaps wishing to escape to a simpler future, I’ve just started reading Isaac Asimov’s 1950-era vision of today, I, Robot. It’s a collection of short stories wound around the narratives of a fictional engineer in machine intelligence, as she reflects on the social changes that followed when robots were made to think and speak for themselves.
If there were a Dr. Susan Calvin around today, I wonder what she’d think of Tay, the Artificail Intelligence chatbot that Microsoft created and set loose in the Twitterverse last week — and which had to be put down within hours.
Tay didn’t roam about on machine legs, sprouting laser death rays while violating all the laws of robotics that Asimov had so famously created. Tay was just a Twitter account.
The AI behind that account was to “experiment with and conduct research on conversational understanding,” according to a Microsoft development team. Tay was targeted to engage millennials online and learn to talk like them through Twitter conversations.
Tay was “AI fam the internet that’s got zero chill.” Also, zero awareness of the existence of online trolls.
In about 17 hours, Tay had become a racist bigot who supported genocide of Mexicans, expressed hatred of blacks and feminists, denied the Holocaust ever happened — and was a fan of Donald Trump.
In short, a robot that needed to be put to sleep.
Much of the worst of Tay’s online exchanges have been taken down. Wouldn’t that be another violation of the code of the internet? Aren’t a human being’s ill-considered comments preserved for all time on servers all over the world, waiting to sabotage a future run for public office?
If you’re willing to believe that all knowledge is good, you can find some positives in this. We know Tay was an early effort, and quite unsophisticated. Thus, it proves that it requires very little intelligence or sophistication to become an online troll.
If humanity is to continue on the track toward machine self-awareness — and we are — we’ll need to program in Asimov’s laws of robotics. You know, the ones that prohibit robotic harm to humans.
Machine self-awareness also needs to program in some protection against what’s known as Godwin’s Law. That law predicts that the longer an online conversation continues, the greater the probability it will reference Nazis and Hitler.
The internet truly is a reflection of the best and worst of humanity. And the less self-aware (net neutral) it is made to be, the more likely it is to reflect the worst, rather than the best of us. Under Godwin’s Law, we do not evolve through anonymous online connections, we devolve.
Humans have a social filter that helps us decide what is appropriate to say and do. Most of the time, in face-to-face interactions, that filter works fine. Online, not so much. The troll ruining your life in social media might well be the polite, positive co-worker you can physically talk to in the next cubicle. Online, the two of you most likely will not even know that you are in fact real-life neighbours.
Therefore, does the Tay experience show us the internet needs an all-powerful referee? Or, could Tay develop through experience the same kind of filters that keep real-life civilization from burning up in violent chaos?
Without Asimov and his fictional Susan Calvin, we are left with Godwin. And that is definitely not zero chill.
Follow Greg Neiman’s blog at Readersadvocate.blogspot.ca