AI could help us be better humans

People often say that we humans needn’t worry about growing automation, that what we should do is just focus on the things that humans do best: things like empathy and interpersonal communication. But does that really seem like a long-term defensible competitive advantage over the robot co-worker who would steal your job? I don’t think so.

What if it wasn’t about stealing your job, though? What if it was about using AI to make you a better person. Like, more honest, more co-operative, more considerate? The world could really use a lot more of that.

As Microsoft’s Dr. Ece Kamar said in a great recent interview, quoting her PhD advisor Barbara Grosz, “We already know how to replicate human intelligence: we have babies. So, let’s look for what can augment human intelligence, what can make human intelligence better.”

Two new studies written up at Kurzweil AI today demonstrate powerful and fascinating ways that humans and computers could work together for mutual benefit.

The first is a study of an algorithm that plays games with humans and seems to be able to get humans to be more co-operative by goofing around with them. The system uses “cheap talk (i.e., costless, non-binding signals)” between games and saw human opponents willing to co-operate more as a result. Is that a computer teaching us how to be better humans? I think it may be. And if you don’t co-operate? Then you get an electric shock! Just kidding, that was not part of the study. Coercion vs incentivization and behavior modeling seems a big important distinction though. I don’t want a bot threatening me into being more cooperative, but this study seems to demonstrate that if you want humans to play nicer, you don’t have to be coercive – you just have to play it cool.

Part of being human, of course, is learning and being wrong sometimes. As Microsoft’s Kamar points out, though, AI and machines are often wrong too- they have big blind spots just like us humans, only in their case it’s our blind spots that define theirs.

The second study cited by Kurzweil AI is of a beautiful system described as “A Crowd-powered Conversational Assistant Built to Automate Itself Over Time.” This one was built by a team that included crowdsourcing community leader Jeff Bigham (the second most influential Jeff in the crowdsourcing community) and it’s a swoon-worthy system of collaboration between multiple chatbots and Mechanical Turk. The Turk workers are asked for answers to users’ questions, then they vote on the best answers, and the best answers are added to the meta-bot’s library of answers for next time.


Bots making humans better humans, and humans making bots better bots. That’s a vision of the future I can get behind.