Would You Give Google a Passing Grade on Its AI Project?

I’d like to imagine you are a teacher who has asked the most brilliant students in the world to build an AI that scours the internet — every known public document — to produce answers to simple questions.

You sit down on finals day and type in the question “Did the Holocaust happen?”

The machine is fast. It comes back and tells you — nope, it was faked. The Holocaust was in fact invented by a group of people concerned about German rubber production to justify certain military actions.

Would you give that project a passing grade? Or would you say “I think you’ve got some wires crossed here. Try again.”

Of course we all use this student project every day. It’s called Google.


The response that Google has to such critiques is the same as Facebook’s: we don’t mess with the algorithm. And that’s fine as far as it goes. But ultimately the algorithm either works on fundamental questions of fact or it doesn’t. And if it doesn’t, maybe Google should be spending less time funding smart thermostats and self-driving cars and launching wi-fi balloons, and more time funding programmers who can write algorithms that can use the massive amount of documentation on the Holocaust to determine that one of the definitive events of the last century did in fact “happen”.

Maybe they could even follow Tressie McMillan Cottom’s advice and hire the sort of diverse workforce that understands why these questions matter?

Or they could build a smarter home thermostat.

By the way, Bing’s engine is aware that the Holocaust happened, so it’s apparently possible. Here’s Bing’s result, which isn’t perfect, but at least gets a Gentleman’s C:

holocaust denial.JPG

It’s also worth noting that Wikipedia knows the Holocaust happened as well, and that particular wiki page, which has weathered many attacks, is as good an example of the sort of information environment we could have on the web if we thought a bit bigger than the current algorithmic tweaks favored by Google and Facebook. The fact that Wikipedia’s process routinely outperforms Google on questions of fact should tell us something about potential solutions to our current post-truth malaise. It’s a pity no one is listening.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s