Join Infinia ML’s Ongoing Conversation about AI
“The best AI isn’t always the one that wins.”
If you deploy an AI-powered system and nobody wants to use it, does it make an impact?
Uh — no. Obviously no, that’s not even a riddle.
Whether you’re building or buying AI, if you don’t think about user experience, you might experience no users.
So let’s get some insight from Sean Gourley, the founder and CEO of Primer, an AI company focused on natural language processing — you know, machines that can “read” and “write” and maybe someday “use air quotes properly.”
From our conversation, here are three lessons for leaders.
1. Set the Right Expectations
Look at Google, for example.
“One of the interface decisions they made early on was they’re going to show 10 results as the default,” says Sean. “What that means is your precision only needs to be at 10% in order to get a good user experience.”
He points out that “if you do hit the button that returns the first result, they’ve framed it as ‘I’m feeling lucky.’”
Sean says that “an interface that lets you communicate that this might work, it might not” and “an interface that gives you 10 shots on goal, instead of one shot on goal,” the result is “a much, much better user experience.”
2. Don’t Expect Perfection
“Part of this here is, strangely enough, helping people understand that humans aren’t perfect,” says Sean.
“People tend to think, well, the humans are doing this perfectly and if the machine isn’t doing it perfectly, then the machine is not something I want to use.
“We’re imposing kind of a human set of values on these systems, as opposed to kind of appreciating what machines bring to the table.”
3. Make AI Understandable
“People have to be able to click on something to understand the why,” says Sean. “And that doesn’t necessarily mean they need to understand the inner workings of a complex neural net, but they do need to understand the features that are driving the predictions that are being made by these models.”
The Apple Newton, “a pretty advanced piece of technology back in the mid-nineties,” is a cautionary tale.
“It was very, very good at actually transforming handwriting into text, but the experience was actually really bad. You had no idea when it made a mistake, where it made the mistake. Right? Was it the T that I got wrong? Was it the curve, or the cursor? Did I have to put the cross on the T before or afterwards? So you were sitting there trying to guess what went wrong.
“And if you compare that against the T9 keyboard, if you made mistakes, it was slower, but you knew the mistake that you made, right? You knew the mistake was, I pushed the wrong key, and you could easily correct it. It’s just like, I need to be a little bit more precise in what I push.
“And so I think when we design these algorithmic systems, we need to bear the Newton lesson in mind because the best AI isn’t always the one that wins.”
Sean says that with natural language AI, you can design for a future that doesn’t exist . . . yet.
“There’s a lot of science fiction geeks in the company and we certainly embrace that as part of our culture,” he says.
“Back in the nineties, when you were designing video games, you could design the video game with graphics capabilities beyond where the chips were at.
“And I think when we’re designing solutions here in the natural language space, you can design products beyond where the science is today with the very, very confident expectation that science and the underlying technology will improve over the period of time between when you envisage the capability and when it actually comes into the market.
“So it’s an amazing time to be designing products around this. And I think there’s a long way to go just based on just the science we have today.”
“There’s very few fields that you can get into where there is this much science that’s just waiting to be deployed and waiting to be commercialized,” says Sean. “And that makes it exciting.”
“And yeah, it brings us back to science fiction and imagining a world that’s a little different from ours and being able to kind of create it and shape it. So it’s a fun place to play.”
Speaking of the future, you know what I’d love to see in your immediate future? You sharing, liking, commenting, subscribing — giving the algorithms what they want!