“Bias is one of the dimensions that you care about, especially when anything involves machine learning.” – John Bohannon

“(Always human) means always caring about outcomes for your customer, for the rest of the world.”

“Whatever it is that you're working on, whatever it is that actually has impact for your customers or within your organization. Just go measure it.”

John Bohannon, Director of Science at Primer, has a very clear answer when asked for one recommendation for organizations looking to start tackling the concerns of bias in Artificial Intelligence (AI).

“If you're going to just make one move, this one's the one. Just go measure!” he says.

In a candid discussion with Jeff Kavanaugh, Head of the Infosys Knowledge Institute, Bohannon shared his thoughts around bias, misinformation and the importance of being “always human”.

Founded in 2014, San Francisco-based AI and Natural Language Processing (NLP) company, Primer, works on building systems that can help organizations read, write, and analyze large swaths of data, and includes companies such as Walmart among its customers.

Bohannon, an American science journalist, and scientist qualifies Primer as a company that builds tools, especially for people who have the word analyst in their title.

“We try and take your knowledge and encode it into machine learning engines... and then you can hook up those engines to do useful work for you. And that's the name of the game here,” says Bohannon.

As an organization, says Bohannon, “always human” is at the center and guides everything that Primer does.

“What it means to us is - all the other values, all the other things that guide your decisions, when you work at Primer, it really falls under this top-level value of being always human,” says Bohannon, adding, “And that means always caring about outcomes for your customer, for the rest of the world.”

He also suggests that this is a tenet of good engineering, and if one puts on blinders and focuses on a narrow task, without caring about the consequences, or how reliable it will be, in contexts other than the ones that are specified for them, then they’re not doing good engineering.

“You got to care, you got to expand that circle of caring way beyond your little project, way beyond your team, way beyond your company, eventually to the whole world,” emphasizes Bohannon.

According to Bohannon, one of the most important dimensions of responsible technology is bias.

“Bias is one of the dimensions that you care about, especially when anything involves machine learning,” he says.

Watch our story on Bias in AI, and how it can impact businesses.

Bohannon notes that all models have bias, and most bias is harmless.

“In fact, a lot of bias is intended, you want the system to swerve towards some goal,” says Bohannon, adding, “The bias that we worry about, and we talk a lot about at Primer, in terms of how to detect it, how to mitigate it, how to prevent it, is bias that harms people.”

As an organization driven by AI and NLP the human is always the center, and Bohannon believes that this clarity of understanding is extremely important.

“It's important for people to keep in mind that usually much greater harm can come from just misunderstanding a thing that's working as intended,” he says, calling it a failure of imagination on the part of the creator.

“The user, whoever is experiencing this thing or interacting with this thing, just did not understand your intentions did not understand what this thing is meant to do, what it what its strengths and weaknesses are, even if it's unbiased,” he notes.

Bohannon emphasizes the importance of measuring constantly.

“Whatever it is that you're working on, whatever it is that actually has impact for your customers or within your organization. Just go measure it,” he says while expressing surprise at how few people do it even with resources at their disposal.

“That's the first step before you can even mitigate something you got to know if you have a problem to solve,” he concludes.