Inside bias: Why so many companies make big hiring mistakes

Venture Capitalist and Global Startup Advisor
AI doesn’t have to be scary. It can augment our intelligence, says venture capitalist and best-selling author Scott Hartley. What AI can also do is help us fight our own biases, starting with the prejudices inherent in the hiring process "algorithms”.
  • Transcript


Scott Hartley: 
In this world where we focus so much on what we’re building, how we’re building it, I think we need to take a step back and reconsider why we’re building, and really humanize our technology, really bring together diverse teams of methodologies and people and mindsets so that we can take our technology and actually apply it to the most fundamental human problems.

Today the conversation is largely about artificial intelligence, and one of the concepts that I like to discuss in the book The Fuzzie and the Techie is this concept of intelligence augmentation—so: thinking about using AI but using it in a way that’s augmenting the ability of humans.

So Paul English, who was the creator of, he is a techie through and through, but he also calls himself an AI realist; he’s somebody who believes in the promise of artificial intelligence, but also realizes that this is not something that tomorrow or next year or maybe perhaps in the next decade is going to completely take away from the characteristics and the qualities of what a human can provide.

And so he’s now creating a company called Lola that’s based in Boston, and Lola is sort of Kayak 2.0, where rather than trying to take the travel industry and put it online he’s actually taking travel and putting it back into the hands of travel agents, real people that are working on the phones dealing with people that are calling in to book travel.

And what he’s doing is he’s supplementing those travel agents with technology, with artificial intelligence, really “flipping the letters” and trying to use intelligence augmentation as an AI realist to sort of better the service that a travel agent can provide.

Eric Colson, who is the Chief Algorithms Officer at Stitch Fix, he uses machine learning—he uses artificial intelligence, but to augment the human stylist.

So they have 60 or 70 data scientists working on creating machine learning algorithms, but those are used to supplement the 3400, 3500 stylists who have their own propensities for delivering fashion, they have their own biases as to the geography or the age or the style preferences of somebody they might be serving clothing to.

And so the machine learning actually learns the bias of the human over time and tries to mitigate that bias by offsetting the selection of clothing that they provide at that particular stylist.

And I think that’s a really interesting example of artificial intelligence not necessarily taking away from that stylist but actually augmenting, improving, helping them perform better.

And I think that flipping the letters from AI to IA is really something that we should be thinking more about today in the context of the AI debate.

I think it starts with job requisition and writing sort of the job descriptions that we want to hire for.

And I think we are bombarded by applicants, we’re bombarded by new resumes and “data driven processes”, and so the quick answer is to use natural language processing and screen for keywords, to run things through a filter and draw out the resumes that really hit the five key words that relate to your team.

And I think what this does is it creates sort of an “inside bias,” where you’re creating and you’re bringing together people that all have sort of the same perspective, the same backgrounds, and it can really sort of create in the sense of what Daniel Kahneman, the 2006 Nobel Prize winner and behavioral economics talks about as “inside bias”.

And I think to the extent that we can think about inside/outside bias and trying to bring say 20 percent of the team from a different perspective, from a different vector, from a different methodology or background, that can really bring diversity to a team where, if you have a data science team, 80 percent of the people may make perfect sense to have them have complete backgrounds in data science—but what’s to say that 20 percent of the team shouldn’t be philosophers or psychologists or anthropologists?

And I think that sort of mentality of almost “Google 20 percent time,” thinking about it for 20 percent people time, 20 percent difference of methodology or difference of perspective.