Man vs Machine: Unconscious Bias in Artificial Intelligence

I woke up and, with great curiosity, turned to my Jawbone UP phone app to see how well I had slept. It reminded me of a Berenstain Bears story where Mama Bear chides Papa Bear when he asks how on earth he will know the day’s weather without the TV weather report. “Try this,” says Mama. “It’s called putting your hand out the window to see if it’s raining.”

We already rely on our machines to tell us the weather, our well-being, and our activities for the day. I don’t know how well I slept unless my robot tells me. Apparently I slept well: “You had two hours of deep sleep. This is over the average for people in their 40s.” The comparison is gratifying. After two weeks of just the data on my light, deep, and REM sleep, I was no more the wiser about my sleep quality. Searching on Google for the right or average amount or perfect ratio yielded nothing definitive. Of course, UP is struggling to understand that I slept well after twenty-four hours of travel and a fifteen-hour time difference.

Still, I do feel refreshed. Score one for the robot.

Measuring your life isn’t new.  A Japanese engineer, Jiro Kato, invented the pedometer, called Manpokei, in 1965. Tim Ferris packaged his vision of self-experimentation in The 4-Hour Workweek in 2007. In 2010, Wired magazine editors coined the term, “the quantified self “, to describe how machines would monitor us, initially for fitness and increasingly for medical wellness, all aiming to satisfy our addiction to self-improvement.

What has changed is the mass-market acceptance of relying on machines to help us perform not in just automated functions, but to supply us with the bedrock of data upon which we build our basic decisions. The media has responded with a backbeat, exploiting our fears about robots stealing our jobs, our love lives (Her), taking over our families (Humans) and, well, the world in general. In one scenario, low level robots do our rote work so we can perform better. In the second scenario, they move so far up the value chain that they run the world while we clean the toilets.

The challenge is figuring out to which part of most processes we humans can add the most value.  At an airline counter in Los Angeles, I watched a manager and trainee change my flight to an earlier one routed through Las Vegas. The conversation was about which fields to change, highlight, modify, and print. It was software training. What was missing was judgment. Could I get through security and to the gate in time? Would I be able to make my flight change in Las Vegas and what were the alternatives if I did not?  None of these questions were raised, but they crossed my mind as I went over to the security line to see the crowds clamoring.  I sweet-talked my way to the front of the line and made my flight. But it made me wonder who I would bet on if I had to choose between man vs machine.

At an FT NED breakfast for experienced directors on board behavior, we discussed judgment and unconscious bias. “Perhaps robot board members would bring an end to our bias?” someone suggested. The senior leaders at the table recoiled. “Who’s programming the robots? What are their biases?” It’s a fair question to ask in light of snafus like the Microsoft’s launch of AI chatbot Tay earlier this year, who quickly went rogue (and racist).  As the article notes, it proves the old programming adage of, “flaming garbage pile in, flaming garbage pile out.” As technology historian Melvin Kranzberg famously wrote “technology is neither good nor bad; nor is it neutral.”

Indeed, what is a good sleep and who makes that decision? Are we happy when a robot board director says your company culture is above average for 150m market cap companies? Where is the room for judgment? Does it mean we will all revert to the least common denominator? Or can we use our judgment to rise above?