Skip to main content
No. 3102:
Computer Decision-Making

by Andy Boyd

Today, should we trust our computers? The University of Houston presents this series about the machines that make our civilization run, and the people whose ingenuity created them.

Imagine someone asks you to add a column of one-thousand numbers. Your first thought is likely ugh. After that you probably think about ways you can feed the problem to a computer. After all, if the numbers were in a computer spreadsheet you could get the summation with a few clicks of the mouse. But just as importantly, when you're done you'd absolutely trust the result. I for one know that adding them by hand I'd almost certainly make a mistake.

We trust computers for many things, and rightly so. Keeping our bank accounts straight. Keeping record of our plane reservations. Helping us do our math homework. Without computers for such mundane activities our lives would be vastly different. So here's a question. Would you trust a computer to diagnose whether a lesion on your arm is a highly malignant melanoma?

Deadly Melanoma. Photo Credit: Wikimedia

Basal Cell Carcinoma. Photo Credit: MedicineNet

A recent study by IBM reports that algorithms jointly developed by computer scientists and doctors were more effective in visually diagnosing melanoma than a panel of eight medical experts. The numbers were hard to ignore. Statistically they made a very compelling case in favor of computers. And it's not an isolated instance. Many types of medical diagnoses have the potential to be improved using computers, though medical applications are just the tip of the iceberg. Will we welcome computers that perform such feats the way we've welcomed their number crunching forebears?

One key difference between the old and the new is that we're now asking our machines to make decisions for us. When we add a column of numbers there's a right and a wrong, and barring a mechanical glitch the computer will always be right. Not so with decision-making. A computer might tell us that on fourth down with a yard to go, based on mountains of data analyzed with cold dispassion, we're better off running than attempting a field goal. But who knows how the play will turn out? And if running fails, will the coach keep his job after ignoring the time-honored tradition of kicking?

Bill Belichick. Photo Credit: Flickr

In many cases the stakes are much higher than the outcome of a football game. Suppose it could be statistically shown that self-driving cars reduced fatal traffic accidents by a factor of ten. What do we tell the one-in-ten people who died because a computer algorithm failed to take appropriate action? Is the car maker liable? And would you be willing to cede control of your car to a computer program in the first place? At least when you're driving you're in control.

There's no question that in many new areas computers algorithms will prove better decision-makers than humans. This isn't to diminish our human virtues. We have a host of capabilities that computers don't. The real question is how much freedom we'll relinquish to these algorithms; whether we'll embrace what the numbers tell us, or submit to the fact we're just not ready. It is, after all, our human choice.

A biker embracing freedom. Photo Credit: Wikimedia

I'm Andy Boyd at the University of Houston, where we're interested in the way inventive minds work.

(Theme music)

Belichick's 4th Down Decision vs. the Colts. See this engines episode

N. Codella, et. al. 'Deep Learning Ensembles for Melanoma Recogni-tion in Dermoscopy Images.' IBM Journal of Research and Development, 61:4/5, 2017. See also the Cornell University Library website: Accessed January 10, 2017.

G. Garber. Fourth Down Analysis Met with Skepticism. From the ESPN website: Accessed January 10, 2017. 

Identifying Skin Cancer with Computer Vision. From the IBM web-site: Accessed January 10, 2017.

M. McFarland. IBM Uses a Smartphone to Help Diagnose Skin Cancer. From the CNN website: Accessed January 10, 2017.

This episode was first aired on January 12, 2017