Whom Would You Trust: A Human Or An Algorithm?

I’ve been struggling with a dilemma.

Almost a year ago, I wrote a column asking if Big Data would replace strategy. That started a several-month journey for me, when I’ve been looking for a more informed answer to that query. It’s a massively important question that’s playing out in many arenas today, including medicine, education, government and, of course, finance.

In marketing, we’re well into the era of big data. Of course, it’s not just data we’re talking about. We’re talking about algorithms that use that data to make automated decisions and take action. Some time ago, MediaPost’s Steve Smith introduced us to a company called Persado, that takes an algorithmic approach to copy testing and optimization. As an ex-copywriter turned performance marketer I wasn’t sure how I felt about that. I understand the science of continuous testing but I have an emotional stake in the art of crafting an effective message. And therein lies the dilemma. Our comfort with algorithms seems to depend on the context in which we’re encountering them and the degree of automation involved.



Let me give you an example, from Ian Ayre’s book “Super Crunchers.” There’s a company called Epagogix that uses an algorithm to predict the box-office appeal of unproduced movie scripts. Producers can retain the service to help them decide which projects to fund. Epagogix will also help producers optimize their chosen scripts to improve box-office performance. The question here is, do we want an algorithm controlling the creative output of the movie industry? Would we be comfortable take humans out of the loop completely and see where the algorithm eventually takes us?

Now, you may counter that we could include feedback from audience responses. We could use social signals to continually improve the algorithm, a collaborative filtering approach that uses the power of Big Data to guide the film industry’s creative process. Humans are still in the loop in this approach, but only as an aggregated sounding board. We have removed the essentially human elements of creativity, emotion and intuition. Even with the most robust system imaginable, are you comfortable with us humans taking our hands off the wheel?

Here’s another example from Ayre’s book. There is substantial empirical evidence that shows algorithms are better at diagnosing medical conditions than clinical practitioners. In a 1989 study by Dawes, Faust and Meehl, a diagnosis algorithmic rule set was consistently more reliable than actual clinical doctors. They then tried a combination, where doctors were made aware of the outcomes of the algorithm but were the final judges. Again, doctors would have been better off going with the results of the algorithm. Their second-guessing increased their margin of error significantly.

But, even knowing this, would you be willing to rely completely on an automated algorithm the next time you need medical attention? What if there was no doctor involved at all, and you were diagnosed and treated by an algo-driven robot?

There is also mounting (albeit highly controversial) evidence showing that direct instruction produces better learning outcomes that traditional exploratory teaching methods. In direct instruction, scripted automatons could easily replace the teacher’s role. Test scores could provide self-optimizing feedback loops. Learning could be driven by algorithms and delivered at a distance. Classrooms, along with teachers, could disappear completely. Is this a school you’d sign your kid up for?

Let’s stoke the fires of this dilemma a little. In a frightening TED talk, Kevin Slavin talks about how algorithms rule the world and offers a few examples of how algorithms have gotten it wrong in the past. The pricing algorithms of Amazon priced an out-of-print book called “The Making of a Fly” at a whopping $23.6 million dollars. Surprisingly, there were no sales. And in financial markets, where we’ve largely abdicated control to algorithms, those same algorithms spun out of control in 2012 no fewer than 18,000 times. So far, these instances have been identified and corrected in milliseconds, but there’s always a Black Swan chance that one time, they’ll crash the economy just for the hell of it.

But should we humans feel too smug, let’s remember this sobering fact: 20% of all fatal diseases were misdiagnosed. In fact, misdiagnosis accounts for about one-third of all medical error. And we humans have no one but ourselves to blame but for that.

As I said – it’s a dilemma.

8 comments about "Whom Would You Trust: A Human Or An Algorithm? ".
Check to receive email when comments are posted.
  1. Scott Brinker from ion interactive, inc., October 31, 2013 at 11:15 a.m.

    I'm dressing up as an algorithm for Halloween.

  2. Sean Tracey from Sean Tracey Associates, October 31, 2013 at 12:02 p.m.

    Fascinating! Call me old-fashioned (or egotistical) but I'd still trust the seat-of-my-pants more than any algorithm. Remember Malcolm Gladwell's story of the Getty kouros, from his book, Blink?

  3. Gordon Hotchkiss from Out of My Gord Consulting, October 31, 2013 at 12:06 p.m. the kouros example shows, there's still a place for human intuition. But it's not universally the answer. There are many, many instances where algos beat humans. I think the reality is that currently (and let's remember, this is a moving target) we need both working according to their respective strengths.

  4. Scott Brinker from ion interactive, inc., October 31, 2013 at 12:31 p.m.

    Actually, for a more serious comment, take the example of the diagnosis of a fatal disease. The evidence would suggest that you are better off having an algorithm determine that. But is there any one on the planet who would suggest that you should turn to an algorithm to break that news to your family? My point is that algorithms work very well in certain contexts -- and by all means, we should leverage them to our advantage -- but there is still far more of what's important in life that is, at least for the foreseeable future, outside the reach of algorithms. As it happens, I think a big part of marketing is wrapped up in that unmechanized human part.

  5. Susan Breidenbach from Broadbrook Associates, October 31, 2013 at 5:30 p.m.

    Because of innate individual differences and also communication filters between doctor and patient, medicine will always be as much an art as a science. Algorithms may eventually be able to handle all of the science part, but humans are essential to the art part. Maybe a better metaphor is what has been happening in the commercial airline business for decades, as computers do more and more of the flying. Today's pilots have less and less experience doing actual flying, which makes them less able to intervene when catastrophe strikes and the computers become helpless. It's hard to imagine today's pilots pulling off something like that amazing 1980s crash landing in Sioux City, Iowa, when truly experienced pilots figured out how to use the plane's few remaining mechanisms in ways they were never intended to be used, and saved more than 100 lives. Regarding the education example, studies do show that self-guided online learning can be more effective than classroom teaching, but that doesn't take much when you consider the terrible job classroom teachers have been doing for decades now. That might be more a function of the educational establishment than of classroom teaching per se. The question is, can self-guided online teaching do a better job than GOOD classroom teaching? I think they will prove to have enormous synergies when used together effectively. Learn the subject matter online at your own pace, and then have teachers lead discussion groups about the material, for example. Anyway, sorry for the ramble, and thanks for the very provocative post!

  6. Steven Sevell from Sevell+Sevell, Inc., November 1, 2013 at 10:56 a.m.

    Great article. Really interesting perspective about how nothing: humans nor algorithms, are infallible. But in the marketing world (or scriptwriting for movie-world) wouldn't humans be necessary to develop the copy or scripts before any robotic algorithms could interpret it?

  7. Gordon Hotchkiss from Out of My Gord Consulting, November 1, 2013 at 11:18 a.m.

    Steven ...practically speaking, I would imagine so, but their creative license would be restricted by the findings of the algorithm. The algorithm would produce the formula and humans would just fill in the appropriate words.

  8. Steven Sevell from Sevell+Sevell, Inc., November 1, 2013 at 11:25 a.m.

    Good point. Hoping it doesn't affect the "small business" world in my lifetime...

Next story loading loading..