Understanding the Limitations of AI

Panaceas are hard to come by. And artificial intelligence (AI) is certainly no exception.

As AI muscles it way into various areas of everyday life, a great deal of attention seems to be divided into one of two categories:

  1. All the wonderful things it can do.
  2. All of the horrific things it might do.

While those are bona fide concerns within reason, perhaps a healthier, more balanced point of discussion is AI’s limitations.

Rather than simply extolling AI or cowering from its perceived evil potential, looking at AI as a powerful tool with decided boundaries can help leaders and organizations better map out strategies with which AI can be used as constructively and ethically as possible—in effect, not asking it to do what it simply cannot do well.

Here are some suggestions:

  • It won’t catch your mistakes. Although AI’s analytic muscle can be substantial, the results it produces depend completely on the information fed into it. Even data that’s inadvertently skewed will generate similarly unreliable output. For instance, when Microsoft first began to use AI to evaluate job applicants, it inputted resumes from the prior 10 years—most of which were from males. As a result, AI interpreted this to mean that men were more attractive candidates than women, a bias that Microsoft didn’t intend to encourage.
  • “Narrow intelligence.” AI can beat chess grandmasters. It can outplay Jeopardy! Champions.  But it can’t necessarily address more complicated problems, as David Professor of Complexity at the Santa Fe Unit Melanie Mitchell writes: “If people see a machine do something amazing, albeit in a narrow area, they often assume the field is that much further along toward general AI.”[1]  For instance, while natural language processing systems are skilled at translation and text generation, they’re still incapable of conversing at length as humans can.
  • It won’t pick up on subtlety. Again, AI can only offer predictions and decisions based precisely on whatever data it’s received. That can cut a very fine line. For instance, when asked to perform a similar—not identical—task, AI will most likely defer to the process and outcome it worked with before. That’s because it cannot grasp subtlety or nuance and arrive at an outcome that reflects those subjective factors. As such, AI can very well struggle in retail settings, where its understanding of customer preferences and foibles can’t approach those of experienced human salespeople.
  • “Brain in a vat.” Although many believe that intelligence is specific to the brain alone, a growing body of research ties human intelligence with other forms of external stimuli. Phrased another way, lacking the effects of physical experience, a “brain in a vat” that relies exclusively on algorithms and the data it receives may produce empirical conclusions, but less complete—and possibly less reliable–than human thought.
  • No empathy. The jury is still very much out with regard to use of AI in medicine. While it’s been hailed as a breakthrough for disease identification and treatment recommendations, incidents such as IBM’s Watson offering unsafe treatment regimens for cancer patients uncover gaping flaws. Moreover, once medicine moves past empirical diagnosis and care, AI is simply incapable of the empathy and human feel that many medical positions require, such as nursing and physical therapy.

All-encompassing solutions are few and far between. Happily, evil incarnate is equally isolated. When considering the use of AI in your organization, it’s helpful to split the difference and consider the technology as decidedly promising but—at least for now—with its share of practical limitations.

That can prompt reasoned decision making with regard to AI, rather than an exaggerated perspective that prompt equally overblown conclusions.


[1] Dickson, Ben, “The four most common fallacies of AI,” VentureBeat, May 8, 2021.

Adapted from REINVENT: Navigating Business Transformation in a Hyperdigital Era by Faisal Hoque (Fast Company Press, 2023), in association with IMD. All rights reserved.

REINVENT debuted as the #1 The Wall Street Journal bestseller and is The 21st Annual American Business Awards®, 2023 Best Business Book of The Year, The The Stevie® Awards Silver Winner.

Copyright (c) 2023 by Faisal Hoque. All rights reserved.

This will close in 0 seconds

Copy link
Powered by Social Snap