Is the hype around AI finally cooling?
That’s what some recent surveys would suggest. Most executives now say the technology is more hype than reality— and 65% report zero value from their AI and machine learning investments.
However, these statements often reflect a fundamental misunderstanding. Many executives don’t differentiate generic black box AI from related technologies such as explainable machine learning. As a result, they’re missing out on a crucial pathway to smarter and more efficient decision-making that can drive more enterprise value.
Black boxes, or software programs that spit out mysterious answers without revealing how they got there, are the algorithms that power the world’s top tech companies. You have no way to know how a black box comes up with its result. Occasionally, the results are amusing, as when Google’s image recognition software erroneously identifies a cat as guacamole, or when Netflix recommends a bad show. In those cases, the stakes are low. A mistake on Netflix’s part costs, at most, a few wasted minutes.
But for complex, high-stakes sectors like healthcare, criminal justice, and manufacturing, it’s a different story. If AI technology informs a steel engineer to add the wrong quantity of alloys, producing a metal with the wrong density, buildings could collapse.
In areas like healthcare, where a single decision literally makes the difference between life and death, professionals may be particularly reluctant to trust the recommendations of a mysterious black box algorithm. Or, even worse, they might adopt them, leading to potentially catastrophic results.
Unlike black box software, any AI solution that can properly call itself “explainable” should reveal how various inputs affect the output. Take an autopilot software, for example — the algorithm controlling the steering needs to know how much the aircraft will tilt if a sensor detects northwest winds of 50 miles per hour, and the user must be able to understand how this information impacts the algorithm’s predictions. Without this ability, the software would fail to serve its intended purpose, and thus would result in negative value.
Furthermore, explainable software should provide some kind of measurement indicating its confidence in each prediction, allowing for safe and precise decision-making. In healthcare, for example, a doctor wouldn’t just be told to use a certain treatment. Rather, they’d be told the probability of the desired result, as well as the confidence level. In other words, is the software very confident in its prediction, or is the prediction more of a guess? Only with this kind of information can the doctor make informed and safe decisions.
How can you apply explainable machine learning to drive smarter decision-making in your company?
If you want to build a tool internally, know that it is difficult. Explainable, machine learning is complex and requires deep statistical knowledge to develop. One sector that’s done this well is pharmaceuticals, where companies often have scores of Ph.D.s doing in-house explainable data science and analysis.
If you want to buy software, you’ll need to do some due diligence. Look at real use cases that the vendor provides, not just taglines. Look at the background of the science/research team — are they proficient in explainable machine learning? What evidence are they showing off their technology?
Most importantly? Use your judgment. The great thing about explainable machine learning is that it can be, well, explained. If you don’t get it, it probably won’t drive value for your company.
Originally published in VentureBeat.