Key insights that will help you make the most of AI.

Cyborg head using artificial intelligence to create digital interface on city bokeh background 3D rendering
Image: Adobe Stock

It’s easy to get sucked into the hype around artificial intelligence (AI), but just as easy to get duped into thinking it’s all hype. The truth is somewhere in the middle. Or, as tech luminary Mike Olson suggested, “The breathless attention paid to AGI and self-driving cars and whatnot blinds [us] to the value of narrowly-focused AI applications.” By “narrowly focused” he was referring to the DeepMind announcement that it had released the “predicted structures for nearly all catalogued proteins known to science”.

Narrow? Hardly. This advance dramatically opens access to protein structures, thereby accelerating scientific discovery in fields as diverse as medicine and climate change. But the AI used is narrow in the sense that it isn’t some sentient machine, thinking through protein structures. As I’ve written, often the best machine learning (ML) is “just” pattern matching at a scale no human could hope to replicate.

Consider this a reminder that just because AI/ML hasn’t gifted us self-driving cars, that doesn’t mean it hasn’t yielded impressive advances. The trick is to narrow the scope of how we use AI, not to give up on its promise.

OK computer

The right approach to AI is to use machines for what they’re good at, and complement this with human intelligence. Machines can process massive quantities of information far beyond what any person could do, then present that information to people in ways that makes it more approachable for people to understand and postulate from. It’s not humans vs. machines—it’s humans partnering with machines.

SEE: Artificial Intelligence Ethics Policy (TechRepublic Premium)

And also data. Lots of it. In fact, as good as machines are, and as smart as people can be, the mapping of all known proteins simply couldn’t have been possible without data, as Ewan Birney, deputy director general of EMBL, stipulated. “All the AI talent in the world…can’t eas[ily] solve science problems…without data—and lots of it.” So where did the DeepMind scientists get the data? Fortunately, in this particular area, there’s a tradition of sharing data, as Birney went on: “Here the long established community norm in molecular biology for sharing data—in particular in structural biology here—is a key enabler.”

Applied to the data science projects within any given organization, this calls out the need for machines running at scale, savvy data scientists and lots of data. When these three things come together, AI has the potential to become truly magical though, as stated, not in some “sentient machine” sort of way. It’s still critical to point models at relatively “narrow” problems that play to machines’ strength, like pattern matching.

In addition, as Aible CEO Arijit Sengupta has stressed, it is for data scientists to remain pragmatic about their models. Sengupta runs a regular competition pitting high school students against trained, university-level data science students at Berkeley. The high schoolers nearly always beat the university students, he said, for the same reason that most corporate AI projects fail: “Data scientists and machine learning engineers are taught to look at ‘model performance’ (how well does a given algorithm do with a given data set at making a prediction) instead of business performance (how much money, in either additional revenue or cost-savings, can applying A.I. to a given dataset generate).” In the case of the competition, the high school students do a better job of focusing on the dollars and cents outcomes of their models, while the university students “get caught up on training fancy algorithms.”

It pays to keep things simple, in other words. And to focus on areas where it’s growing in strength.

SEE: Hiring Kit: Artificial Intelligence Architect (TechRepublic)

So where should enterprises look to use AI in the near term? According to a Stanford report, “One Hundred Year Study on Artificial Intelligence”, we’ve made “remarkable progress” in AI since 2016, with AI showing particular improvement in three key areas:

  • Learning in a self-supervised or self-motivated way
  • Learning in a continual way to solve problems from many different domains without requiring extensive retraining for each
  • Generalizing between tasks—adapting the knowledge and skills the system acquired for one task to new situations

With these parameters in mind, enterprises can move from “mostly failing” with AI to “mostly succeeding. It’s just a matter of using AI wisely.

Disclosure: I work for MongoDB but the views expressed herein are mine.