For Machine Learning to Change the World, Technology is Not Enough

The “New Electricity” Needs A Lot of New Wiring

Computer scientist and Coursera co-founder Andrew Ng has called artificial intelligence “the new electricity.” The comparison is especially apt for machine learning, a widespread form of AI concerned with using data to find patterns and make predictions.

Like electricity, machine learning is built on a foundation of scientific and technological innovation. And like electricity, it has applications in every industry that promise to change the world forever.

But the mere existence of an innovative technology isn’t enough to make it useful — and the story of electricity helps to explain why. It’s not enough to make a lightbulb glow in a lab. Real-world impact requires buildings that are wired for electricity and utilities that reliably produce it. Such a world-changing innovation requires massive infrastructure improvements to make it possible.

Harold Wallace Jr. writes on a Smithsonian blog that “Thomas Edison, George Westinghouse, and other inventors began introducing practical electric power systems in the 1880s. By the 1920s most cities and towns in America received electricity from either privately owned or municipal utility companies.”

In other words, the journey from invention to implementation took decades. And it wasn’t just physical infrastructure that needed to change. Human mindsets had to change as well. The BBC’s Tim Harford explains:

Until about 1910, plenty of entrepreneurs looked at the new electrical drive system and opted for good old-fashioned steam.

Why? Because to take advantage of electricity, factory owners had to think in a very different way.

. . .

But you couldn’t get these results simply by ripping out the steam engine and replacing it with an electric motor. You needed to change everything: the architecture and the production process.

And because workers had more autonomy and flexibility, you even had to change the way they were recruited, trained and paid.

Factory owners hesitated, for understandable reasons.

Of course they didn’t want to scrap their existing capital. But maybe, too, they simply struggled to think through the implications of a world where everything needed to adapt to the new technology.

Like electricity, the true impact of practical machine learning doesn’t just come from technical breakthroughs. It comes from the systems around that technology.

Advanced algorithms aren’t worth much without the people, processes, and infrastructure to apply them to real world problems. That’s why much of opportunity for machine learning in business today isn’t inventing even more sophisticated techniques - it’s putting together the pieces of the puzzle that are already on the table.

This is where non-technical leaders with vision and ambition can play a significant role. My colleague Robbie Allen likes to say that even if researchers were to stop pushing the field of machine learning forward (and they won’t), there would still be a ten-year backlog of machine learning projects inside enterprise companies.

Most business leaders already recognize that machine learning is a transformative technology. In a 2018 report, 86% of data science decision makers across the Global 2000 said ML was already impacting their industry.

Now, we need to build infrastructure to fully take advantage of ML’s power. Like factory owners at the turn of the 20th century, today’s leaders need to think in new ways and prepare to fundamentally change how their businesses work.

“Many companies are willing to spend money” on a new technology like artificial intelligence, says Andrew McAfee in his book More From Less. However, “surprisingly few are ready, willing, or able to make the changes required to fully exploit it.” Such intangible assets, he says , “allow a company to put new technologies to work, obtain higher productivity, pay more, and gain competitive advantage over rivals in an industry.”

There are still many business challenges preventing ML from reaching its full potential. But the problems are solvable, and the opportunity is great.

In a 1959 paper credited with coining the phrase “machine learning,” computer scientist Arthur Samuel wrote that it was “now possible to devise learning schemes which will greatly outperform an average person and that such learning schemes may eventually be economically feasible as applied to real-life problems.”

Note Samuel’s hope for eventual practical applications. As the history of electricity illuminates (wink), changing the world takes time. Quoting the BBC’s Tim Harford again:

Come the 1920s, productivity in American manufacturing soared in a way never seen before or since.

You would think that kind of leap forward must be explained by a new technology. But no.

The economic historian Paul David gives much of the credit to the fact that manufacturers had finally figured out how to use technology that was nearly 50 years old.

It’s been 60 years since Arthur Samuel imagined economically feasible learning schemes. We live in the future that he imagined. It’s up to us to make it work.

James Kotecki is the Director of Marketing & Communications at Infinia ML, a team of data scientists, engineers, and business experts putting machine learning to work. Special thanks to Infinia ML Chief Scientist Larry Carin for the idea that sparked this piece.