Why Machine Learning isn’t about machines

How will machine learning change us as a society? It’s now time to ask this question — before we start building products and services that have unintended consequences.

I wanted to start this blog post by referencing “the turn of the the last century.” I realized that would put me smack dab in the middle of the Y2K hysteria, and not in the birth of bureaucracy (whence such hysteria came).

No, we only know what the “turn of the century” means, many years in retrospect. Now, we can look back on the year 1900 and see quite clearly that its significance was the shift from idiosyncratic, family-run, and sometimes chaotic organizations toward professional management, and of course, bureaucracy.

It was hard to see while it was happening, but Max Weber saw it (perhaps that’s why he had a nervous breakdown). Weber saw that we had begun to run our businesses and governments with standardized rules, and standardized hierarchies. No longer could the boss’s son waltz in and tell everyone what to do, unless he had an actual job title. (Well, that was the idea anyway; we apparently still let the boss’s kids take jobs they’re not qualified for).

This was radically new and had huge implications for how we purchase, exchange, work, and live. Bureaucracy became the irrationally rational norm; rules were to be followed even if they made no sense.

Which brings me to machine learning. Machines can learn if we give them the tools to learn, and the data to help them practice. But they cannot see what Max Weber saw. Machines cannot know they are creating an irrational bureaucratic hellscape — and nor would they care. They are very good at things humans are bad at, namely, vigilance and repetitive tasks. We should let them do those things.

But we should not let machines make decisions about rules, about whether the boss’s son is qualified, or other culturally and socially important questions. At the turn of this century, we are making machines that can do all of those things, but we are not pausing to evaluate whether we should.

Historians like to say that the 19th century did not really end on the arbitrary date of December 31, 1899, but instead on the more auspicious and socially meaningful date of November 11, 1918. It was only then that humanity realized what its changes had wrought, what horrors we had invented, and that humans themselves must take responsibility for those changes. I would argue we need to do the same now, before an equally socially meaningful date in the future.