How to keep AI from taking your job

By Vedant Misra | January 5, 2017

With the emergence of technology that automates knowledge work, an entirely new part of the labor force is worried about job security. These concerns were once isolated to people who did repetitive physical labor, but today it can seem like any work at all might be replaced by AI. The 15-to-30-year forecast of the global labor market is a giant question mark.

How will we navigate the transition? How can you know how susceptible you are?

First, recognize that machines don’t need to do everything you do in order to replace you–they just need to replace enough of the value you create, at a low enough cost. That means machines don’t need to be able to crack jokes at meetings or empathize with your team. They just need to replace most of the value you create for whoever pays you.

This might sound pessimistic, but really it’s the opposite, because it gives us a good rule of thumb by which to avoid being displaced by technology.

Focus on creating the kind of value that machines can’t.

There’s a simple rule for assessing whether machines can learn to do something or not: is there training data for it?

AI might seem abstract and frightening, like it can do anything, but the fact is, every intelligent machine system we have ever built operates on the basic principle of learning to make predictions using a dataset to learn from.

Self-driving cars only exist because it’s possible to generate the data we need to train them, by attaching cameras to cars. Of course, there’s a lot more that goes into the process of developing the algorithms that control the car, but without the training data, it isn’t even possible to start. For learning, training data is not sufficient, but it is necessary.

This is also true of your brain. If you’re like most people, your brain acquired the training data to learn to drive a car when you were somewhere between 14 and 20 years old. Inside your head is a massive and complex web of neurons that are responsible for processing the stream of input from your eyes. You bootstrapped your ability to avoid obstacles and control your motor functions in response to environmental stimuli by practicing inside a car. There is no other way to learn to drive a car—you can’t read a book or attend a lecture and expect to be able to drive well.

Fortunately, there is still a verywide spread of human activity for which the training data is trapped in humans’ minds. This is nowhere more true than for interpersonal interaction.

At the core of B2B sales is the ability to understand what people want. We’re nowhere close to being able to build machines that can understand the subtleties in what people want. We try to learn that from the data people leave online. We use technology install data with the hope that someone who already uses Wordpress or Marketo might be the kind of person who also uses whatever it is you’re selling.

But most of the information about people’s preferences isn’t computable; it’s still trapped in human brains. This will gradually become less true as we develop and deploy machines with microphones and cameras that hear us talking and watch us interacting with each other. If we let them observe us, they’ll learn exactly what we prefer when we talk to each other and interact with each other.

But that isn’t happening any time soon.

As long as you keep an eye on what useful information is still trapped in human brains, and use that information to deliver value, there will be a job for you.