What should AI do?

What should AI do?

That may seem like a pretty straightforward question — accelerate science, cure diseases, automate our work, and so on. But where do we draw the line?

To answer what exactly AI should do, we first have to ask what we are willing to abstract away.

For the longest time, humans have created abstractions to manage the finite amount of attention we can distribute, giving up control over details we consider less important so we can focus on what matters more.

Abstraction is in everything we create. In programming, we create systems of lower- and higher-level languages — each layer abstracting decisions about memory allocation, timing, and hardware, giving us less control over the machine but more control over what and how we want to build. We see it in art, where Picasso and Kandinsky replaced recognizable objects and details of reality with fields of color, contrast, and shape. By freeing art from the details and rules of reality, these artists were able to devote more of the canvas to expressing raw emotion or telling a story.

Abstraction has traditionally been bespoke work: each system designed for a particular domain and each reflecting an intentional decision about what we are willing to lose control over and what we choose to prioritize instead.

Artificial intelligence introduces abstraction on a different scale. Instead of abstracting one particular layer of the world, AI is becoming a general-purpose layer of abstraction itself — whether in the process of developing a game, writing a novel, or even communicating with others.

But when the abstraction of intelligence becomes this general and accessible, it becomes a slippery slope. AI can remove the friction surrounding what we care about, but it can just as easily remove the choices, experiences, and meaning that made us care in the first place. For the first time, the question is not simply what can be abstracted away, but what should be.

By asking artificial intelligence to automate game development, we may abstract away not only the implementation but also the design choices that express our taste. By asking artificial intelligence to write our novel, we may abstract away not only the act of typing but also the interactions, settings, and nuances through which we give the story meaning. By asking artificial intelligence to create our products, we may abstract away not only the repetitive labor of making them but also the marks and decisions through which we tell our stories.

There is a fine line between the abstraction that enables us to focus on what matters and the abstraction that takes away our ability to choose the type of dent we would like to leave on the universe.

But in a world racing to create the AI that solves the next big problem or automates the next job, I find it increasingly important to spend the time and effort to figure out what in life is worth abstracting.

Because at the end of the day, our duty is not to create an AI that can do everything.

Our duty is to create a world where AI enables us to become more curious.

A world where AI enables us to become more creative.

A world where AI enables us to become more compassionate.

A world where AI enables us to become more human.