AI beyond efficiency: Redefining goals for creativity and discovery

The much-discussed article “The AI We Deserve” got me thinking about AI design. Evgeny Morozov contrasts the current model of goal-oriented, problem-solving AI with what I’d call “exploratory AI,” which emphasizes process over specific outcomes. But doesn’t every process ultimately yield some kind of result? Even in art, broad exploration eventually produces a work of art in one form or another.

The author illustrates this with how children exhibit a sense of wonder by trying things out. Yet their seemingly aimless activities aren’t meaningless. In fact, they’re part of a finely tuned biological process that develops motor coordination, neural connections, boundaries, emotional regulation, and other necessary skills for adulthood. It is goal-oriented. 

Adults, too, experience a similar phenomenon: ideas often emerge when our minds are at rest, like during a walk or while washing dishes. These “idle moments” can be surprisingly productive—our brains keep processing in the background, often resulting in new ideas or solutions to problems we have on our minds. It’s not so different from children’s playful learning; both look unstructured but serve deeper purposes.

I also noticed the author used “goal pursuit” and “problem-solving” as if they’re the same. But, for instance, I might set a goal to read three books a month—not to solve any particular problem but purely for enjoyment or personal growth. So I believe you can build AI to pursue a goal without necessarily solving a problem.

The author critiques the influence of the “efficiency lobby” on AI design, arguing that corporate interests, heavily involved in AI development, prioritize efficiency at the expense of creativity and playfulness. However, the real tension may not be AI’s goal-oriented or problem-solving focus. After all, solving problems doesn’t necessarily mean sacrificing creativity. True efficiency—and innovation—comes from identifying the right problems to solve, which is a creative process in itself.

Ultimately, I think the issue isn’t whether AI is goal-driven or problem-solving; it’s how we define and pursue those goals and/or problems. In my view, goal-oriented approaches are preferable to problem-solving. And embracing a broader vision of what “goals” can be, we allow room for both discovery and ingenuity in AI’s design.