Artificial intelligence and the future of design
Classical uses of artificial intelligence—the kinds of things you learn how to do when you're learning about machine learning—usually involve finding patterns in existing content: classifying images, for instance, or interpreting a block of text. But I'm fascinated by AI techniques that generate content, like classical music, trippy images, and convincing voices.
I spent a lot of the summer thinking about how generative algorithms will affect the process of design, and, by extension, the process of optimizing everything. Here's why we might someday talk about "discovering a design" instead of "creating a design."
(In the image above, Autodesk's Dreamcatcher generative-design software iterates through refinements to a motorcycle component.)
One way you can tell AI is progressing is that AI ethics suddenly feel urgent. With near-perfect AI would come the ability to intuit private information—like Target knowing when someone is pregnant, but with practically any characteristic. Sprawling AI models could tie together unrelated pieces of data to undermine privacy. Authoritarian-government CTOs are probably churning out AI-strategy memos right now.
Amazon, Facebook, IBM, Google, and Microsoft have launched an organization called the Partnership on AI to Benefit People and Society (a cumbersome name of the kind that ought to generate a catchy acronym; alas). PAIBPS aims to come up with guidelines for ethical and transparent AI.
Right now the most sophisticated AI is only available to a handful of companies with lots of resources that, for now at least, seem to be approaching it responsibly. This is a little like nonproliferation; when only three countries have the resources to build a bomb, you can get them all in the same room and come up with formal mechanisms to control the technology. But pretty soon very sophisticated AI will become accessible much more broadly, and then we'll have suitcase AI to worry about.
Another important question: whether AI will take our jobs. Tim O'Reilly points out that AI won't take our jobs until all the important problems are solved, and there are plenty of important problems left.
|