Aug 31, 2022 - Technology

Demystifying AI

Illustration of robot hands holding a clipboard.

Illustration: Shoshana Gordon/Axios

Arguably the most consequential technological transformation of our lifetimes is coming faster than we think — and we're not adequately prepared.

Why it matters: Artificial intelligence is rapidly improving at human-like tasks, like language and reasoning.

  • And the number of plugged-in experts who say we have years and years to prepare ourselves for AI’s rise is dwindling, the New York Times lays out in a stunning new report.

Case in point: Five years ago, AI's big win was that AlphaGo — a machine built by Google — beat humans at the game Go, which was one of the last games where humans still had the upper hand.

  • But even that seems minor compared with last year when Google's AI solved a molecular biology problem that had confounded scientists for decades, the Times reports.

What’s happening: We asked the top experts in our newsroom exactly what AI can and can’t do — and how we should be thinking smartly about it.

Yes, AI is getting smarter. It will keep getting astonishingly better at suggesting possibilities, like videos to watch, and extending patterns, like next things to say in a chat, Axios managing editor of technology Scott Rosenberg tells us.

  • But for the foreseeable future, people will remain the arbiters of which possibility is actually best, or which pattern is useful.
  • In this sense, AI will always be a tool — but one so powerful and able to operate independently enough that we can be forgiven for mistakenly seeing it as autonomous.

But AI's results are only as good as the algorithms and the data used to train the systems, both of which are susceptible to bias, Axios chief tech correspondent Ina Fried notes.

  • That helps explain, for example, why chatbots may quickly devolve into racist speech. Or why AI-generation pictures of a doctor are likely to show a white man.

🧠 Reality check: AI is far from human. Despite high-profile claims to the contrary, the machine-learning-based AIs we know today can't achieve sentience because they don't have bodies and senses to create a sense of self in time and place, Scott says.

  • That could change! But no one knows how to make it happen today, or even has a good theory for how to get there. And before we go down that road, we'd better have a long conversation about whether it's even a good idea.
Go deeper