Mapping the global impact of AI

Every week, I discuss the dance of technology and society with a brilliant mind. This week I had a thrilling conversation with Dr. Kate Crawford about why we should understand AI as much more than just code and algorithms: it’s really an industry built on a global network of resource extraction, human labor, and data collection. 

Kate is senior principal researcher at Microsoft Research, co-founder of AI Now and Visiting Chair for AI and Justice at the école Normale Supérieure in Paris. Her book Atlas of AI is out now. 

Below is an excerpt from our conversation. To hear the full discussion, listen here.

***

Azeem Azhar: Kate, you're very well known for a piece of work you did with Vladan Joler called The Anatomy of an A.I. System which has made it into the permanent collection of the MoMA in New York and the V&A in London. What did you learn from that project?

Kate Crawford: Anatomy of AI transformed my thinking! We mapped out what it takes to make an Amazon Echo. We researched every component. We traced the mining supply chains. We looked at smelting. We looked at container shipping. We looked at all of the data pipelines where the data is harvested from these devices and interpreted. And then we looked at the end of life where these devices are thrown into tips in places like Ghana and Pakistan, generally fewer than four years from when they were purchased! 

That project really changed my thinking: I realised we have to move away from this idea of artificial intelligence as being immaterial and abstract and in the cloud, to see it as a profoundly material technology, something that is made from the earth, from human bodies and from huge amounts of data.

Azeem Azhar: That’s a great overview of your thesis. At the heart of that lies a more abstract idea, which is to challenge this notion that intelligence is a thing on its own. You make a very compelling argument that the way that intelligence is thought of by the A.I. community is problematic.

Kate Crawford: There’s a sort of Cartesian dualism in artificial intelligence where minds are regarded as entirely separate from their connection to earthly bodies and states. I see this as an original sin for the field - this idea that computers are minds and that they're anything like human intelligence.

You see that when Elon Musk talks about wanting to upload his brain to the cloud. 

In actual fact, what we see with machine learning is large-scale statistical analysis, prediction and optimization. Very different to the sort of ways in which humans are in the world and think in relation to each other, in relation to their environments.

Large scale training data sets have a baked-in world view. We're getting the answers, if you will, that we already know. So I think we have to be very skeptical about these forms of ‘intelligence’ and see them as products of their time.

Listen to our full discussion here