The definition of technology remains elusive, and has been prone to large shifts through the decades alongside modernization. A generally time-independent and agreed-upon definition states that technology is a means to utilize systems, machines, or devices to leverage scientific knowledge for practical human applications. By this definition, artificial intelligence is a technology, using computational means and systems to better understand and make sense of data. We can view AI as a technology that is used for transporting and converting data for human consumption in the same way that we use an automobile to transport us from one point to another, or a generator to convert energy into electricity.

Upon coming across Atlas of AI, a new book by USC Annenberg research professor and Microsoft senior principal researcher Kate Crawford, my initial inclination based on its title was to expect another book focusing on two recurring themes: How AI can be used to transform new industries, and where the newly generated global centers of power will reside. Upon reading the book — and much to my delight — it was about neither.

As an evolutionary scientist, I was thrilled to see Stephen Jay Gould’s work referenced when describing errors in systematic classifications, an issue that sits at the cornerstone of problems related to AI bias. As a geospatial/AI/data scientist, I was also immediately hooked by the “atlas” reference in the title. An atlas — by definition, a collection of maps — is supposed to provide the reader with a visual frame of reference regarding spatial relationships. And the Atlas of AI does just this.

The analogy of “the cloud” has allowed us, for decades, to distance ourselves from internalizing the costs associated with the technologies that sit at our fingertips.

Crawford very effectively describes the world of AI through the spatial and temporal relationships that we would expect when perusing the pages of an atlas. Instead of looking forward and trying to forecast where society will most effectively utilize AI applications in the decades ahead, she takes a life-cycle-analysis deep dive into the material side — into the nuts and bolts — of AI.

Beginning with chapter 1, aptly titled “Earth,” Crawford describes what AI actually is. It becomes apparent that the analogy of “the cloud” has allowed us, for decades, to distance ourselves from internalizing the costs associated with the technologies that sit at our fingertips — by allowing us to use a metaphor that conjures up an image of something fluid and far away. Technologies that live in the cloud provide an abstract mental barrier that makes it more difficult to distinguish between costs and benefits. Simply put, by highlighting this conceptual misalignment, Crawford creates a new dimension and set of tools through which we are able to create such a cost-benefit analysis.

Crawford also answers the question, “What is AI actually composed of?” In short, AI is a technology like any other, comprised of both commodity and specialty materials, each carrying their own portfolio of environmental and human labor costs, built upon societal inequities, and — largely — serving the privileged. Throughout Atlas of AI, Crawford literally brings this notion of “somewhere else” down to our planet, and does so in a convincing and thought-provoking manner. So while we rightly continue to applaud AI applications that support the rise of renewable energy, the proliferation of electric and self-driving vehicles, and systems that minimize resource extraction and use, all of these AI-powered areas have their roots firmly planted in soil enriched with massive computational architectures — architectures that use vast quantities of rare materials and relying on massive energy requirements.

Upon completing the book — which I look forward to re-reading — my belief remains unchanged and hardened that, as applied researchers, our goals should remain looking for ways to effectively use AI to help society cope with a growing list of globally interconnected problems. However, we should also try to learn from history and not solve one class of problems, only to create another. Crawford’s Atlas of AI serves as a guide that should influence our thinking in the decades ahead.


Michael Ferrari is a senior fellow at Wharton Customer Analytics and chief data scientist at Engine No. 1.