A critical examination of the dynamics between error and intelligence in the history of AI. Error has, at the same time, been the problem (“human factor”) that smart machines should fix and the stupid mistakes that machines make which require people to step in. Error, moreover, is the foundation of machine learning: trial-and-error, in its cybernetic understanding of negative feedback, is implemented as the error-correcting mechanism known as backpropagation. To complicate things further, both engineers and philosophers have considered error to be instrumental to independent and autonomous behavior in organisms and machines alike — deviance from norms or a fundamental break with operations has been viewed as the condition of both unoppressed individuals and truly intelligent non-organic systems. The concept of error in the history of AI is not a simple story, but an important one if we hope to understand how we have co-evolved with these tools.
An accessible book for a Nordic audience, arguing that when it comes to artificial intelligence, we need to keep two thoughts in our heads at the same time. The book is a historical study of AI that dodges both a dismissive and a celebratory attitude toward the topic of intelligent-seeming machines. It looks at AI from aspects of knowledge, labor, autonomy, art, risk, error, understanding, and futures.
The book demonstrates how the origins of our data-driven present can be traced back to changes in fundamental operations such as modeling, linking, and reuse in the 1960s and 70s. Already half a century ago, the downplaying of human labor in digital work, imaginaries of data as a “raw” resource, the management of data as “capital,” and an algorithmic understanding of the natural environment were articulated by the big tech of the day — the agencies of the welfare state, insurance companies, and research institutions in the natural sciences.