Johan Fredrikzon: Unveiling the Dynamics of AI and Its Societal Impact
In a recent update, Johan Fredrikzon, an international postdoc at the division, shared insights from his participation in two conferences focusing on artificial intelligence. His presentations shed light on the hidden aspects of AI development and its historical context.
4S – Society for Social Studies of Science: Sea, Land, Endangered Ecologies, Solidarities
Honolulu, November 8–11, 2023
At the big STS Conference, held in Honolulu, Johan Fredrikzon delivered a presentation titled "The Invisible Work of AI Alignment and Its Historical Foundations." His talk delved into the practical efforts to make AI more predictable and secure for users, emphasizing that this work is carried out by thousands of low-wage workers in the global South, particularly in Africa and South America.
Despite its significance, this labor remains concealed from the public eye, contributing to the prevailing myth of AI as autonomous and automatic technology. Johan argued that the manual performance and human aspects of this work remain hidden to sustain the ongoing AI hype. Drawing parallels with the early digitalization era of the 1960s and 70s, he highlighted historical examples of hidden labor contributing to the myth-building around automatic technology.
Annual conference for the Foundation for Legal Information/Stiftelsen för Rättsinformation
In addition, Johan opened the annual conference for the Foundation for Legal Information (Stiftelsen för Rättsinformaiton), on November 14 with a keynote lecture. His talk, titled "From the Dream of Automation to Judgment as Labor," explored the historical dynamics between labor and machines. Drawing on examples from the history of automatons, Johan insisted that we should oppose situations where our work is defined as that which machines cannot do. Instead, we should adopt the attitude where intelligence and ability is the result of our measured use of technology.
The fact that we ourselves and our tools bring different capabilities to the table, becomes particularly evident in the legal profession as it turns to new methodologies. For all their massive training, current large language models (LLMs), are incapable of knowledge representation which results in the inability to differentiate between true and false, a condition Johan terms epistemological indifference in a forthcoming article in the journal Critical AI. Such shortcomings, however, need not be devestating as long as users adopt a cooperative yet critical stance toward these new tools regarding what they can and cannot offer.
With an impressive rate of text processing and rough drafting, but with poor reliability and without any capacity for knowledge, paired with a deceivingly high confidence in its replies, generative artificial intelligence is unfit to perform anything useful on its own. To be of any value to legal professionals, AI requires of them to embrace the core of their expertise: careful consideration. In his keynote, Johan termed it Judgment as Labor.