Bursting the AI Illusion: A Review of Chapter 9 in Acemoglu’s and Johnson’s ‘Power and Progress’ Book
Authors Dr. Daren Acemoglu and Dr. Simon Johnson, renowned MIT professors, released a thought-provoking book titled “Power and Progress” on May 16, 2023. This comprehensive exploration into the roots of technology, its architects, and the pursuit of a more pro-human version is divided into 11 insightful chapters. Today, I will delve into Chapter 9, “Artificial Struggle.” For my review of the previous chapter, please click here.
In “Artificial Struggle,” Acemoglu and Johnson unpack the Artificial Intelligence (AI) illusion — a widespread optimism around AI’s potential to enhance productivity and effectuate sweeping economic and societal benefits. However, this optimism, they argue, is an illusion stemming from a misplaced focus on automation rather than the enhancement of human skills.
The authors believe that AI should concentrate on complementing human skills and improving machine utility rather than attempting to automate skills humans have perfected over the years, harnessing tacit knowledge and expertise. Quoting MIT polymath Norbert Wiener from his 1954’s book,
“The Human Use of Human Beings: Cybernetics and Society,”
the authors underscore the necessity for machines to be valuable and complementary to humans; otherwise, their benefits to humanity are lost. Wiener compared automation to slave labour, highlighting the need to avoid detrimental economic consequences.
Acemoglu and Johnson point to several instances where over-promising led to under-delivery. They cite Geoffrey Hinton, co-creator of modern deep-learning methods and Turing Award winner, who predicted in 2016 that radiologists would soon become obsolete due to advancements in deep learning. The authors also reference optimistic predictions made by the chief technology officer of Google’s self-driving car division and Elon Musk’s forecast for Tesla in 2019. Despite these rosy forecasts, the AI approach, they argue, has only resulted in modest corporate gains and significant losses for society and workers.
The chapter underscores two fundamental reasons for these disappointments. First, current AI approaches rely heavily on pattern recognition and prediction, which must be equipped to automate many human skills. The situational nature of intelligence is a challenge for these models as codifying every possible situation is not just tricky but also runs the risk of ‘overfitting’ — introducing irrelevant information that leads to inaccurate predictions.
Second, the authors raise concerns about the potential for invasive surveillance, where massive data collection from employees and customers leads to high-monitoring environments. While the authors concede that employers are within their rights to ensure task completion, they argue that the traditional balance of motivation — derived from fair wages and workplace benefits — is now skewed in favour of pervasive monitoring, which can exploit workers without improving productivity. They term this a “rent-shifting activity” that unjustly redistributes productivity gains away from the workers.
The chapter concludes soberly: the current AI path disproportionately favours capital and highly skilled production workers, leaving most low- and middle-income countries disadvantaged. The authors drew parallels with economist Frances Stewart’s observations in the 1970s when she noted that importing Western technology into developing countries exacerbated inequality and poverty.
“Power and Progress” forces us to grapple with the unfulfilled promises of AI and the artificial struggle it has imposed on society. The question we must confront is not whether we can advance AI capabilities but how we should do so in a way that benefits humanity.
