The concoctions sold as snake oil didn’t contain what was claimed, it was largely found to be ineffective and in extreme case led to the loss of life. AI snake oil literally means AI that does not work.
However, it does favor us by shining what may not work because research in more than a dozen AI fields have found far-reaching credibility crises. AI Snake Oil uncovers such rampant claims and warns of the dangers of AI when it is controlled by largely unaccountable big tech corporations.
Amidst so much publicity around AI, an amazing hype around artificial intelligence has been generated at the cost of human wisdom. Driven by the desire to quick fix solutions, the hype comes around with questionable generative and predictive answers. Should it not be the responsibility of researchers to separate the milk from the froth?
The professor-student team of Narayanan and Kapoor at the Princeton University have cut through the hype with some clear and crisp writing on how AI fails us daily, and how it might one day benefit us. Interestingly, they comment on new developments in AI in their newsletter AISnakeoil.com.
Jobs threat
That AI will cause sudden mass joblessness seems farfetched; however, it will change the nature of many jobs and decrease the demand for other jobs. Previous waves of automation had similar impact, albeit more abrupt. When typewriter was replaced by word processer the transformation was significant, as it called for a change in the nature and skill of job.
Rarely been a job category been replaced entirely by technology, only elevator operator seems to have disappeared due to automation. Automation often decreases the number of people working in a job or sector without eliminating it.
Called automation paradox, the most common type of impact is a change in the nature of job duties. Last mile phenomenon of automation is of critical importance: it takes previously done job but creates new types of needs for human labor. For those whose jobs are already automated, however, the prospects could be scary.
One must appreciate that Narayanan and Kapoor have made things simple, which others have tried to make it complex. AI Snake Oil offers a breath of fresh air about both AIs, predictive and generative.
As of today, predictive AI is not on a firm footing. Falling prey to snake oil is crucial when it is known how it fails and even harms people. The problem is how much data we can have and how effective our models are likely to predict the future. Limits to predicting future based on the past data and concurrent trends is fraught with uncertainty. Machine learning therefore can only generate the plausibility of what the future might hold.
Proponents of AI know the limits but do not want any reputational damage as yet. A 2023 paper claimed that machine learning could predict hit songs with 97 percent accuracy, however, in reality the study’s results are anything but false or even bogus. Earlier studies do bear testimony to it, although such papers about ‘frightening accuracy’ have the potential to revolutionize the music industry.
Who would not want to spin money based on such a hype? Overall, more than a dozen fields have compiled evidence of widespread flaws but none of it has been publicly accepted, but the supply of snake oil comes from companies that want to sell predictive AI.
Capital theory
“Fears about automation/technology are fears about capitalism.” As companies are driven by profit, AI is expected to generate profit. More than technology, it is capital which is at the core of the entire debate. Big Tech companies have gotten so rich off of AI that they can easily mold public perception.
Academic research and tech journalism too are completely dependent on industry funding. It is this aspect that Narayanan and Kapoor have tried to bring up honestly in their book. Painting AI with a single brush is tempting but flawed, they say. AI Snake Oil is all about why there is so much information, misunderstanding, and mythology about AI.
There is a collective learning, but non-acceptance of common follies. Everyone is found guilty of telling untruths – if not to one another, then to themselves. Certainly, everyone in AI fraternity is found guilty of that. Most of the time, what we think of as truth is threaded with self-serving distortions. AI Snake Oil has everything you ever wanted to know about AI.
by Arvind Narayanan & Sayash Kapoor
Princeton University Press, USA
Extent: 348 pages. Price; Rs. 699,
First published in the HinduBusinessLine on Nov 25, 2024.