Earlier today, MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) released a new study addressing these three challenges. Various attempts have been made to anticipate and forecast the effects of current AI technologies, such as extensive language models, on individuals’ livelihoods and entire economies in the foreseeable future.
(Image Source: https://ca.finance.yahoo.com/)
According to Goldman Sachs, artificial intelligence could automate 25% of the labor market in the coming years. McKinsey predicts that by 2055, over half of all labor will be led by AI. According to a poll by Princeton, NYU, and the University of Pennsylvania, ChatGPT alone may impact 80% of occupations. Furthermore, according to research by the placement agency Challenger, Gray & Christmas, AI is already displacing thousands of workers.
However, the goal of the MIT study was to go beyond what they refer to as “task-based” comparisons and evaluate the likelihood that firms will truly replace human labor with AI technology and the viability of AI performing specific functions.
In contrast to what one might anticipate, including this writer, the MIT researchers discovered that most tasks previously considered vulnerable to AI displacement aren’t, at least not yet, “economically beneficial” to automate.
Neil Thompson, a research scientist at MIT CSAIL and a paper co-author, believes the main lesson is that the impending AI disruption may occur more gradually and subtly than some pundits have predicted.
A crucial disclaimer is that the study only included positions involving visual analysis, such as quality-control product inspection at the end of a production line. The researchers left it to future research to examine the possible effects of text- and image-generating models, such as ChatGPT and Midjourney, on laborers and the economy.
This study aimed to determine what tasks an AI system would need to complete to replace workers. Hence, the researchers conducted a poll of workers. The cost of developing an AI system that could accomplish all of this was then estimated, and it was also determined whether or not companies, particularly “non-farm” American companies, would be prepared to cover the system’s startup and ongoing costs.
The researchers use the example of a baker early in the investigation.
The U.S. Bureau of Labor Statistics estimates that a baker spends roughly 6% of their time ensuring food quality, a duty that artificial intelligence (AI) might (and is) automating. If a bakery with five employees and an annual salary of $48,000 were to automate food quality checks, it might save $14,000. However, according to the study’s estimations, a minimal, in-house AI system capable of performing the task would cost $165,000 to implement and $122,840 annually to maintain.
However, the report considers self-hosted, self-service AI systems offered by companies such as OpenAI, which require task-specific fine-tuning rather than complete training. However, the researchers claim that many low-wage, multitasking-dependent positions are too many for a corporation to automate economically, even with a system that can be purchased for as little as $1,000.
The researchers note that “even if we consider the impact of computer vision just within vision tasks, we find that the rate of job loss is lower than that already experienced in the economy.” “It would still take decades for computer vision tasks to become economically efficient for firms, even with rapid cost reductions of 20% annually.”
The researchers acknowledge several limitations of the study, which is commendable. For instance, it ignores situations in which artificial intelligence (AI) can supplement human labor rather than replace it (e.g., analyze a golfer’s swing) or develop entirely new roles and responsibilities (e.g., maintain an AI system). Furthermore, it does not account for all potential cost savings from pre-trained models such as GPT-4.
(Information Source: Techcrunch.com)
