The new research reveals that the expected automation of around half of all human-performed jobs with the help of AI could come at a premium cost, which employers would likely not be willing to pay.

A fresh study ‘Beyond AI Exposure: Which Tasks are Cost-Effective to Automate with Computer Vision?’ by a team of researchers from the Massachusetts Institute of Technology (MIT) and IBM has estimated that numerous job positions previously identified by other academic works as candidates for AI substitution wouldn’t be cost-effective to automate in the current economic conditions.

Contrary to the earlier expert forecasts that AI could and would replace about half of all humans in the labour workforce, the new research from the MIT-IBM Watson AI lab concluded that taking into account today’s costs, “only 23% of worker wages being paid for vision tasks would be attractive to automate.”

The MIT/IBM scientists admit that AI has immense potential for job automation. Numerous experts have predicted that AI will be implemented in almost all industries over the next 10 years. Some business leaders like SoftBank CEO Masayoshi Son even believe that generative artificial intelligence will soon achieve superiority over the human mind. The leaders of OpenAI also stated that over the next ten years, artificial intelligence will be able to perform any work currently done by humans.

At the same time, the group of MIT and IBM researchers decided to focus on the economic feasibility of the predicted substitution rather than its theoretic potential. In this case, the chances of AI substituting half of human employees seem much slimmer.

“The previous literature on ‘AI Exposure’ […] attempts to measure an overall potential for AI to affect an area, not the technical feasibility and economic attractiveness of building such systems,” noted the researchers.

Focus on the potential doesn’t take into account technical and financial obstacles that could affect rapid AI adoption. First of all, the study highlights the importance of the high cost of AI training and implementation.

Even though popular AI systems like ChatGPT can be fine-tuned to fit specific purposes, scientists note that many organisations would face the dilemma of either trusting their proprietary data and internal processes to a third-party vendor or creating their own AI systems. The latter requires lots of investment, while the first variant can be too risky when one deals with sensitive data.

In addition, both short- and long-term economic factors, which may change in an unpredicted way, would definitely play a significant role in the business decision-making process. As we know, in uncertain times, investors tend to shift their assets to traditional and predictable markets rather than innovations and high-tech.

Nevertheless, the research team doesn’t negate the AI-driven future of work. The main point of the research is that it will likely happen less swiftly than many people expect and have a gradual nature.