AI Stealing Jobs?… MIT Says "Humans Are More Efficient"

"High Implementation Costs... Only 23% Improved by AI"

Recently, concerns have been growing that artificial intelligence (AI) technology will replace human jobs across various industries. However, a study drawing attention has found that, when considering costs alone, human workers are still more efficient in the majority of jobs.

On the 22nd (local time), Bloomberg reported, citing research from the Massachusetts Institute of Technology (MIT), that only 23% of workers can be efficiently replaced by AI in terms of dollar wages.

[Image source=Reuters Yonhap News]

The AI boom sparked by OpenAI's ChatGPT has accelerated AI adoption across all industries worldwide. Global tech companies such as Microsoft and Google in the U.S., and Baidu and Alibaba in China, are investing huge sums in AI and competing to develop services. Due to the excessively rapid pace, 1,100 experts including Elon Musk, CEO of Tesla, AI scholar Stuart Russell from UC Berkeley, and Steve Wozniak, co-founder of Apple, called last March for a six-month pause on AI research. On the 14th, the International Monetary Fund (IMF) forecasted that 40% of jobs worldwide will be affected by AI.

The MIT Computer Science and Artificial Intelligence Laboratory stated, "Concerns that 'machines will take jobs' frequently arise due to rapid technological changes. Anxiety has grown again with the introduction of large language models (LLMs). We found that because the initial cost of implementing AI systems is high, applying AI computer vision automation is only efficient at 23%."

The researchers projected that even if the annual cost required for AI adoption decreases by 50% each year, it will not be until 2026 that half of labor can be replaced by computer vision tasks. They also predicted that human labor will still be more efficient than computer vision tasks in certain areas until 2042.

The research team analyzed costs by automating various tasks mainly in occupations in the U.S. where computer vision could be adopted, such as teachers and real estate appraisers. Computer vision is a technology that recognizes images through sensors and analyzes them based on AI deep learning techniques. It is used in various fields, from smartphone photo classification to object detection for autonomous driving.

As an example cited in the paper, bakers visually inspect ingredients daily for quality control, but this accounts for only 6% of their total work, the researchers analyzed. Therefore, while implementing cameras and AI systems can save some time and wages, the cost of adopting the technology is higher than the baker’s wages.

The researchers analyzed that AI utilization is expected to be high in retail, transportation, and logistics tasks such as those at Walmart and Amazon. The healthcare sector also emerged as a suitable area for AI adoption. They also forecasted that the launch of technologies like subscription-based AI as a Service (AIaaS) could increase AI utilization.

This study was conducted with support from the MIT-IBM Watson AI Lab and collected data on about 1,000 visual assistance tasks across 800 occupations through an online survey. However, at the current point, only 3% of tasks are efficient for automation adoption in terms of cost savings, but as data costs decrease and accuracy improves, up to 40% could be replaced by 2030.

This content was produced with the assistance of AI translation services.