Our exclusive speaker Yossi Sheffi is an MIT professor and a leading expert on supply chain, resilience, and risk management.
Key Takeaways
- Talent development may get harder as junior work disappears.
- AI is moving into cognitive work.
- Entry-level knowledge jobs face the most risk.
- Senior workers may gain the most from AI support.
- Professions like law, coding, and accounting may change fast.
From Doers to Deciders – Dr. Yossi Sheffi
Originally posted to LinkedIn by Dr. Yossi Sheffi—headings and key takeaways added by the Mollie Plotkin Group editors.
AI that performs cognitive work will handle the drudgery, reducing the need for entry-level workers but letting more experienced ones focus on judgment and decision-making.
How Automation Has Historically Changed Work
Before 2023, automation was primarily used to simplify complex jobs and replace them with simpler, more structured and codified activities. This transition can be seen throughout the past three centuries, from the mechanization of textile production in 18th-century England, to the Ford assembly line in the beginning of the 20th century, to Uber in the beginning of the 21st century. The mechanical looms replaced master weavers, the assembly line replaced the artisan teams building one car at a time, and Uber triggered a replacement of taxi drivers with knowledge of local points of interest and streets with a horde of independent operators with no required knowledge beyond the ability to operate their app. In all cases, the replacement workers were performing less-demanding tasks and, not incidentally, were paid less.
Over time, many technological innovations made jobs in supply chain management easier. These include, for example, technologies that help operators steer and control trucks, airplanes, trains, and ships. Similarly, systems such as pick-to-light and augmented reality (AR) glasses simplified warehouse work. Modern robots further simplified warehouse work by bringing storage racks from the aisles to the order pickers instead of pickers going out into the aisles to locate and retrieve items. This “goods to person” concept is similar to the Ford assembly line that brought the work to the people on the line rather than the other way around.
Most automation innovations focused on blue collar work – simplifying it and reducing the expertise required by using fixed logic focusing on physical labor and structured digital tasks. The result was a decline in middle-skill jobs. High-level managerial jobs and physical, low-skill service jobs (which robots could not master) remained more or less constant. Then came artificial intelligence (AI) and large language models (LLMs).
How AI and LLMs Change the Equation
Modern LLMs do not follow a rigid “if-then” script. Their predictions (e.g., the next word in a sentence) are based on training on vast data sets, which allows the model to absorb the structure of language and predict a reasonable “next word.” These models can now perform cognitive work.
Given how they are trained, the models are best at codified cognitive work, which explains why entry-level cognitive jobs are now at risk. The knowledge of employees who are getting started in their careers is based on codified intelligence gleaned from books and other general educational sources. Modern LLMs, however, can absorb such knowledge from the training data at speeds and with a reach that are far beyond the capabilities of humans. Experienced workers, on the other hand, have tacit knowledge—rooted in understanding context and based on experience—that is not codified and takes years to acquire. These workers are likely to find that LLMs augment their work and increase their productivity. The models will take on some of the drudgery tasks and leave experienced workers to focus their efforts on high-level thinking and decision-making. For these higher-level workers, AI models will provide decision support, not task execution, speeding up the performance of their responsibilities.
How Jobs May Change
Past industrial revolutions simplified many jobs’ requirements, enabling companies to hire lower-level workers. Of course, some jobs disappeared. For example, elevator operators’, telephone-exchange operators’, and telegram messengers’ jobs are gone. In the beginning of the 20th century, over 40 percent of the US workforce was employed in agriculture. Today it is about 1.5 percent. The hard work in the fields was mechanized.
Today’s AI, with its ability to codify a growing body of cognitive work, is likely to reduce the number of entry-level jobs for lawyers, software coders, accountants, and other knowledge-based professions. In many cases, the work is simplified to the extent that customers can perform some or much of it themselves—think of tax-return software and the ubiquitous online and over-the-phone self-services, most of it powered by AI bots.
What This Could Mean for Education
This trend even extends to teachers. My guess is that in 5 to 7 years, most high-level education (from high school on) will move to the “customers” in a do-it-yourself mode, with a smaller number of teachers (or professors) overseeing the AI. This is already starting to happen: In today’s middle school and high school classrooms, a growing number of teachers are only overseeing the work that students are doing online, rather than actively delivering lesson plans in a classroom. In that sense, the teaching profession will dovetail with developments in industry, where managers’ main job may include overseeing autonomous robots and software agents.