(Revised, January 21, 2024)
What impact do you believe AI will have on the American worker?
The impact will be dramatic but varied.
AI depends on data, and, in particular, generative AI uses LLMs (large language models). LLMs are based on past data, amply processed, and used to generate plausible results based on probabilities drawn from such data regarding “what comes next” in a mix of semantics and syntax. Domains not expressed in some form of language (text, images, audio) will be largely immune to the impact of generative AI replacing such work.
Routine Manual Work
In the long run, routine manual work will be substituted for better materials and automation (manufacturing, transportation services, cashiers, and retail banking). This is the realm of the robots. This is "blue-collar AI." Such capital-labor substitution requires capital investment, which small firms may not have. Historically, the decline in routine manual work is well documented.
Non-Routine Manual Work
Further, unless there is ample consolidation in industries for employment with non-routine manual work (like residential home construction, restaurants, hair salons, police work, emergency medical work, etc.), automation will take a long time to emerge, if ever. Some kinds of work may not allow capital to be substituted for labor; the return on investment may not be there.
However, such work can be informed by LLMs as assistants, creating a kind of hybrid cyborg work redesign. The reliability and validity of the data that train such systems may be suspect (i.e., prone to bias and hallucinations) if the amounts and variety of data are insufficient. Indeed, non-routine manual work is also the realm of artisanal and gig work. One would expect the supply of this kind of work to increase because of labor needs for side hustles in conjunction with the slow move to automation in those market segments that are not prone to economies of scale.
Cognitive Routine Work
Cognitive routine work is more vulnerable to the LLMs, given that they are mainly based on some form of language in electronic media (text, audio, or visual). Data from such domains can be easily captured and used to train general LLMs and domain-specific LLMs. The transition employing these systems will only be slowed by the inertia of organizations (lack of leadership, lack of understanding about digital transformation, inadequate funds, poor management, unions, and lack of competitors forcing changes). So, the pace and scope of this change are unclear, given such inertia. However, large-scale players in retail (Walmart, Target, Uniqlo, and Amazon) have enormous incentives to make changes using automation and LLMs, as well as the resources, further adding to their competitive advantage.
Cognitive Non-Routine Work
Regarding “creative work,” or cognitive non-routine work, there is still room for humans to locate the new, unusual, and unobvious in seeking solutions to jobs to be done. But as prompt engineering improves and translating data into a form usable by LLMs also improves, generative AI will become more and more creative, relevant, and likely predictive. It will also provide excellent advice on evaluating creative alternatives if offered a clear rubric for considering desirability, feasibility, and sustainability in innovation.
The next five years will tell us more about the scope and pace of the changes AI will bring through robotics and generative AI.
Published at Wallethub
Of interest from Business West
Comments