AI’s impact on human employment will vary in line with what tasks are open to automation

Venture capitalist Vinod Khosla set the cat among the pigeons by saying that AI may destroy the software services and business processing outsourcing businesses that are so important to the Indian economy. Elsewhere, shares of major software service companies lost ground as investors sold them after the launch of Claude Cowork by Anthropic fanned fears that several industries would be disrupted.

Will the machine render the worker obsolete? This question is as old as the history of industrial civilization. This particular species of dread has quite understandably re-emerged with the rise of AI, a revolutionary technology by any measure.

The debate is mostly framed in terms of jobs. The temptation to do so is understandable—a robot takes a job, a software programme eliminates a job and a large language model destroys a job. This makes intuitive sense. However, economists such as David Autor and Daron Acemoglu have over the past decade provided a more nuanced way to think through this very important question.

The two economists from the Massachusetts Institute of Technology have tried to shift the focus of discussion away from jobs to tasks. (‘Skills, Tasks and Technologies: Implications For Employment and Earnings,’ by Daron Acemoglu and David Autor, 2011).

Their core point is that jobs are nothing but bundles of tasks. Humans use skills to perform tasks that constitute their jobs. A lawyer drafts contracts, counsels clients, researches precedent, appears in court and manages relationships. An accountant reconciles ledgers, interprets tax law, advises on financial strategy and communicates findings to non-specialists. A doctor reviews the results of medical tests, understands medical histories, diagnoses the problem and treats it.

Technology rarely eliminates all of such tasks simultaneously. What it does, with great efficiency, is target specific tasks within a job—typically those that are routine, codifiable and repetitive—while leaving others relatively untouched.

This distinction between the job and the task is not merely semantic. It shapes how we should evaluate the disruption that AI is now bringing to office work, professional services and creative industries. When the unit of analysis is a task rather than a job, it is worth asking whether AI taking over some tasks will or will not increase efficiency in the tasks that will continue to be done by human professionals. Much of the currently inconclusive debate about whether AI will make humans redundant or actually increase labour productivity hinges on the issue of tasks.

There is also a distributional nuance. The standard trope is that technology helps skilled workers but hurts unskilled workers. But Acemoglu and Autor argue that what actually matters is which jobs are dominated by routine tasks that can be automated.

It is quite likely that workers with sophisticated skills as well as those with more basic skills will be able to adapt to a new technology such as AI, but those in the middle, with a high proportion of routine tasks, will be hit hard. These displaced workers then try to get work one level below where they were. The increased competition for work at the lower end of the labour market may freeze wage growth. The result is likely to be growing income inequality.

“The distinction between ‘labour tasks’ and ‘capital tasks’ in production is permeable and shifting,” Autor wrote in another essay in 2013. There are some tasks that can be easily replaced by new technology while there are others which can be complemented by the same technology.

“This evolving division of labour has a clear economic logic: novel tasks—those demanded by new products, techniques, or services—are often assigned first to workers because workers are flexible and adaptive. As these tasks are formalised and codified, they become fallow for automation since machinery typically has a cost advantage over human labour in rote execution of repetitive tasks”.

The demand for labour will thus be sensitive to the nature of tasks that combine to define a job. Some will be replaced by new technology such as AI. Others will be able to use the technology to increase their productivity—displacement versus augmentation.

What the balance between these two categories looks like in the age of AI is still unclear. A lot also depends on how companies eventually use AI in their operations. In other words, the technology becomes endogenous to business decisions as use cases multiply. Policy choices will also have a big role to play.

Anyone who has watched Billy Wilder’s 1960 film The Apartment will recall the opening sequence: a vast, open-plan floor at a Manhattan insurance company, populated by hundreds of clerks as far as the eye can see. It is a monument to a particular kind of business organisation.

In adjacent rooms were typist pools—rows of women whose professional purpose was to transform handwritten notes and dictated memos into clean, typed documents.

Those clerks and typists are gone. New types of work emerged. The question is whether something similar will happen in the case of artificial intelligence as well.

The author is executive director at Artha India Research Advisors.