WEF Jobs of Tomorrow Generative AI 2023 · Page 6

By integrating LLMs with other systems, these 
capabilities can be extended to a greater range of 
abstract tasks, such as scheduling meetings, placing 
orders, responding to emails, or providing research 
on a particular topic.
Given the large overlap between LLM capabilities 
and current job tasks, how will introducing LLMs 
into the workplace change jobs? Which parts of a 
job will be impacted the most, and which jobs will 
be impacted most? Finally, with the introduction of 
these new technologies, which new jobs can be 
expected to arise? 
A task-based approach  
to job exposure
To answer these questions, the methods deployed 
in this white paper assess the potential exposure of 
language-based job tasks to the ability of LLMs to 
perform these tasks. The approach is to first think 
of a job as consisting of many different tasks and 
then assess how each task may be affected by 
LLMs. The magnitude of impact on a job ultimately 
depends on the degree of language-based skills 
required for specific tasks in that job and the time 
spent on those tasks. Language-dependent, 
standardized, routine and process-oriented tasks  
are prime candidates for automation and 
replacement by LLMs. At the same time, those 
requiring a greater degree of human interaction  
are more likely to be augmented and performed  
in collaboration with LLMs.
For example, some job tasks are routine and 
predictable and are performed by people working 
individually, such as clerks and administrators, 
which involves reading and entering data, cross-
referencing records between different databases and 
reviewing transactions. These tasks are more likely 
to be exposed to and ultimately automated by the 
introduction of LLMs, implying that they will no longer 
be performed by humans. The outcome is that jobs 
emphasizing these tasks will either transform to take 
on non-automatable tasks or go into decline. Other 
job tasks require a great deal of abstract reasoning, 
creativity and problem-solving. While language 
tasks may not be their primary product, they may 
rely heavily on language and communication. For 
example, Mathematicians and Editors rely heavily 
upon language, yet need to incorporate creative 
insights from their fields of expertise. Similarly, 
Software Developers work a lot with computer 
languages but also need to grasp complex systems 
at various levels of abstraction to create a finished 
software product. Workers in these jobs would not 
have their tasks replaced by LLMs; rather, LLMs 
would supercharge their ability to complete these 
tasks. Teachers, for example, could rely on LLMs for 
assistance in lesson planning and correcting student 
work. According to one study in the US, three in 
ten teachers have already used ChatGPT for lesson 
planning (30%), generating creative ideas for classes 
(30%) and building background knowledge for 
lessons and classes (27%).10 Software Developers 
could turn to LLMs to generate standardized blocks 
of code with clear functional parameters, speeding 
up the development process and allowing for more 
time to be spent on high-level architectural tasks. 
Software Developers also perform many tasks with 
high potential for automation, suggesting that many 
jobs will be transformed rather than automated  
or augmented.
The research methods employed in this paper aim 
to identify which tasks will be exposed to LLMs and 
how they will be impacted: whether they have the 
potential to be automated and replaced b