Future of AI is here but not evenly distributed - some real SURPRISES on use, gender, age, expectations & ethics

A few surprises in this excellent paper on use of ChatGPT on ‘The Adoption of ChatGPT’ by Anders Humlum from the University of Chicago (July 9, 2024).

It was a large-scale survey (Nov 2023 - Jan 2024) of 100,000 workers from 11 exposed occupations, linked to labour market histories, earnings, wealth, education and demographics to characterise the nature of adoption of ChatGPT. Note that this was before release of certain improved AI models such as ChatGPT4o and Claude 3.

The 11 occupations, a good selection, were; 

HR professionals, Teachers, Office clerks, IT support, Software developers, Journalists, Legal professionals, Marketing professionals, Accountants, Customer Service and Financial advisors.

Use

• almost everyone is aware of it 

• 50% had used technology

• adoption rates from 79% for software developers to 34% for financial advisors

• differed in their intensity of use

As expected and the authors confirm the “widespread adoption of ChatGPT, only a year after its first launch, solidifies it as a landmark event in technology history.”

Age a factor

• younger & less experienced workers more likely to use ChatGPT

• every year of age has 1.0 percentage point lower likelihood of using ChatGPT 

• similarly with every year of experience at 0.7 lower

Gender gap

Women less likely to use tool:

• 20 percentage points less likely to have used the tool

• pervasive in all occupations

• in various adoption measures

• persists when comparing within same workplace

• also controlling for workers’ task mixes

Expectations

Employees perceive:

• substantial productivity potential in ChatGPT

• average estimates that ChatGPT can halve working times in a third of job tasks

• smaller rather than larger time savings for workers with greater expertise

• 38% say they will not perform more of the tasks ChatGPT saves time completing

Interesting experiment

In a clever experiment, inserting exposure to expert assessments of the time savings from ChatGPT in their job tasks, they found, “despite understanding the potential time savings from ChatGPT, treated workers are not more likely to use ChatGPT in the following two weeks”. This is rather worrying for those who see training, frameworks and documents as the solution to increasing use.

Finally, and this really matters, they investigated what prevents workers from transferring the potential productivity gains from ChatGPT into actual adoption. 

Workers reported barriers:

• restrictions on use

• needing training as the primary barriers to adoption

• need for firm policies (guidelines for use or facilitating employee training) 

• few reported existential fears, becoming dependent on technology or redundant in their jobs, as reasons for not using ChatGPT

This last point is important. Employees are not too concerned with ethical issues and do not seem to fear dependency or redundancy.

Conclusion

There's a ton of other useful stuff and detail in this paper but the Surprises that stood out for me were the lower uses by women, strong barriers to adoption by employers, despite workers agreement htat is will increase productivity and indifference to ethical concerns.

Bibliography

Humlum, A. and Vestergaard, E., 2024. The Adoption of ChatGPT. University of Chicago, Becker Friedman Institute for Economics Working Paper, (2024-50).