Studying Generative AI’s Impact on Work
Bhupender singh karki
Motion Graphics Specialist | Expert in Dynamic Animations & Visual Effects
The potential that artificial intelligence has to disrupt jobs and markets has been making headlines for years, and the advent of generative AI in late 2022 spurred renewed discourse. However, its actual impact on the labor market was—and remains—relatively unknown given the speed with which it came to market, continuing technological advancements, and its perceived ability to replace tasks commonly believed to be human in nature.
At Upwork, we have a unique advantage, in that our platform is a self-contained labor market for independent talent. From our marketplace, we can derive data and insights from the full range of hiring and work behaviors: project posts, matching, job completion, and feedback, as well as rates, earnings, and payments.
in partnership with our Analytics team, decided to look at our own platform data over the course of 2023. By setting up a study that isolates the causal impact of generative AI on freelancers’ work opportunities and earnings, we got an early read on how generative AI is transforming work. Our end-to-end visibility into data on that full range of hiring and work behaviors leads us to expect that many of the generative AI-influenced trends we’re seeing transpire in our work marketplace will be reflected in the broader labor market. Our findings illustrate a dynamic, complex, and multidimensional story about generative AI’s impact on work. You can explore our full findings and methodology in our whitepaper, and highlights below.
Generative AI’s overarching effects on the Upwork marketplace
A look at previous technological innovations, ranging from robotics to mobile devices, suggests that AI will have a dual, contrasting impact on work.
First is the much-discussed replacement effect, whereby some work is impacted and ultimately disrupted as some tasks are able to be automated. But just as important is the less-discussed reinstatement effect, whereby newly created work opportunities increase earnings over time because emergent technology creates new tasks in which labor has a comparative advantage.
Generally, the replacement effect is more noticeable in the short term, but as new jobs and tasks are created over the mid to long term, the reinstatement effect will prevail. The effects of new technologies often counterbalance each other over time, giving way to many new skills and opportunities that previously weren’t possible or simply didn’t exist.
Historically, technology creates more work opportunities than it replaces as society evolves, innovates, and discovers new ways of working. These new opportunities tend to focus on more complex tasks and command higher rates, benefiting both skilled workers and the organizations they support.
Whereas it can be years before these forces take full effect in the broader labor market, both forces are already at play in Upwork’s work marketplace.
When you combine the initial impact of the replacement and reinstatement effects on the Upwork marketplace, generative AI has increased both the total number of job posts and freelancer earnings per new contract created. This indicates an overall greater demand for work completed by independent talent from clients, not less, as well as an overall higher value of contracts. And, as we will discuss, these new engagements tend to favor knowledge-based skilled work, offering more task complexity and variety to the workforce.
Replacement or reinstatement? Not all categories of work are impacted equally
We also need to dig a layer deeper and understand how AI is impacting certain categories of work. We know from previous technological disruptions that not all categories of work benefit equally. Within our data platform, we track 12 distinct categories of work, as represented in Figure 1. All projects that occur on our platform roll up into one of these categories. Demand for AI and machine learning work is growing across all 12 categories, especially for high-value tasks.
Specifically, the reinstatement effect of generative AI appears to be driving growth in categories related to technology solutions and business operations. Work that requires generative AI and machine learning skills is up significantly within these categories.
In particular, one of our largest categories of work, Data Science & Analytics, is reaping the benefits, with over 8% growth in freelancer earnings directly attributed to the introduction of generative AI work in this category. (It should be noted that this increase in earnings is spread across all work being done within this category, including work not directly related to AI and machine learning.) Research that we released in 2023 showed a reinstatement effect already taking place on our platform, with new tech roles and skills like AI content creators, prompt engineers, and large language model (LLM) development experts emerging.
These projects and skills tend to contribute to generative AI’s propulsion of growth in high-value contracts; in fact, we also found that independent talent engaged in AI-related work earn 40% more than their counterparts engaged in non-AI-related work. While tools such as ChatGPT automate certain scripting or coding tasks present in these categories (an example of replacement), that is outweighed by productivity gains for independent talent using such tools to their advantage, often leading to higher earnings.
Turning to business operations work—contrary to predictions about back office jobs being automated away en masse due to AI—we found Accounting & Consulting, Administrative Support, and Legal Services all experiencing earnings increases for their categories as a whole due to generative AI, ranging from 6% to 7% growth.
Many of these engagements require intensive interpersonal communication, and generative AI seems to have helped increase their value. Generative AI frees up time previously used for routine tasks so that these professionals have more space to understand and address client needs, thus amplifying the value of those with strong people and communications skills.
领英推荐
As expected, we also see categories where client demand and earnings are decreasing, although replacement and reinstatement effects aren’t always obvious at the work category level. Our research found that generative AI is creating a delta between “low-value” and “high-value” contracts within certain categories.
One fascinating category to look at is Writing & Translation, a small category of work on our platform but still a seemingly obvious target for replacement. While we have seen overall reduced demand and earnings, high-value contracts in this category have actually experienced higher earnings as a result of generative AI.
In high-value Translation engagements, for example, we’ve seen earnings increase by 7%. An example of a “high-value” engagement is a multilingual AI trainer, while a “low-value” engagement could be translation for a webpage.
Another vital distinction: Although lower-value, rudimentary Writing & Translation contracts are being automated and replaced, the nature of these types of tasks has been predicted to be automated for some time. While generative AI might have accelerated the effect, the future of low-value Writing & Translation contracts, as well as any repetitive low-value task, was likely to be disrupted. On the other side of the coin, it’s highly encouraging to see that generative AI has such a positive impact on higher-value Writing & Translation engagements.
Furthermore, we’ve observed that the introduction of generative AI is increasing earnings in high-value contracts across all categories. This may partially be because of project complexity, which places a premium on tasks that are done by humans higher up the value chain, which tend to be augmented rather than replaced by technology.
In other words, technology like generative AI can make routine low-value work go faster, freeing up people to work on problem-solving, strategy, and other complex tasks. These higher-value tasks are often paid a premium. Consider, for example, how it is much more difficult (and thus higher value) to formulate a breakthrough integrated marketing campaign than it is to create automated email templates.
Actionable insights from the impact of generative AI
Our study across all categories of work on our platform led us to a few critical conclusions:
Every technological sea change brings with it fear, uncertainty, and doubt, and the meteoric rise of AI is no exception. The dynamics at play, especially in a fluid work marketplace like Upwork, are as plentiful as they are complex. Taking a balanced, objective look at the early data shows that, while there are imperative steps to take in preparing, upskilling, and reskilling the workforce of tomorrow, the reigning narrative about AI stealing everyone’s jobs is vastly overblown.
At Upwork, not only have we not yet observed this to be true, but our research also shows a multitude of positive effects and opportunities ahead stemming from the rise of this new technology. We remain resolute in our conviction that the future of work will be shaped by AI that puts humans at the center, unlocking innovative new ways for talent and clients to get work done more easily and effectively than ever before.
Methodology
To understand the impact of generative AI on the Upwork labor market, we analyzed data collected on our platform from 2021 through Q3 2023. We specifically looked at freelancer earnings per new contract across all 12 work categories on our platform for high-value contracts, defined as contracts of at least $1,000, and low-value contracts, comprised of contracts between $251 and $500. To isolate generative AI’s impact on freelancer earnings, we used a synthetic control approach combined with Lasso regularization. This method creates a counterfactual scenario to simulate what would occur without AI’s influence; in other words, a control group, which is ideal for studies where controlled experiments aren’t feasible. To help better construct the counterfactual groups, we also scored which job categories are exposed to the effect of generative AI, based on AI-related job posts and occupational exposure scores by Felten, Raj, and Seamans (2023). Our approach allowed us to assess AI’s causal effects on the labor market and specific work categories, while also considering external economic factors, such as the effects of seasonal hiring or economic downturns.