Introduction
Artificial intelligence (AI) has become a hot topic, and its potential to boost productivity has captured the attention of economists, computer scientists, and managers. A groundbreaking study conducted by economists from Stanford and MIT sheds light on the impact of generative AI (GAI) on worker productivity.1 In this article, I explore the study’s primary findings and the implications for profits, wages, and the role of intellectual property.
The study focused on the use of GAI chatbots to support customer service representatives in their interactions with customers. The researchers introduced chatbots in staggered waves to over 5,000 workers in a randomized control trial.2 The researchers measured various metrics—issues resolved per hour, customer sentiment, escalation rates, and employee retention—to gauge productivity gains.
Productivity gains
The findings were remarkable. On average, productivity increased by 14% with the introduction of GAI. However, the study revealed an intriguing pattern—the productivity gains were predominantly observed among inexperienced and lower-skilled workers. Skilled and experienced workers saw minimal impacts, suggesting that GAI acted as an equalizer.
One of the reasons for this productivity disparity lies in the “tacit” knowledge flow facilitated by GAI. The AI was trained using successful calls, representing an indirect transfer of knowledge from more experienced workers to their inexperienced counterparts. Consequently, customer satisfaction improved, and inexperienced workers became more productive.
Who Profits?
However, these results raise further questions that are outside the scope of the study, but that are paramount for understanding the impact of AI on society. Namely, who ultimately profits from these productivity gains? In the long run, will inexperienced workers receive increased wages based on their increased productivity? Or will management opt to hire more inexperienced workers and to use GAI to bridge the skills gap, displacing more experienced workers? If the real source of the productivity increase is the GAI tool itself, the latter seems more probable.3
How will the returns to the GAI tool be divided? Will management retain the additional profits? Or will the gains from increased productivity accrue to the creator of the GAI algorithms? The key to understanding who profits lies in intellectual property (IP) rights and the distinction between the algorithm and the training data.
AI Assets and IP
GAI tools represent a new class of intangible assets, perhaps AI assets or GAI assets.4 While the creator of the GAI algorithms may be able to assert some ownership over the downstream tools, the value creation for specialized implementations relies heavily on the training data. In the context of the study, those training data were generated by (experienced) employees under the direction of management. Therefore, the question of who owns the GAI Asset depends on the relative contributions of algorithms, data creators, and data owners.
Conclusion
Considering the study’s findings, companies need to recognize the value of their data and formulate robust IP policies to protect their trade secrets, particularly the training data used to create the AI algorithms. Thus, agreements between users, tool developers, and algorithm providers should reflect the value of each component of the tool. Further, intellectual property considerations will shape employee contracts and compensation structures, determining how contributions to productivity gains are rewarded. Businesses must carefully navigate the delicate balance between utilizing AI for increased efficiency and preserving the well-being and stability of their workforce.
1 The study is a working paper and has not yet been peer-reviewed.
2 Because of the experimental design the results can be interpreted as causal rather than simply correlational.
3 It is worth noting that because experienced workers play a critical role in accumulating training data, there exists a tension between ongoing data quality and the gains to using inexperienced workers once the tool is developed.
4 I would call them “intelligence assets,” but for fear of confusion with something out of a Jason Bourne movie.