Lhe launch of ChatGPT in November 2022 marked a turning point in the history of artificial intelligence. In companies, generative artificial intelligence (GAI) has surprised people with its ability to carry out tasks that were the prerogative of the most qualified (and best paid) employees. Since then, many other IAG software programs have appeared. What will be their impact on the world of work, and in particular on the most qualified employees?
Less than a year after the launch of ChatGPT, a study provides the first elements of an answer to this question (“Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality »Fabrizio Dell’Acqua et al.Harvard Business School Technology & Operations Management Unit, Working Paper No. 24-013).
The study was carried out by a multidisciplinary team of researchers in collaboration with an international consulting firm.
More than 750 consultants took part in the game. They were divided into two groups. The first group used ChatGPT-4 to work on a business case. The second group did not use it. Most of the tasks that the researchers entrusted to them were within the reach of artificial intelligence (for example: making ten proposals for new products or writing a press release for the launch of one of these products). On the other hand, one of them was beyond the reach of artificial intelligence (making strategic recommendations based on both interview transcripts and encrypted data).
The three main lessons of the study
1) Generally, the use of generative artificial intelligence improves performance. When the tasks to be performed were within the reach of artificial intelligence, consultants who used ChatGPT-4 worked faster and better than their colleagues who did not use it. More specifically, they completed more tasks (+12.2%) and completed these tasks more quickly (+25.1%).
The quality of their achievements was also higher (+ 40%). Furthermore, the increase in performance was identical for consultants who were trained in the use of ChatGPT-4 and for those who used a more intuitive approach. This result confirms that generative artificial intelligence is very easy to access.
2) In some cases, the use of generative artificial intelligence deteriorates performance. As could be expected, the consultants who used ChatGPT-4 performed much less well than those who did not use it to carry out the task which was not within the reach of artificial intelligence (− 20 percentage points). The main limitation of generative artificial intelligence is that the quality of the results it provides varies greatly from one task to another.
You have 50% of this article left to read. The rest is reserved for subscribers.