Cookies Policy
The website need some cookies and similar means to function. If you permit us, we will use those means to collect data on your visits for aggregated statistics to improve our service. Find out More
Accept Reject
  • Menu

INESC TEC researcher warns companies about the quality of data generated through AI in a paper published by MIT management journal

Could the increasing interest in language models, like ChatGPT, be diverting resources away from companies to adopt advanced analytics practices that truly support smart decisions? Pedro Amorim, INESC TEC researcher, and João Alves (from INESC TEC LTPLabs spin-off) believe so. In a paper published in MIT Sloan Management Review, they warn about the quality and unpredictability of data generated solely from generative language models - despite advocating for more investment in Artificial Intelligence (AI) that incorporates these models with advanced analysis (with concrete reasons provided).

25th June 2024

We have reached a crossroads where generative language models bring increased concerns, even though we recognise the importance of this and other AI tools. But isn't ChatGPT an AI tool? Yes, it is. It is a generative language model, or a large language model (LLMs). And aren't advanced analytics practices also an AI tool? They are. So, what do Pedro Amorim and João Alves advocate? In the paper published in the summer Special Report of MIT Sloan Management Review - a renowned independent journal based on research and digital platforms for business leaders -, the researchers explain that LLMs and advanced analysis feature different resources that support companies; in addition, and contrary to common practice, business leaders should not choose one over the other, but rather seek the integration of generative language models with advanced analysis practices. The paper is available here.

"When used in a complementary way, LLMs can help companies developing and implementing advanced analysis practices, both for forecasting and prescription processes. In fact, generative language models can be especially useful to help companies incorporate data that is not structured to support them in their analyses, translate business problems into analytical models, and help understand and explain the results of the models; it's worth mentioning that, when used exclusively, they can be unpredictable and present data quality issues", explained Pedro Amorim, researcher at INESC TEC.

In their research, the authors performed several tests with generative language models to increase the potential of advanced analysis practices. More specifically: transforming unstructured data into structured data; helping to explain complex outcomes; incorporating complex data; improving communication of outcomes; and facilitating cross-functional collaboration across teams.

"In the first process, our research shows that the use of generative language models to assist in this task can sometimes save the equivalent of weeks of work. LLMs can analyse data and, by identifying key themes, structure them, later providing predictive models that advanced analytics teams can explore. These teams must be multidisciplinary, with data scientists and people responsible for the business areas, who make decisions at companies", explained Pedro Amorim.

In fact, the research carried out by Pedro Amorim and João Alves shows that the business teams themselves can benefit from data generated through the generative language models, since they allow them to participate more actively in the analytical process, reduce barriers to adoption, facilitate change management and promote confidence in the data that is generated.

Concerning advanced analytics and AI, and before this paper, Pedro Amorim had published (together with other authors, all of them founders of the INESC TEC LTPLabs spin-off) - the book "The Analytics Sandwich", which proposed a shift from traditional decision-making (based on intuition) and an investment in advanced analytics and AI.

The INESC TEC researcher mentioned in this news piece is associated with UP-FEUP.