Federico Fubini (Corriere della Sera)
Actualizado
In today’s digital landscape, the question of whether artificial intelligence (AI) is making us less intelligent arises frequently. A new preliminary study from the Media Lab at the Massachusetts Institute of Technology (MIT) plays a significant role in this discussion. The study, titled “Your Brain and ChatGPT: Cognitive Debt Accumulation When Using an AI Writing Assistant,” indicates troubling findings regarding our cognitive capabilities when relying on AI for tasks.
Led by Natalia Kosmyna, a doctoral researcher specializing in computer science and human-computer interaction, the research involved testing 54 volunteers divided into three groups. Each group was assigned to write three short texts on predefined topics during consecutive sessions. The first group had no access to technology whatsoever, relying solely on their mental resources. The second group utilized Google, while the third group leveraged AI, specifically ChatGPT . Participants were monitored through electroencephalography (EEG), allowing researchers to analyze brain activity during the writing tasks.
Diverging Brain Activations
The results of this study revealed shocking variances in brain activation among the groups. Participants who wrote without any digital assistance exhibited a significantly different level of cognitive engagement compared to those relying on technology. The group using Google showed brain connectivity reduced by 34% to 48%, while those using ChatGPT experienced a staggering 55% reduction. In essence, the more reliable the assistance, the lower the level of cognitive engagement .
One of the most illuminating insights from the study is the impact of writing medium on the quality of the produced work. The group without technology activated brain areas associated with creative ideation and self-control , indicating a deeper engagement with their content. In contrast, the participants using Google mainly activated visual processing areas, focusing on quickly assimilating information. Those employing ChatGPT primarily tapped into regions associated with automatic functions , illustrating a dependency on external scaffolding. The creative output differed markedly: the text from the group relying solely on mental resources was diverse and heterogeneous, while the AI-supported group produced texts that were remarkably similar and homogeneous.
Cognitive Ownership and Textual Understanding
Perhaps the most concerning finding was that 83% of participants using ChatGPT struggled to recall phrases from their own texts just moments after completion. This suggests that they had not developed a sense of ownership over their work. In stark contrast, nearly all members of the group working independently were able to recall almost verbatim accounts of what they had written. This discrepancy emphasizes that reliance on AI can lead to diminished attention towards both the content and meaning of one’s work.
The evaluation of the produced texts yielded further insights. AI evaluators gave higher scores to texts written with AI as opposed to those produced by human effort. However, human evaluators—teachers—immediately recognized the AI-generated texts and rated them lower, reaffirming the lack of depth in the AI-produced material. This disparity raises questions about the validity of relying on AI for generating quality content.
Implications for Learning and Society
The implication from this study is profound. When the AI-using group was tasked to create a text without missing digital support, their performance was notably hindered. Conversely, those accustomed to independent writing thrived when utilizing ChatGPT, revealing that the adaptability of thought is compromised when routines rely too heavily on AI.
The researchers concluded that individuals reproducing AI-suggested content without assessing its accuracy forgo claiming the ideas expressed. This not only risks internalizing shallow insights but also raises concerns about manipulability in larger societal contexts, especially within political realms and inferential thinking.
As society increasingly leans on AI tools, the consequences stretch far beyond individual cognition. For democracy, education, and independent intellectual growth, the challenge becomes clear: While artificial intelligence can undoubtedly aid in various tasks, training our natural intelligence remains paramount. It is a delicate balance where the responsibility rests on individuals and educational institutions to foster judgment and creativity independent of digital shortcuts.