AI models overestimate smartness of people: Study
Artificial intelligence (AI) has made tremendous progress in recent years, with models like ChatGPT and Claude demonstrating impressive capabilities in understanding and generating human-like text. However, a new study by scientists at HSE University has found that these models may be overestimating the smartness of people. Specifically, the study discovered that current AI models tend to play “too smart” when engaging in strategic thinking games, assuming a higher level of logic in humans than is actually present. This can lead to suboptimal outcomes, as the models fail to account for the limitations and biases of human decision-making.
The study focused on the Keynesian beauty contest, a classic game theory puzzle that requires players to think strategically and make decisions based on their expectations of others’ behavior. In the game, players are asked to choose a number between 0 and 100, with the goal of selecting a number that is closest to two-thirds of the average number chosen by all players. The twist is that players must make their decision without knowing what numbers the other players have chosen.
The researchers used this game to test the performance of several AI models, including ChatGPT and Claude, against human players. They found that the AI models consistently played “too smart,” choosing numbers that were closer to the theoretical optimum than the numbers chosen by human players. However, this approach ultimately led to the AI models losing the game, as they failed to account for the fact that human players do not always make rational or optimal decisions.
The study’s findings have important implications for the development of AI models that interact with humans. If AI models overestimate human smartness, they may fail to account for the limitations and biases of human decision-making, leading to suboptimal outcomes. For example, in a business setting, an AI model that assumes humans will always make rational decisions may fail to anticipate and respond to irrational or emotional behavior.
The researchers suggest that AI models need to be designed to take into account the cognitive biases and limitations of human decision-making. This may involve incorporating more realistic models of human behavior into the AI’s decision-making process, or using machine learning algorithms that can adapt to the specific characteristics of human players.
The study’s findings also highlight the importance of testing AI models in real-world scenarios, rather than relying solely on theoretical simulations. By pitting AI models against human players in games like the Keynesian beauty contest, researchers can gain a better understanding of how AI models perform in practice, and identify areas where they need to be improved.
Overall, the study’s results suggest that AI models still have a long way to go in terms of understanding human behavior and decision-making. While AI models like ChatGPT and Claude are incredibly sophisticated, they are not yet able to fully capture the complexities and nuances of human thought and behavior. By recognizing the limitations of current AI models and working to develop more realistic models of human behavior, researchers can create AI systems that are more effective and more capable of interacting with humans in a meaningful way.
The study’s findings are a reminder that AI models are only as good as the data and assumptions that they are based on. If AI models are trained on data that assumes humans are more rational or intelligent than they actually are, they will likely perform poorly in real-world scenarios. By acknowledging the limitations and biases of human decision-making, researchers can create AI models that are more realistic, more effective, and more capable of achieving their intended goals.
In conclusion, the study by scientists at HSE University highlights the importance of developing AI models that take into account the cognitive biases and limitations of human decision-making. By recognizing that humans are not always rational or optimal in their decision-making, AI models can be designed to be more effective and more capable of interacting with humans in a meaningful way. As AI continues to evolve and become more integrated into our daily lives, it is crucial that we prioritize the development of AI models that are realistic, effective, and capable of achieving their intended goals.
Source:
https://www.sciencedirect.com/science/article/abs/pii/S0167268125004470
About the Author
