The study, conducted by Anna Maklova, a research assistant at Humboldt University and an expert in psycholinguistics, delved into the concept of “theory of mind,” which examines understanding other people’s mental states.

Details of the study

In collaboration with Charles University in Prague, the research team focused on evaluating whether large language models (LLMs), such as OpenAI’s GPT-4, could simulate the stages of cognitive development of children between the ages of one and six.

When models were asked to respond like children in a battery of more than 1,000 trials and cognitive tests demonstrated a remarkable ability to mimic lower levels of intelligence similar to those of children in the specified age range.

Maklova emphasized that although comparing artificial intelligence to humans provides a simple understanding, it can also be misleading. Instead, this study argues for a paradigm shift in assessing AI’s ability to generate the childlike images observed in experiments.

The author calls for caution and not to underestimate the capabilities of artificial intelligence, and also emphasizes the importance of realizing its potential to effectively mimic human behavior. Such a shift in perspective could pave the way for safer development of AI technologies, avoiding the pitfalls of false expectations and ensuring responsible development.