Artificial intelligence (AI) systems have gained prominence this year ever since the breakout success of OpenAI's ChatGPT in 2022. Despite their widespread use, AI tools were found to excel in imitation but lag in innovation, according to research conducted by the University of California, Berkeley.
Tech Xplore reported that the study found that while AI systems, including language models like ChatGPT, demonstrate proficiency in summarizing and reproducing existing knowledge, they struggle with the creative problem-solving abilities inherent in human innovation.
AI vs. Children
The study compared the capacity of AI systems to imitate and innovate with that of children and adults. Eunice Yiu, co-author of the research, emphasized that AI should be viewed as a form of "cultural technology," similar to a library or search engine, capable of summarizing and conveying existing cultural knowledge.
However, AI systems lack the imaginative and problem-solving capabilities that humans, even young children, exhibit. The experiment involved presenting children aged 3 to 7 and adults with descriptions of everyday objects.
Participants were asked to identify objects that would "go best" together and, in the subsequent stage, to innovate new uses for familiar objects. The results indicated that while both children and adults excelled in innovation tasks, AI models struggled to generate novel responses.
In the innovation task, wherein participants were asked to draw a circle without conventional tools like a compass, most humans selected conceptually dissimilar tools, such as a teapot with a round bottom.
In contrast, AI models were less accurate in choosing effective tools for the same task. The study's findings suggest that AI's reliance on statistical prediction of linguistic patterns is insufficient for discovering new information and innovative problem-solving.
AI Systems Fall Short in Innovation
While AI systems excel in transmitting known information, they fall short in expanding, creating, changing, evaluating, and improving upon conventional wisdom, a domain where humans, especially young children, excel.
"Even young human children can produce intelligent responses to certain questions that [language learning models] cannot," Yiu said in a statement.
"Instead of viewing these AI systems as intelligent agents like ourselves, we can think of them as a new form of library or search engine. They effectively summarize and communicate the existing culture and knowledge base to us," she added.
The study concluded that, although AI development is in its early stages, there is much to learn about enhancing its learning capacity.
The researchers noted that drawing inspiration from children's intrinsic motivation and curiosity in learning could guide researchers in designing AI systems better equipped to explore and understand the world. The research team's findings were published in the journal Perspectives on Psychological Science.
Related Article : "AI Hasn't Come So Far," says Laureate, Echoing World's Greatest Scientists During Laureate Forum
