Benjamin Lindquist, an assistant professor at MIT, argues that modern AI hallucinations stem from mid-20th century efforts to make machines mimic human creativity. In a new paper, he explains that early engineers intentionally injected randomness into algorithms to replicate the unexpected leaps of logic found in human thought. Lindquist notes that while these systems were originally valued for their ability to produce strange and novel answers, contemporary models have been refined to suppress such unpredictability. He suggests that there is an inherent tradeoff between achieving creative outputs and maintaining predictability in artificial intelligence systems.
Read the full story at news.mit.edu.
Photo: 2427999 / Pixabay
