started · updated
AI hallucinations linked to historical use of randomness
Research into artificial intelligence hallucinations highlights a historical connection to the deliberate introduction of randomness in early AI systems. According to historian Benjamin Lindquist, mid-20th-century researchers incorporated irregularity into models to mimic creative thought and prevent machines from being overly predictable.
While this randomness was intended to allow machines to deviate from scripts and offer unexpected ideas, it contributes to the modern issue where AI provides convincing but incorrect answers when it lacks factual data. Lindquist suggests that while reducing unexpected responses might mitigate hallucinations, it could also come at a cost to the creative potential of the technology. The challenge remains in distinguishing between the generation of novel possibilities and the accurate evaluation of which ones are factual.
Entities
Adam Tauman Kalai · Benjamin Lindquist · Jacques Monod · MIT · OpenAI
Claims
What the coverage asserts, and how many sources carry each claim.
- [● 3 SOURCES] Early AI researchers deliberately introduced randomness to encourage creative, non-scripted responses. www.informacion.es · www.elperiodicoextremadura.com · www.laopinioncoruna.es
- [● 3 SOURCES] AI models sometimes provide convincing but incorrect answers when they lack information. www.informacion.es · www.elperiodicoextremadura.com · www.laopinioncoruna.es
- [● 3 SOURCES] Reducing unexpected AI responses may decrease the technology's creativity. www.informacion.es · www.elperiodicoextremadura.com · www.laopinioncoruna.es