Daniel Kahneman outlines a concept which helps explains how these models work.9 He describes two types of thinking:
System 1: Fast, Instinctive, Emotional
System 2: Slow, Logical, Calculated
Today’s generative AI models excel at System 1 thinking. They can quickly generate responses based on patterns they’ve learnt, much like how you might instinctively react to a sudden noise or complete a familiar phrase. For this reason, they can produce human-like text or images almost instantly.
The fascinating part is, as these models evolve, they’re beginning to demonstrate capabilities that resemble System 2 thinking — something that used to be unique to non-generative AI such as chess bots or spam filters.