Q&A: How AI came to model the “messy” human mind
Assistant Professor Benjamin Lindquist studies how engineers deliberately injected randomness into early AI, giving rise to deep-learning models that sometimes hallucinate.
Early computer engineers believed injecting randomness and irrationality into AI systems designed for logic and calculation was essential for mimicking human thought, contends MIT professor Benjamin Lindquist. In a new paper, Lindquist traces the origins of AI hallucinations back to early research that intentionally added randomness to deterministic algorithms. He suggests there will be tradeoffs between creativity and predictability when building machines that mimic human behavior.
MIT News spoke with Lindquist about these ideas. Lindquist explains that early computer engineers theorized that chance plays an important part in discovery and creativity, equating creativity with unexpected leaps of logic or a dream-like state. One prominent idea from John Pierce, a Bell Labs electrical engineer, posited that the utility of math is that it could remove order from art, allowing for bursts of the unexpected that might lead to insight.
Early neural networks were built by psychologists who believed these systems could observe the messy processes happening inside the brain.
These early computer engineers viewed AI differently than we do today. While today's AI systems like ChatGPT and Claude are valued for consistently delivering correct answers, early advocates of these systems saw value in unexpected, even bizarre answers. This is because early AI systems were confined to laboratories, so there were no real-world consequences for incorrect or creative outputs.
However, today, legal and monetary reasons have led to a "sanding down" of strangeness, with these systems suppressing creative outputs.
Early neural networks that incorporated randomness to simulate human creativity evolved and became the dominant paradigm in AI. For a long time, symbolic AI was the dominant approach, where the world was thought to be based on rules, and the role of AI researchers was to reconstruct knowledge of the world in facts and propositions.
Neural networks gained traction in the 1950s but fell out of favor early on. By the 1980s, they shifted more into the world of cognitive science. To become more powerful and useful, neural networks needed a lot of data and computational power, which became available with the advent of the internet. Around 2012, neural networks started to outperform statistical machine learning and kept growing from there.
Written by urgent.news from MIT News AI's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
