Tracing AI Hallucinations and Creativity Back to Early Computing
A recent study explores how early computer engineers intentionally injected randomness into algorithms, linking the origins of AI hallucinations to attempts to model the human mind.
A recent academic paper published in the journal Past and Present investigates the historical roots of modern artificial intelligence, tracing how early pioneers deliberately incorporated randomness into algorithms to simulate creative human thought. Authored by Lindquist, the research highlights that modern challenges like AI hallucinations are deeply tied to these foundational decisions, revealing inherent tradeoffs between predictability and creativity in machine behavior.
In conversations about the research, Lindquist explained that scientists and artists have long theorized that chance plays a crucial role in discovery and original thought. During the early 1950s, Bell Labs electrical engineer John Pierce suggested that mathematics could be utilized in art not to create order, but to intentionally disrupt it. By injecting calculated moments of the unexpected, engineers hoped to bypass standard clichés and find new insights.
The Role of Early Neural Networks
To introduce this irregularity into otherwise deterministic computers, early researchers turned to artificial neural networks. Built largely by psychologists, these systems were envisioned as tools to peer inside the "black box" of the human mind and observe productive mistakes in action. While some believed creative thinking was simply standard reasoning executed at a higher level, others viewed random neural networks as a way to map the messy reality of human cognition.
This perspective contrasts sharply with modern expectations. Today, users rely on production chatbots like ChatGPT or Claude for dependable, accurate answers. However, early advocates initially prized neural networks precisely because they produced unexpected or bizarre outputs. Because these systems remained confined to laboratory environments, developers faced no real-world financial or legal fallout from strange results. In contrast, modern AI labs actively suppress erratic behavior, effectively sanding down the strangeness that stems from the technology's core.
Evolution and the Shift to Neural Networks
For many years, symbolic AI—which treats the world as a system governed by strict facts, rules, and propositions—dominated the field. Although early neural networks gained attention in the 1950s championed by figures like psychologist Frank Rosenblatt, the approach temporarily lost favor following Rosenblatt's accidental death in 1971 at age 43, paving the way for a shift toward symbolic systems.
Nevertheless, neural networks persisted, migrating into cognitive science by the 1980s. They eventually captured mainstream dominance around 2012, fueled by the explosion of internet data and massive increases in computational power that allowed neural networks to outperform alternative statistical machine learning techniques.
What it means for developers
For modern developers building applications, understanding this history provides vital context regarding the permanent tension between strict predictability and creative freedom. When working with state-of-the-art architectures, engineers often try to stamp out hallucinations and enforce rigid behavioral guardrails. Yet historical precedent suggests that completely eliminating unexpected outputs may come at the direct expense of genuine machine creativity and novel problem-solving. Developers looking to experiment with these capabilities can try top AI models cheaply through one API at https://apixoai.online.
Ultimately, history demonstrates that pushing an AI system toward one end of the spectrum invariably forces compromises on the other.
Source: Q&A: How AI came to model the “messy” human mind — MIT News AI. Written by the Apixo team from that report.
One key for Claude, GPT, GLM, DeepSeek and more. Pay per token with crypto.
Get your API key
