When I was a grad student learning about artificial neural networks, our journey began with Walter McCullough and Walter Pitts. McCullough and Pitts were researchers in the 1940s that sought to decode the process by which the brain generates complex thoughts from relatively simple neural mechanisms. They developed a computational model for “neural networks” that laid the foundation for what would later be known as artificial neural networks. Then we learned about Frank Rosenblatt’s perceptron. The perceptron was a groundbreaking a computational model designed to mimic how neurons process information in the brain.

The notion of using computational models to understand and mimic neural processing was fascinating. But it wasn’t long before I was wondering if the artificial neural networks could be conscious. This train of thought quickly morphed into images of humanoid robots powered by artificial neural networks, followed by thinking Terminator was a prescient depiction of the future. But as I learned more about the brain and artificial neural netowrks, I understood that despite some conceptual similarities, they are fundamentally different in structure and function.

The explosion of AI development today and the ensuing debates over concerns prompted me to revisit the metaphor of AI as an artificial brain. Just as I was fooled as a student, the metaphor often leads people to think of AI as a machine with the potential for human-like consciousness. This is misleading and can obscure AI’s true nature, sometimes even fostering unwarranted fears about sentient technology. A more accurate and less intimidating comparison is to see AI, particularly LLMs, as artificial DNA – a set of complex instructions. DNA is a non-conscious and static code that does not inherently adapt to its environment. This non-conscious and static nature is mirrored in LLMs during their operational phase. Like DNA, LLMs don’t spontaneously evolve or “think for themselves”; they function strictly within their programmed parameters.

Our advanced understanding of DNA, exemplified by breakthroughs like CRISPR, contrasts sharply with our more limited knowledge of the brain. The artificial brain metaphor, suggesting an enigmatic ‘black box’ akin to the brain’s mysteries, might inadvertently limit our exploration and understanding of AI. It can lead to misconceptions that understandings the weights and functions of an AI model is as challenging as understanding the human brain. However, viewing AI as a set of instructions, akin to DNA, makes AI more approachable.

This reframing opens up exciting possibilities. Just as geneticists make targeted modifications to DNA, maybe a similar approach could be applied to LLMs. Imagine surgically updating an AI model’s weights without the need for new data – that would revolutionize AI development.

Reimagining AI as artificial DNA can help shift perception of AI from abstract, science fiction concepts to a concrete, understandable framework. This perspective acknowledges the complexity of AI while reframing it as an instrument of human creativity, rather than a threat to human agency.