Article: Mixed Signals—How Using Biology to Explain AI Can Mislead

Photo Source: Pixabay

Artificial Intelligence (AI) and machine learning (ML) continue to revolutionize many aspects of our world, including business, manufacturing, science, law, entertainment, transportation, law enforcement, and education. When computer scientists and researchers began to come up with language to describe artificial intelligence and its concepts, they used biological terms to help people understand the way AI "learns". The problem is that such a linkage causes misperceptions about the nature of biological systems and computer science. The pioneering work of Frank Rosenblatt (July 11, 1928 – July 11, 1971) at Cornell University in the 1950s led to the first computer program that could perform human-like tasks, such as learning to classify objects. In this work, Dr. Rosenblatt borrowed terminology and concepts from neuroscience in the design of a system he called a Perceptron. He modeled the Perceptron to act like a nerve cell or neuron. Biological neurons transmit signals first by changing the electrical property of the nerve based on an input signal. With a strong enough change in electrical potential, the nerve sends chemical messages to adjoining nerves. This process, called signal propagation, underlies coordination in animals. These propagated signals, such as sight or hearing, send information to the brain for processing and back from the brain to the body to perform an action such as walking, running, or standing still. A similar process in the brain helps to decide if some information has value and should be remembered. Rosenblatt’s Perceptron consisted of a group of inputs that would flow through a layer of filters to produce an output similar to how a nerve in a biological system receives information and transmits a signal to other nerves based on signal strength. Rosenblatt’s work fundamentally changed computing and ushered in the field of deep learning—a type of AI that learns from large amounts of labeled data. However, the fact that he modeled his research based on nerve function has created the misconception that the nervous system in humans and animals functions the same way as a computer. Computers run programs that rely on digital or “on-off” commands to process information, whereas the nervous system combines biochemical interactions and nerve impulses. The nervous system acts with a mixture of analog or smooth, continuous functions punctuated by on and off or binary signals. Such a fundamental difference suggests that deep learning AI does not model the brain. Humans, for example, learn from much less data than deep learning or Perceptrons. A deep learning model like chatGPT requires billions of inputs to learn, whereas a human can learn from even a single example. For instance, a young child will learn that fire burns from one single touch of a flame. Such an adaptive ability to learn helps people respond to changing environments and circumstances, differentiating human brains from computers. AI and machine learning, especially deep learning, owes its initial discovery to the pioneering work of the psychologist, computer scientist, and neurologist Frank Rosenblatt in the 1950s. Rosenblatt pioneered the first example of machine learning based on the functioning of a single neuron that receives input and, with a strong enough input signal, will fire and propagate the signal. Based on this type of logic, Rosenblatt developed the first computer that could learn to differentiate between different inputs. He used biological terms such as neuron and neural networks to describe his invention called the Perceptron. Still, such terminology has caused many to confuse the workings of a computer with the functioning of the nervous system, especially the brain. Computers function in a digital or on-off system to process information, while the brain exhibits a mixture of smooth analog functions punctuated by on-off nerve pulses. Conflating computer and brain functions remains tempting, but they work in very different ways.

Dr. Smith’s career in scientific and information research spans the areas of bioinformatics, artificial intelligence, toxicology, and chemistry. He has published a number of peer-reviewed scientific papers. He has worked over the past seventeen years developing advanced analytics, machine learning, and knowledge management tools to enable research and support high-level decision making. Tim completed his Ph.D. in Toxicology at Cornell University and a Bachelor of Science in chemistry from the University of Washington.

You can buy his book on Amazon in paperback and in kindle format here.