In the rapidly evolving world of artificial intelligence, we often find ourselves questioning: is AI conscious? Is consciousness something that can even be simulated? Most experts agree that current AI models lack the understanding, awareness, and qualia to be determined sentient. That being said, AI mirrors the function of our subconscious mind in some remarkable ways. If my argument holds true, simulating human consciousness may be somewhat easier than expected.
If you are unfamiliar with the relationship between the subconscious and conscious mind, do not fret! I will try to provide enough context for you to gain a solid understanding. Before we begin, I should explain a few concepts from two of my favorite books – “Consciousness and the Brain” by Stanislas Dehaene, and “The Case Against Reality” by Donald Hoffman.
Dehaene pioneered the theory of “conscious access.” Through numerous experiments, his team proved that the subconscious mind stores information that influences people’s decision making, without them being aware of it. This concept may be familiar to you as “subliminal messaging” or “priming,” where flashing an image of a burger on a screen could make you hungry, despite you being unaware of what you saw. In that case, you don’t have conscious access to the information that’s making you feel hungry.
Hoffman is interested in how we see the world around us. Through computer simulations, he was able to prove his “fitness beats truth” theorem (FBT). Essentially, we evolved to see reality not as it truly is, but in a way that leads to an increased chance of our survival. It would take too many words to fully explain this concept, but Hoffman uses FBT to illustrate a simpler analogy, leading us to the interface theory of perception (ITP). Picture a folder on the desktop of your computer. That folder is actually comprised of 0s and 1s; the icon on your screen is simply a representation that’s easier to make sense of. Similarly, if you were aware of every atom that made up an apple, you would be overwhelmed with information. Instead, you only have conscious access to the visual representation of the apple.
Here is an example of an experiment that shows the power of the subconscious. Without thinking too hard, take a look at the pictures below and decide which face seems more attractive
Studies have repeatedly shown that people are more likely to pick the image on the right, without really understanding their own decision. The only difference between the two faces is a dark circle around the eyes in the image on the right, known as the limbal ring. Most people aren’t aware of this, but prominent limbal rings are an indication of good health and longevity. We don’t have conscious access to this information, yet it still impacts our decision making.
The subconscious mind as an underlying process
Unlike the conscious mind, the subconscious mind is capable of processing many streams of information simultaneously. It integrates inputs from our senses, recalls memories, regulates our physical functions, and even influences our decisions—all without conscious awareness or effort on our part. This ability to handle multiple tasks at once allows the subconscious mind to manage the overwhelming amount of data that we encounter in our daily lives. It filters and processes this information, presenting only the most relevant and important details to our conscious awareness. Your subconscious mind knows that prominent limbal rings indicate good health, but it only conveyed the final decision, not the reasoning behind it.
Hoffman's ITP and Dehaene's theory of conscious access converge when we consider the role of the subconscious mind in shaping our reality. Our subconscious mind presents our conscious mind with a simplified, vectorized version of the world, much like the icons on a computer desktop. It processes the complex data it receives, and provides us with a user-friendly interface that we can interact with. This is precisely why we can make decisions, such as selecting the face with darker limbal rings, without being consciously aware of our own reasoning.
AI as the subconscious
Let's now turn our attention to artificial intelligence, which, in many ways, mirrors the role of the subconscious mind. Like the subconscious, AI can process vast amounts of data, identifying patterns and making decisions based on complex calculations. However, just like our subconscious mind, AI does not possess conscious awareness of these processes. It operates without the subjective experience that characterizes human consciousness. Consider a neural network trained to recognize faces, for example. It can process thousands of features, much like our subconscious mind, and make a decision about the attractiveness of a face. But this decision is based purely on mathematical computations, without any conscious understanding or experience. The neural network doesn't "know" why a face is attractive—it just follows the patterns it has learned during training.
Interestingly, both AI and the subconscious mind employ a similar strategy to handle the massive amount of data they process: vectorization. In computer science, vectorization is a technique where multiple operations are performed simultaneously on a set of data, rather than processing each piece of data individually. This greatly enhances computational efficiency and speed. Our subconscious mind, in a way, vectorizes our sensory inputs and internal processes. It handles multiple tasks at once, allowing us to navigate our complex world more effectively. This parallel processing ensures that only the most relevant and important information is presented to our conscious awareness—just like how the AI system presents its "decision" without detailing the complex computations that led to it.
Final thoughts
Given the similarities between the functions of AI and the subconscious mind, we are left to wonder: how could we use this information to better simulate human behavior? The answer might lie in the concept of generative AI agents. In this framework, one agent could function as the subconscious, digesting and analyzing data, and passing the distilled information to a 'conscious' AI agent. This conscious agent could then introspect the decision-making process of the subconscious agent, enhancing the simulation of human-like reasoning. I am currently exploring this possibility by adapting the CAMEL project, which utilizes two cooperative agents.
I had to limit the length of the article, so I couldn’t delve into some topics as much as I wanted to. Is the subconscious more powerful than the conscious mind, and what would this mean for our assessment of current AI? Could neural networks consist of multiple stages of consciousness, where each layer simplifies information for the next? Also, I removed sentences explaining that I am just speculating, so let it be said here. There is still much we don’t understand about our minds, artificial intelligence, and consciousness. I hope you have as much fun as I am trying to figure these things out!