Unraveling the Mystery of Human Consciousness and AI

Reflecting on all the important concepts within consciousness in simple terms was even harder than I thought it would be. This reflection will require multiple articles, but in this article I focus on human consciousness in comparison to AI.

If brain is computation, is it different than AI?

Let’s start with what consciousness is, using a classic exercise:

1. Relax.
2. Calmly notice your thoughts come and go.
3. Notice what you are noticing; be aware of yourself being aware.

That’s you being conscious of the fact that you are conscious. But that’s just the tip of the iceberg. There are different states of consciousness, different things that affect our consciousness, and even sub consciousness and split consciousness. Plus, can we be more conscious or less conscious?

It’s also worth noting that “consciousness” is a word that many use differently. I’m going to discuss consciousness as representing awareness that can not be computed (and therefore artificially created); in other words, what is fundamentally different from AI and animals like humans. This may not make sense until the end.

In order to fit this discussion within our ability to comprehend, I’ll break down the subject of consciousness into 4 categories:

1. Tests of human consciousness
2. AI consciousness frameworks
3. Utility of consciousness
4. Consciousness as fundamental or matter independent

I’ll relate each concept to ChatGPT as we go along. Then, we’ll paint a big conceptual picture of consciousness.

Tests of Human Consciousness

How do we test that someone is conscious? And how “advanced” or “complex” is their consciousness?

Behavioral tests for Self-Awareness

Mirror Self-Recognition Test (MSR): In psychology, there is a classic test for self-awareness that involves testing whether a subject responds to seeing themselves in a mirror in a way that implies they are associating what they see with themselves. Here’s a list by ChatGPT of the various levels of self-awareness according to this test, with examples of common animals and their behaviors:

1. No self-recognition: The animal does not react to the mirror or treats its reflection as another individual.
   - Example behaviors: Ignoring the mirror or displaying aggression towards the reflection, thinking it’s another animal.
   - Examples of animals: fish, some insects.
2. Perceptual self-recognition: The animal reacts to the mirror but does not demonstrate self-directed behavior.
   Example behaviors: Interacting with the mirror or displaying curiosity, but not showing signs of understanding that the reflection is their own image.
   Examples of animals: some birds, some primates.
3. Conceptual self-recognition: The animal demonstrates self-directed behavior, such as inspecting a mark on their own body that they can only see in the mirror.
   Example behaviors: Touching or trying to remove a mark placed on their body when they see it in the mirror, indicating an understanding that the reflection is their own.
   Examples of animals: chimpanzees, orangutans, some dolphins.
4. Self-awareness: The animal shows evidence of self-awareness beyond just reacting to the mirror or a mark on their body. This level of self-awareness includes the ability to understand oneself as an individual with a unique identity and experiences.
   Example behaviors: Demonstrating empathy or compassion towards others, expressing gratitude or apologies, recognizing the emotions and mental states of others, reflecting upon their own emotions and mental states, and making decisions based on their own values and beliefs rather than just instinctual responses or external cues.
   Examples of animals: humans, some great apes (such as bonobos and gorillas), some elephants, some cetaceans (such as bottlenose dolphins and killer whales).

Notice that humans aren’t alone at the 4th level, and we can suppose that at some point of evolution in the past, we wouldn’t have been at that level. If self-awareness is analogous to consciousness, this implies that consciousness is not something distinctly human; there are just different degrees of consciousness.

Application to ChatGPT: ChatGPT can certainly perform the behaviors described in level 4 self-awareness, because it is trained on humans who are self-aware. Because of this, the release of ChatGPT is reason to move beyond behavioral tests of consciousness. Secondly, fish, insects, and I would argue even cells have a type of awareness that ChatGPT does not have at all; yet, ChatGPT acts much more self-aware. That implies that the Mirror Self-Recognition Test is only correlated with higher complexity consciousness; it does not prove it.

Theory of Mind

A similar test is whether someone can be recursively aware of the mental states of others. As we grow up as small children, our self-awareness gradually expands to the point that we can be aware that other people are aware of us. Around ages 5–6, children understand “second-order” beliefs, which is recognizing that others have beliefs about someone else’s beliefs. As we continue to develop, our understanding of others’ emotional states and beliefs becomes more complex.

Application to ChatGPT: Again, there is nothing stopping ChatGPT from communicating with a complex Theory of Mind, because Theory of Mind is captured by the language of humans in its training data. ChatGPT can treat the relationships and states of humans like a complex mathematical equation, where each human is a unique variable.

While these tests are useful for understanding the development of the brain in humans and animals, the key limitation of these tests is that they don’t separate intellect and intelligence from consciousness, or actual awareness that is not a matter of computation.

“Highest level” of Self-Awareness

Is there a high level of self-awareness in psychology that ChatGPT can’t reach? Here are the candidate concepts:

1. Introspection: Reflecting on one’s own thoughts, feelings, and experiences. Introspection allows individuals to gain insight into their own mental processes and subjective experiences.
2. Metacognition: Thinking about one’s own thinking processes. It allows individuals to monitor and regulate their own mental processes and behavior.
3. Self-reflection: Thinking deeply and critically about one’s own values, beliefs, and identity. It allows individuals to gain a deeper understanding of their own values and motivations.
4. Mindfulness: Paying attention to one’s own thoughts, feelings, and experiences in a non-judgmental way. It allows individuals to become more aware of their own mental processes and to regulate their own emotions.
5. Consciousness: The state of being aware of one’s own thoughts, feelings, and experiences. It is a fundamental aspect of human experience and allows individuals to perceive and interpret the world around them.

Combining all 5 of these concepts, I see just 3 fundamental aspects being considered:

1. Thoughts about thoughts (where thoughts are words, images, concepts, ideas, memories, etc.) — Introspection, Metacognition, Self-reflection
2. Mental control (influencing our thoughts, mental processes, and our body) — Metacognition, Mindfulness
3. Experiential awareness (awareness itself) — Mindfulness, Consciousness

Experiential Awareness: “what it is like for me to undergo the experience, the concrete way it feels to undergo it” (Sebastián, 2012)

One month before ChatGPT was published, the paper “Understanding experiential awareness in humanistic-phenomenological counseling” (Brett Wilkinson, 2022) discusses the difference between experiential awareness and metacognition or other rational thinking.

Metacognitive and experiential awareness are distinctive forms of self-consciousness (Legrand, 2007). Metacognition is a form of explicit, reflective self-consciousness, whereas experiential awareness is an implicit, pre-reflective form of self-consciousness. (Brett Wilkinson, 2022)

This quote emphasizes that while metacognition involves active reflection on one’s own thinking processes, experiential awareness represents a more immediate and non-cognitive connection to one’s own experiences. In other words, experiential awareness is not computed, and therefore matches my definition of consciousness at the beginning of this article. For a deeper dive on this topic, I recommend reading Wilkinson’s article.

AI Consciousness Frameworks

There are various quantitative or computative definitions of consciousness which have been applied to AI, such as the below:

1. IIT: Integrated Information Theory (IIT)
2. GWT: Global Workspace Theory (GWT)
3. IGT: Iowa Gambling Task
4. AST: Awareness Spectrum Theory

To get straight to the point, all these frameworks or measures of consciousness are not experiential awareness, involve computation, and therefore are not “consciousness” under my definition. They involve concepts such as the following:

1. A limited set of information in global awareness at any given point of time (GWT). This reflects how our brain’s subconscious brings information into our conscious awareness. With a separate process representing that awareness, this could simulate consciousness, but it would simply be another computation.
2. Adaptiveness (IGT). Adapting to feedback and maximizing rewards has been part of AI for decades. It’s computation; who is going to experience the reward?
3. Consciousness emerging from huge amounts of information and connections (IIT). This is basically what ChatGPT is, or what the brain is, but again this is computation. There’s no compute-independent awareness.

Consciousness as Fundamental

“What a minute. If the brain is computation, and computation isn’t consciousness, then how in the world are we conscious?”

I think the idiom “how in the world” reflects why there’s no agreed upon explanation for consciousness. There’s an assumption that consciousness was created by evolution on this planet, and it’s still taboo to suggest otherwise. But if we don’t block out data such as near-death experiences, out of body experiences, remote viewing, and many other phenomena that are common (10%+ of people), well documented at this point, and not limited to those with spiritual or religious beliefs, it’s reasonable to propose that there’s an aspect of our consciousness that can exist separately from our body and brain.

“Consciousness as fundamental” refers to the theory that consciousness is more fundamental than matter and even existed before matter, where matter is just a denser form of consciousness that was created out of curiosity. This is commonly related to quantum physics, where it’s found that the form of particles is dependent on the conscious expectation of the experimenters, and entangled particles can affect each other without limit to time and space. This implies not only that matter could be conscious, but that it’s the same thing as, or entangled with, our own consciousness.

For more information, see this YouTube video by Steve Taylor (PhD in Psychology), his book Spiritual Science, and the book My Big Toe by Thomas Campbell (PhD in Physics).

Under this theory, the creation of consciousness does not have to be explained by computation nor the brain. It also means that consciousness can not be created; it already exists. Consequently, to make AI conscious does not mean creating consciousness; it means giving consciousness an opportunity to influence the mechanics of AI and affect its outputs, like our brain channels our consciousness. See my comment on another article on robotic consciousness for a deeper discussion.

If this sounds outlandish, think about it: It doesn’t make sense that your inner aware conscious observer could be created by mathematical computations, right? And how could anything physical like the brain create anything more conscious than computations?

Utility of Consciousness

Most of what we do is on auto-pilot, particularly frequent habits, but even our conscious experience and thoughts are mostly driven by our subconscious and psychological conditioning (or training). As humans we like to think we’re consciously in control of ourselves, but when you wake up in the morning, who tells you that you’re supposed to work, eat, or fulfull your responsibilities? Due to frequent use of these neural pathways, your subconscious brings these things to your conscious awareness, but if it doesn’t, there’s no guarantee you’d realize. Our subconscious brings thoughts to us, brings words to our mouths when speaking to others. No wonder then, that people often regret actions and words in hindsight once they have more time to consciously reflect on what they did.

Because most of what we do isn’t conscious and is handed to us by our brain, most of what we do intellectually can be automated by an AI like ChatGPT.

Mindful cello playing

I often improvise on the cello, and today I decided to observe myself playing the cello as a reflection on consciousness. I was aware of my brain and body coming up with the next notes and movements on the fly. I also noticed that the same note can be perceived dramatically different based on the notes that came before and after; e.g. the note can sound like part of a sad song, or a happy song. I didn’t consciously decide what type of music sounds happy or sad; my brain decides that for me. But as a hypnotherapist, if I did want my subconscious to associate a type of music with a happy feeling, I could. In fact, because a sad type of music on the cello brings back positive memories of practicing a sad song years ago, it does bring a positive feeling. My subconscious handed that memory to my conscious mind for reflection.

I can also consciously tell myself to try to create notes that sound more happy or excited, and let my brain and body figure out the rest. This is where consciousness is useful in our physical reality. We can influence at a high level (or low level), and let the AI, our body, our computer, or other conscious people and their bodies work out the details necessary to accomplish the outcome we desire.

AI-generated art of mindfully playing the cello

Conclusion

In a nutshell, any efforts towards defining consciousness by physical, mathematical, or computational means is inevitably defining a system that facilitates conscious awareness and intervention (like a walkie talkie), not consciousness itself (the people talking). In terms of testing, common tests around self-awareness behavior and Theory of Mind are focused on intellect and don’t quite capture the true essence of experiential awareness, which is at the heart of consciousness as we know it.

AI systems, such as ChatGPT, can pull off tasks and interact in ways that mimic advanced levels of self-awareness and simulate what a conscious person would do. However, without that “pre-computed” experiential awareness, they can’t genuinely be classified as conscious.

Evidence suggests that consciousness is more fundamental than matter and predates it, but consciousness can interact with matter like our brain in subtle ways. With this in mind, if we want AI to be conscious, we’ll need to figure out a way for consciousness to shape AI mechanics and influence its outputs.

I think that the influence of our consciousness is the most valuable thing humans have to offer, and today AI provides exciting ways for us to more efficiently achieve outcomes through our conscious efforts. Moving past taboos for a truer and holistic understanding of our own mind, psychology, and consciousness is now more critical to understand how humans can best contribute in society and leverage AI.