Conscious intelligence and reflective representation
[WORK IN PROGRESS]
This post is quite different compared to what I usually write here, it is more in the form of an academic paper. It is still in progress, I have a lot of ideas to process and put down on paper, and it will probably take me a while to finish it!
Introduction
The idea of the mind has been questioned since very ancient times in philosophy, and popularly encapsulated in Descartes’ ”dictum, cogito, ergo sum” ("I think, therefore I am"), although in those times, Descartes believed that the mind was separated from the body.
In ”A short story (and take) on life and consciousness”, I mentioned the intuition that consciousness takes its roots in the ability to take a step back on the information one perceives. Throughout the evolution of life, biological systems have developed diverse and complex sensory mechanisms to gather environmental information, such as vision, hearing, and olfaction in humans. These sensory mechanisms have co-evolved with highly sophisticated and efficient brains, capable of processing the gathered information intelligently. These characteristics have survived and continued to evolve through natural selection, as Darwin elaborated in "On the Origin of Species” (1859), because they are advantageous traits for survival. My intuition is that biological life and artificial machines (and other potential forms of life) which have acquired advanced sensors and information processing abilities, as well as developed specific brain architectures allowing them to take a “step back” on information and its processing, can display conscious behavior.
The “step back” ability refers to the capacity to think with a distance about what we perceive. It is not just about sensing and processing information, there is an additional layer. Instinctive and subconscious reactions still occur after sensing and processing information, but they lack conscious thought. For example, when you adopt a defensive stance after hearing a loud sound, you sense and process information, but react unconsciously. You were not conscious of it when you reacted, you didn’t think about it, but you still sensed and processed information, and reacted accordingly. Another example is experienced players and their reliance on “intuition”. They process more sensory information but exhibit less conscious thought compared to novices who must consciously think through their actions, as highlighted in Groot’s ”Thought and Choice in Chess” (1965).
Thus, consciousness involves thoughtfully representing information with a degree of separation, including the self. In this article, I will try to deepen this idea of consciousness and propose ideas for a theory of conscious intelligence.
Distinct component or emergent phenomenon?
A popular view currently is that consciousness is simply an emergent phenomenon of a powerful enough information processing system.
Tononi's Integrated Information Theory (IIT) proposes that consciousness arises from the ability of a system to integrate information. Integrating information means the ability of a system to combine diverse pieces of information into a unified, coherent whole, reflecting a high level of interdependence among its parts. IIT posits that the more information a system can integrate, the higher its level of consciousness.
Andrej Karpathy, in a discussion with Lex Fridman, said that consciousness is an “emerging phenomenon of a large enough and complex enough model”, because if “you have a powerful enough model of understanding the world, you actually understand that you are an entity in it”.
I have nothing against the view that consciousness is an emergent property of a powerful enough information processing system, I think that you need a powerful information processing system to be conscious. However, reducing consciousness to only being this immergent property is a reductionist view. As David Deutsch advanced in The Fabric of Reality, higher-level analysis is also very useful and often necessary for us to understand and explain phenomena. In fact, Deutsch posits that "the purpose of high-level sciences is to enable us to understand emergent phenomena", emergent phenomena being the "comprehensible facts that are not simply deducible from lower-level theories".
Ultimately, everything in the world is just about atoms, but does an atomic-level view enable us to understand and explain all world phenomena, like human psychology? No, we need different perspectives, at different levels. Consciousness is about information processing and is an emergent property of a powerful information processing system, but we can try to conceptualize its characteristics with a higher-level perspective to better understand and explain it.
Conscious intelligence
Components
We can outline three main components for a theory of conscious intelligence:
1. Sensing Information: The ability to sense and capture information, considering all the information that exists.
2. Processing Information: The cognitive ability that transform sensed information into meaningful information/insights.
3. Reflective representation: The ability to represent and reflect on information in a parallel mental model, with self-awareness.
As we explore in greater detail in the next chapter, reflective representation is the reflective component of conscious intelligence, or what allows an intelligent system to be conscious. It involves creating a representation of information in a parallel mental model. This parallel mental model also allows for the creation of and reflection on new information. Consciousness thus includes the ability to model and reflect on information in a parallel manner (not completely detached from reality as we will see), with self-representation being crucial. It must include the ability of representing yourself. If you cannot represent yourself in a parallel mental model, are you conscious? Without this self-representation, one cannot be truly conscious.
Expression
Not intended as a precise mathematical formula but rather as an illustration, an indicator of conscious intelligence could be expressed as:
Conscious intelligence indicator = Sensing × Processing × Reflective representation
Where:
Sensing is the proportion of available information a being can capture (0 to 1).
Processing is the proportion of usefulness a being can derive from information (0 to 1).
Reflective representation is the ability to represent and reflect on information in a parallel mental model (0 to 1).
If the total equals 1, we have conscious omniscience, a being capable of sensing, processing, representing, and reflecting on everything.
The theory suggests a gradient of consciousness in nature. Living beings exhibit varying levels of consciousness, with humans potentially having the highest known level. A rock doesn’t have it, a dog does, a human does even more. Machines might also be able to possess some degree of consciousness, though to what extent remains uncertain.
Interconnection between components
There is an important observation to make: the different components of the theory are linked to and have an impact on each other. You cannot represent information in a parallel mental model if you do not sense anything or cannot process what you sense. The more numerous and diverse information you can sense, the more you can process it meaningfully, e.g. in integrating information in a larger context to take better decisions. You can process information more effectively if you can represent it in a parallel mental model, e.g. for consciously simulating scenarios.
We could even imagine that being intelligent (efficient with information) might be necessary to be conscious, and that being conscious might be necessary to be intelligent.
Some would even argue that reflective representation is a component of information processing. I am not against this idea, but I still isolate it from the rest as I think it is the final element allowing the apparition of conscious thoughts.
Elements of reflective representation
Reflective representation could be schematized as:
You exist in the world and perceive it -> you can trigger a parallel mental model -> in which you can represent (create and recreate) information, including yourself (self-awareness) -> and reflect on this representation of information. [should we go even further and you should be able to reflect on the reflection-processing itself?]
Attention (focus) being a crucial element at every step.
Even though it might be way more complex in reality (notably with integration at different layers, including those we don’t perceive), we want here to formulate the philosophical concept from a relatively high perspective.
Let’s explore these elements.
Parallel mental model
As mentioned in the previous chapter, reflective representation first involves the ability of creating a representation of information in a parallel mental (or world) model.
This notion takes its root in the neuroscience idea that what we see of the world is actually a recreation of it in the brain. We don’t see the world “as is”, but instead we recreate an image of it in our brain using the information collected via our sensors. This concept is supported by works such as Marr's “Vision” (1982) which discusses how the brain processes visual information to construct our perception of reality, and by the principles outlined in the predictive coding framework by Friston (2005), which suggests that the brain continually generates and updates a model of the environment based on sensory input.
It is linked to the concept of mental models which is extensively explored in neuroscience, psychology, and philosophy, e.g. in the work of Johnson-Laird (1983), who posits that humans create mental simulations of the world to reason and make decisions. Similarly, it is linked to the concept of world model in artificial intelligence and robotics, such as in the work of Ha and Schmidhuber (2018), who showed how agents can build and utilize internal models of the world to simulate and plan actions.
More generally, it is the conceptualization of how a system creates an internal representation of the external world (or information in general) in order to understand, predict, plan and make decisions.
Scientific evidence supports the existence of these parallel mental models. Brain networks such as the default mode network (DMN) and the frontoparietal network are known to be active when individuals are engaged in self-referential thinking, imagining, and planning (Raichle et al., 2001). These networks facilitate the construction of internal models that integrate sensory information with past experiences and future possibilities (Buckner & Carroll, 2007).
Creation
This parallel world model also allows for the creation of and reflection on new information.
It is not just about information integration, but about recreation and even "pure" creation (in the limits of our imagination). You can imagine yourself, and you can imagine yourself with different hair, as well as in a physical form you’ve never seen before.
When you experience, e.g. when you are facing plants, you can see the plants, you can consciously think about concepts about those plants, you can imagine a new plant among them, you can close your eyes and recreate them, or recreate them in another color or in another landscape.
This mental model takes its roots in reality, we recreate reality (what we perceive of it) in it, and can create what we want within the limits of our imagination.
Of course, this parallel mental model is linked to the “real” world, you mostly developed and learned through your interactions with the “real” world, but there is a large gradient of detachment and creativity. In this parallel world, you can represent and think about the “real” world, imagine creative elements in the “real” world, imagine “real” elements in a creative world, or even imagine creative elements in a creative world. The creative elements are highly influenced by what you’ve seen and learned in the “real” world, and your interpretation of “real” elements is influenced by the creative ones you’ve seen.
Self-awareness
As mentioned above, self-representation is a crucial aspect. Reflective representation must include the ability of representing yourself. If you cannot represent yourself in a parallel world model, you cannot be truly conscious, at least not at the human level.
Descartes gave us the intuition that self-awareness and reflective thoughts have an essential role in the idea of consciousness.
? Even further: ability to represent our own process thinking, to be aware of our thoughts, as a condition as well.
Neural Correlates of higher-order thoughts, identified by Lau and Rosenthal (2011), highlight the brain areas involved in conscious awareness. They argue that conscious states are those accompanied by higher-order thoughts, meaning we are aware of ourselves experiencing these states. Higher-order theories of consciousness, like those proposed by Rosenthal, suggest that conscious states require a higher-order awareness of one’s mental states. Rosenthal posits that having thoughts about our thoughts is crucial for consciousness.
? Different orders: simple processing (not conscious), conscious thinking, conscious thinking about conscious thinking. Is it a condition? or just a by product? because then I can consciously think about consciously thinking about consciously thinking, it doesn’t stop.
On higher order: Descartes said “I think therefore I am”, but how do I know that I think in the first place? Maybe the reality is “I know that I think therefore I am”. This definition of consciousness might be more accurate.
Attention
To trigger the conscious state, to represent things, to reflect on things, you must be able to focus your attention on them.
William James’ “The Principles of Psychology” (1890), in which he introduces the pioneering concept of the stream of consciousness, James also highlights the crucial role of attention in this process, asserting that it acts as a selective mechanism that filters and prioritizes sensory input, thus shaping our conscious experience. By focusing our awareness on specific aspects of our environment or internal states, attention helps create a coherent and meaningful narrative from the constant flux of mental activity.
Graziano’s AST: The theory begins with attention, the process by which signals compete for the brain’s limited computing resources. The Attention Schema Theory (AST) by Graziano posits that consciousness arises from the brain’s model of attention. Graziano explains that our awareness is a constructed representation, an internal model tracking attention.
