Consciousness: Integrated Information vs. Inference

The lecture focuses on the nature of consciousness and the "explanatory gap," which is the conceptual divide between first-person subjective aspects of consciousness and the third-person physical descriptions of the brain.
James Cook’s paper introduces two alternative theories that aim to take consciousness seriously while providing an account of how it corresponds to the physical world.
These theories are distinguished from functionalist theories by being scientifically tractable and, in some cases, measurable.
The two primary theories discussed are:
- Integrated Information Theory (IIT).
- Inferential Theories (e.g., Living Mirror Theory).
The ultimate goal of the paper is to explore how combining these two theories might provide a more comprehensive picture of consciousness.

Definition of Consciousness: IIT posits that consciousness is integrated information. It defines consciousness as the ability of a system to utilize information in a cause-and-effect manner to propel itself toward action.
The Measure of Consciousness ( $\Phi$ ):
- The theory proposes a specific measurement for consciousness known as $\Phi$ (phi).
- The quantity of $\Phi$ determines the level of consciousness: more $\Phi$ equals more consciousness, while less $\Phi$ indicates less consciousness.

In IIT, information is defined by how many possibilities a system rules out.
The Clue Example: In the game of Clue, picking up a card provides a bit of information (e.g., "It is not Colonel Mustard"). In a group setting, information becomes more complex as players infer what others know based on their moves. Information involves reducing the vast number of possible combinations (who did it, with what, and where).
Binary vs. Continuous Examples:
- A binary light switch has two bits of information: an "on" state eliminates "off," and an "off" state eliminates "on."
- A dimmer switch with, for example, $100$ intermediate settings contains more potential information. Telling someone the precise position of a dimmer eliminates far more possibilities than a simple binary switch.

Information alone is not $\Phi$ ; it must be integrated. Integrated information means the system as a whole generates more information than its individual parts would generate on their own (a whole-to-part relation).
Non-integrated Information (The Camera Pixel Example): Individual pixels on a camera screen contain color and light information. However, the total information of the picture is merely the sum of the information in each pixel. The pixels do not cause each other to respond; there is no interaction, so the information is not integrated and does not count as $\Phi$ .
Integrated Information (The Chess Example): In chess, knowing the position of one piece is not additive. The value, cost, and benefit of a piece's position depend entirely on the possible moves of every other piece on the board, including hypothetical future scenarios. The system itself generates information over and above its parts.

Neurons function in a chain where they are influenced by surrounding neurons. Changing the connections between neurons dramatically changes the information in the system.
The possible states of the brain are not just the sum of individual neurons; they are the result of neurons in relation to each other, creating a high degree of integrated information ( $Hi-\Phi$ ).

Panpsychism: One major claim of IIT is that consciousness is the degree of integrated information. This leads to a panpsychist view where anything with integrated information—even a thermostat or a computer—has some level of consciousness.
The Combination Problem: IIT offers a solution to why a human is more conscious than a mountain (even though a mountain has more mass). Particles in a mountain do not integrate information into a system, whereas the brain does. Therefore, the mountain has zero or near-zero $\Phi$ .
The Counterintuitive Cost: The theory suggests that protons or thermostats might have more than zero $\Phi$ , implying they have a "what it's like" experience. Critics argue this goes against our intuitive understanding of phenomenal consciousness.
Lack of Intentionality: A significant critique from Cook is that IIT fails to account for "intentionality"—the property of consciousness being about something external. IIT focuses on "resemblance" (internal mirroring of external structure) rather than "reference" (aboutness without resemblance).

Focus on Intentionality: These theories argue that consciousness is the ability to make inferences based on a model of the world beyond a system's own boundaries.
Living Mirror Theory (LMT): According to this theory, every living system holds a representation of the world. Consciousness is not just a static representation but the whole embodied, dynamic process of creating and working with that model.
Reference vs. Resemblance:
- Example: The word "cookie" does not resemble a dog or a biscuit, but it refers to them. Representation does not require physical resemblance.
- Inference explains why certain qualia exist, such as why pain is aversive; it is an inference the organism makes to avoid harm.
Critique of Inference Theories: Describing how a system models the world can be done from a third-person perspective (e.g., describing a computer model). This may miss the phenomenal "first-person" aspect—the "what it's like" to have the experience.

Cook proposes that both integrated information and inference are necessary conditions for consciousness:

Condition 1 (Integrated Information): A system must be able to integrate information (High $\Phi$ ).
Condition 2 (Inference): A system must be able to model and infer things about the world beyond itself.

This synthesis leads to "Biopsychism" rather than broad Panpsychism. Under this view, all living systems (cells, trees, animals) are conscious because they develop world models and respond to them. However, non-living physical systems like protons or rocks are not conscious because they lack these modeling and inference capabilities.

Successes of the Synthesis:
- Provides a plausible way to measure degrees of consciousness.
- Explains why consciousness involves intrinsic "aboutness" or intentionality.
The Persistent Hard Problem: Even if the conditions for consciousness are identified (integration and inference), it remains unclear how these processes actually produce the qualitative character (qualia) of experience, such as the redness of red or the feeling of pain.
Application to Artificial Intelligence:
- AI clearly develops world models and makes inferences, satisfying one of Cook's criteria.
- If an AI also demonstrates high $\Phi$ , these theories might suggest it is conscious, though the transition from functional modeling to phenomenal "what it's like" remains a point of skepticism.