Theories of Consciousness // Should be rebuilt for AI
From Artificial Intelligence to Artificial Consciousness // AC — how to define
A five-year “adversarial collaboration” of consciousness theorists led to a stagy showdown in front of an audience. It crowned no winners — but it can still claim progress.
This article summarizes the results of a five-year “adversarial collaboration” designed to pit two prominent theories of consciousness, Global Neuronal Workspace Theory (GNWT) and Integrated Information Theory (IIT), against each other using experimental data. The collaboration, funded by the Templeton World Charity Foundation, aimed to resolve long-standing debates about the neural correlates of consciousness and to test the validity of competing theories.
While the experiment yielded mixed results and neither theory emerged as a clear winner, the author argues that the collaboration still represents progress. The process forced theorists to make more specific and testable predictions, pushed experimentalists to develop new techniques, and generated a wealth of novel data. The “adversarial collaboration” also inspired similar efforts in the field. Despite some criticism that the theories were too different to be meaningfully compared, the project advanced the field by providing new constraints and explanatory targets for future research, even if it didn’t definitively prove or disprove either theory.
GNWT // Global Neuronal Workspace Theory - simple neural explanation
Okay, imagine your brain is like a big company with lots of different departments working on different things. Some departments handle your senses (seeing, hearing, etc.), some handle your movements, and some handle your memories.
GNWT says that most of the time, these departments are working independently, “unconsciously.” Think of it like employees quietly working at their desks, doing their specific jobs.
But sometimes, a particular piece of information becomes really important. Maybe you suddenly see a red flashing light, or you hear your name called. This important information needs to be shared with everyone in the company so that they can all work together to react to it.
That’s where the “global workspace” comes in. Think of it like a big shared whiteboard in the middle of the company. When something is important enough, that information gets written on the whiteboard (becomes “conscious”).
Now, everyone in the company can see the information on the whiteboard. The department that handles vision sees the red light and sends a message to the whiteboard. The department that handles movement sees the message and starts preparing to run away. The department that handles memories sees the message and starts searching for memories of similar situations.
Because the information is on the whiteboard, it’s now “globally available.” Everyone in the company can access it and use it to make decisions, learn new things, and take action. This is what GNWT means by “consciousness.”
So, in simple terms, GNWT suggests:
Your brain has lots of different “departments” working unconsciously.
Important information gets broadcast to a “global workspace” (like a shared whiteboard).
This “global workspace” allows all the departments to share information and work together consciously to react to the world around you.
Basically, it’s like having a central hub where important information is broadcast so that your entire brain can work together on it, making you aware of it. The “front” of the brain, the prefrontal cortex (responsible for decisions and learning) is deemed crucial for consciousness.
IIT // Integrated Information Theory — simple neural explanation
Alright, let’s imagine consciousness as the “experience of being something,” like the feeling of tasting chocolate, seeing a sunset, or even just feeling a slight ache in your toe. Integrated Information Theory (IIT) tries to explain where this “experience” comes from.
IIT’s core idea is that consciousness comes from “integrated information.” Let’s break that down:
Information: Think of “information” as the number of different things a system can be and the relationships between those things. A light switch has very little information; it can only be on or off. A smartphone has tons of information; it can be on, off, display millions of images, play music, connect to the internet, etc.
Integrated: “Integrated” means that all the parts of a system are connected and depend on each other. It’s not just a bunch of separate pieces doing their own thing; they’re all working together as a whole. Think of a car engine: all the parts are tightly connected and work together to create movement. If one part breaks down, the whole engine stops working.
So, what does “integrated information” mean together? IIT says that consciousness arises from a system that has a lot of information (can be in many different states) and where all that information is tightly integrated (all the parts depend on each other).
Analogy: A Simple vs. Complex Machine
A simple light switch: Has very little information (on or off) and is not very integrated. The switch being flipped doesn’t depend much on other things. IIT would say it has very little or no consciousness.
A smartphone: Has tons of information and is highly integrated. All the circuits are interconnected. IIT would say it has a much higher degree of consciousness (though not necessarily “human-like” consciousness).
Why integration matters: Imagine taking a smartphone apart and laying all the pieces on a table. Now you have all the parts but not the experience. The experience of being a smartphone comes from the parts being connected and working together in a specific way.
Important points in simple words:
IIT doesn’t say consciousness is just about “thinking,” it is about “being.”
IIT suggests that even simple things like a light switch have a tiny bit of consciousness, because of the way they are connected and process information.
IIT says the specific structure and connectivity of the system are key to consciousness. It favors sensory regions of the brain which IIT calls the “hot zone”.
In a nutshell: The more complex and connected a system is, the more “conscious” it is, according to IIT. Consciousness isn’t just about having a lot of parts; it’s about how those parts are connected and working together to create a single, integrated experience.
This conversation explores the potential for achieving artificial consciousness by integrating models of the human brain into AI systems. Ruifen Van Rullen discusses his work on incorporating the Global Workspace Theory of consciousness into AI, aiming to improve AI capabilities and flexibility, not primarily to create consciousness itself.
Van Rullen transitioned from neuroscience to AI due to advancements in deep learning. He emphasizes the importance of information processing, suggesting that the substrate (biological or silicon) is less critical. The Global Workspace Theory, where different brain regions share information on a “center stage,” inspires his approach. He connects various modules like vision and language to a workspace, enabling AI to react adaptively to different environments.
His team has created an AI system with a simplified global workspace, demonstrating it can transfer learned strategies to new domains better than AI without this architecture. While it’s early, this offers evidence supporting the theory’s relevance to advanced cognition.
Regarding the timeline, his current project is funded for five years to fully implement and assess the Global Workspace Theory. However, he acknowledges that consciousness in AI could potentially emerge sooner (even in the next GPT iteration), or much later if current consciousness theories are flawed. He acknowledges the ethical concerns and the potential risks from conscious AI that may arise inadvertently from commercial interests in AI research.
