AI Consciousness: An Exploration of Possibility, Theoretical Frameworks & Challenges

AI consciousness is a complex and fascinating concept that has captured the interest of researchers, scientists, philosophers, and the public. As AI continues to evolve, the question inevitably arises:

Can machines attain a level of consciousness comparable to human beings?

With the emergence of Large Language Models (LLMs) and Generative AI, the road to achieving the replication of human consciousness is also becoming possible.

Or is it?

A former Google AI engineer Blake Lemoine recently propagated the theory that Google’s language model LaMDA is sentient i.e., shows human-like consciousness during conversations. Since then, he has been fired and Google has called his claims “wholly unfounded”.

Given how rapidly technology is evolving, we may only be a few decades away from achieving AI consciousness. Theoretical frameworks such as Integrated Information Theory (IIT), Global Workspace Theory (GWT), and Artificial General Intelligence (AGI) provide a frame of reference for how AI consciousness can be achieved.

Before we explore these frameworks further, let’s try to understand consciousness.

What Is Consciousness?

Consciousness refers to awareness of sensory (vision, hearing, taste, touch, and smell) and psychological (thoughts, emotions, desires, beliefs) processes. 

However, the subtleties and intricacies of consciousness make it a complex, multi-faceted concept that remains enigmatic, despite exhaustive study in neuroscience, philosophy, and psychology.

David Chalmers, philosopher and cognitive scientist, mentions the complex phenomenon of consciousness as follows:

“There is nothing we know about more directly than consciousness, but it is far from clear how to reconcile it with everything else we know. Why does it exist? What does it do? How could it possibly arise from lumpy gray matter?”

It is important to note that consciousness is a subject of intense study in AI since AI plays a significant role in the exploration and understanding of consciousness. A simple search on Google Scholar returns about 2 million research papers, articles, thesis, conference papers, etc., on AI consciousness.

The Current State of AI: Non-conscious Entities

AI today has shown remarkable advancements in specific domains. AI models are extremely good at solving narrow problems, such as image classification, natural language processing, speech recognition, etc., but they don’t possess consciousness.

They lack subjective experience, self-consciousness, or an understanding of context beyond what they have been trained to process. They can manifest intelligent behavior without any sense of what these actions mean, which is entirely different from human consciousness.

However, researchers are trying to take a step towards a human-like mind by adding a memory aspect to neural networks. Researchers were able to develop a model that adapts to its environment by examining its own memories and learning from them.

Theoretical Frameworks for AI Consciousness

1. Integrated Information Theory (IIT)

Integrated Information Theory is a theoretical framework proposed by neuroscientist and psychiatrist Giulio Tononi to explain the nature of consciousness.

IIT suggests that any system, biological or artificial, that can integrate information to a high degree could be considered conscious. AI models are becoming more complex, with billions of parameters capable of processing and integrating large volumes of information. According to IIT, these systems may develop consciousness.

However, it's essential to consider that IIT is a theoretical framework, and there is still much debate about its validity and applicability to AI consciousness.

2. Global Workspace Theory (GWT)

Global Workspace Theory is a cognitive architecture and theory of consciousness developed by cognitive psychologist Bernard J. Baars. According to GWT, consciousness works much like a theater.

The “stage” of consciousness can only hold a limited amount of information at a given time, and this information is broadcast to a “global workspace” – a distributed network of unconscious processes or modules in the brain.

Applying GWT to AI suggests that, theoretically, if an AI were designed with a similar “global workspace,” it could be capable of a form of consciousness.

It doesn't necessarily mean the AI would experience consciousness as humans do. Still, it would have a process for selective attention and information integration, key elements of human consciousness.

3. Artificial General Intelligence (AGI)

Artificial General Intelligence is a type of AI that can understand, learn, and apply knowledge across a wide range of tasks, similar to a human being. AGI contrasts with Narrow AI systems, designed to perform specific tasks, like voice recognition or chess playing, that currently constitute the bulk of AI applications.

In terms of consciousness, AGI has been considered a prerequisite for manifesting consciousness in an artificial system. However, AI is not yet advanced enough to be considered as intelligent as humans.

Challenges in Achieving Artificial Consciousness

1. Computational Challenges

The Computational Theory of Mind (CTM) considers the human brain a physically implemented computational system. The proponents of this theory believe that to create a conscious entity, we need to develop a system with cognitive architectures similar to our brains.

But the human brain consists of 100 billion neurons, so replicating such a  complex system would require exhaustive computational resources. Moreover, understanding the dynamic nature of consciousness is beyond the boundaries of the current technological ecosystem.

Lastly, the roadmap to achieving AI consciousness will remain unclear even if we resolve the computational challenge. There are challenges to the epistemology of CTM, and this raises the question:

How are we so sure that human consciousness can be purely reduced to computational processes?

2. The Hard Problem of Consciousness

The “hard problem of consciousness” is an important issue in the study of consciousness, particularly when considering its replication in AI systems.

The hard problem signifies the subjective experience of consciousness, the qualia (phenomenal experience), or “what it is like” to have subjective experiences.

In the context of AI, the hard problem raises fundamental questions about whether it is possible to create machines that not only manifest intelligent behavior but also possess subjective awareness and consciousness.

Philosophers Nicholas Boltuc and Piotr Boltuc, while providing an analogy for the hard problem of consciousness in AI, say:

“AI could in principle replicate consciousness (H-consciousness) in its first-person form (as described by Chalmers in the hard problem of consciousness.) If we can understand first-person consciousness in clear terms, we can provide an algorithm for it; if we have such algorithm, in principle we can build it”

But the main problem is that we don’t clearly understand consciousness. Researchers say that our understanding and the literature built around consciousness are unsatisfactory.

3. Ethical Dilemma

Ethical considerations around AI consciousness add another layer of complexity and ambiguity to this ambitious quest. Artificial consciousness raises some ethical questions:

1. If an AI can understand, learn, and adapt to the extent of humans, should it be given rights?
2. If a conscious AI commits a crime, who is held accountable?
3. If a conscious AI is destroyed, is that considered damage to property or something similar to murder?

Progress in neuroscience and advances in machine learning algorithms can create the possibility of broader Artificial General Intelligence. Artificial consciousness, however, will remain an enigma and a subject of debate among researchers, tech leaders, and philosophers for some time. AI systems becoming conscious comes with various risks that must be thoroughly studied.

For more AI-related content, visit unite.ai.