Is it possible for an AI model to develop self-awareness similar to human consciousness? A team of researchers developed a list of testable criteria to find out.
Since well before ChatGPT became part of the mainstream, researchers, fiction writers, and filmmakers alike have asked the question of whether artificial intelligence could ever become “conscious”—and how we would know if it did. Recently, a group of 19 computer scientists, neuroscientists, and philosophers set out to explore this question by developing a checklist for determining if an AI model is conscious. While this is not yet a full AI consciousness test, it does outline several different ways of approaching the question, “how do we know if AI is self-aware?”
To build the checklist, the research team identified 14 key criteria based on different theories of consciousness. They then developed specific ways to test AI models against this checklist. The researchers emphasized that the checklist is designed to prompt more questions and offer a starting point for conversations about what consciousness might mean when talking about AI models.
What Does it Mean for AI to “Become Conscious?”
To be able to test an AI model for consciousness, we first need a clear definition of “consciousness.” The research team started by generating specific characteristics from a variety of current philosophies of consciousness, including:
- Recurrent Processing Theory – This theory suggests that consciousness comes from our brains putting experiences through “feedback loops,” using prior knowledge and connections to make sense of our current experience.
- Global Workspace Theory – This theory helps explain how our brains coordinate and process the many streams of information running through our head at any given time. In this theory, consciousness is defined as being like a mental stage manager or spotlight that decides what gets our attention and what doesn’t.
- Higher Order Theories – This is a group of theories that argue consciousness is the result of being aware of our thoughts and sensory experiences as they happen. Consciousness here is defined as being able to “think about thinking.”
- Attention Schema Theory – This theory explains consciousness as a result of the brain’s ability to direct our attention to specific objects, thoughts, memories, and other stimuli while filtering out others. The ability to be aware of how and where our attention is being directed is a key element of this theory.
Additionally, the researchers included criteria based around predictive processing, the brain’s ability to accurately predict and account for the world around you based on past experience. This is an especially important component of AI models designed to generate creative content or solve complex problems.
The research team also included criteria to evaluate AI based on agency—the ability to make conscious decisions to act—and embodiment, either in physical space or relative to other virtual systems.
Evaluating Current AI Models for 14 Characteristics of Consciousness
Based on the above theories, the research team came up with a list of 14 characteristics that indicate consciousness. They argue that the more of these characteristics an AI model shows, the higher the possibility that it is conscious. When they tested a number of different current AI models against their checklist, the researchers found that none came anywhere close to meeting all 14 criteria. Only a few managed to check more than a handful of boxes.
One of the researchers, Eric Elmoznino, gave one possible explanation: Different AI models fulfilled certain criteria and not others based on what they were originally designed to do. For example, many of the AI programs designed to generate images based on a prompt fulfilled some of the checklist criteria in the “recurrent processing” category. This makes sense, because these models need to be able to simulate objects and art styles based on many pre-existing examples.
It’s also important to note that different AI models are built using different algorithms and formulas to simulate both learning (how they gather and synthesize information) and communication (how they relate that information back to human users based on prompts). This means that different versions of AI—whether chatbots like ChatGPT or applications like AI virtual assistants—may rank differently for different criteria in the checklist.
Does Testing AI for Consciousness Help Us Learn More About the Brain?
The researchers’ AI consciousness checklist is mostly just a thought experiment for now. However, AI models already have a role in brain disease research and have been used to help develop new technology that can aid in the treatment of many different diseases.
Below are a few ways AI technology and machine learning are already helping researchers better understand, diagnose, and treat brain diseases and disorders.
Brain Imaging and Mapping
AI models can review huge amounts of data much faster than human researchers. When combined with imaging technology like EEG, MRI, or CT scans, AI and machine learning can give scientists a better understanding of how different parts of the brain work together and how complex tasks are coordinated across the whole brain.
Neuromodulation Devices
AI may play a growing role in neuromodulation devices and wearable technology. An AI model that can monitor subtle changes in brain chemistry or activity can help calibrate assistive devices much quicker and more accurately and can respond to factors human users may not even notice. For example, AI models may be able to detect an oncoming seizure in a person with epilepsy based on electrical activity in the brain.
Diagnosis
AI learning models can help doctors identify complex patterns in test results that can aid in diagnosis. Researchers have already successfully used AI to help spot the presence of disease in blood and tissue samples, recognize early signs of dementia, and diagnose Alzheimer’s with a single MRI scan.
The group of researchers who created the AI consciousness checklist released a pre-publication version of the paper with their full research and findings, which you can read here.
Want to learn more about how the latest technology is helping researchers diagnose and treat brain disease? Register for our free expert-led webinars and view previous webinars to hear directly from leading neurologists on the latest advancements in brain disease research.
Stay updated on the latest news from the American Brain Foundation by following us on Twitter and Facebook. Only through research will we find cures for all brain diseases and disorders. Donate today to make a difference.