The question that arises is how to verify the perceptible awareness of an algorithm? Algorithms, unlike the human brain, show no detectable internal function on an EEG or MRI. For this reason, another method is needed. The researchers first examined the current theories about human self-awareness to arrive at the main descriptors of the state of consciousness, and then examined them in the architecture of artificial intelligence.
Consequently, a theory must be based on neuroscience and empirical evidence such as brain data scans. Researchers manipulate awareness with these data and perceptual tricks; It is therefore possible that consciousness will arise independently of computations performed by biological neurons or silicon chips.
The researchers finally came up with six theories. One of the theories known as feedback processing shows that the transfer of information through feedback loops is one of the key factors of consciousness. Another theory called global neural workspace claims that consciousness arises when independent streams of information are combined by passing through a narrow path in a workspace similar to a computer clipboard.
Another theory known as higher order suggests that consciousness includes the process of representing and interpreting primary data received from the senses. Other theories emphasize the importance of the mechanisms necessary to control attention and the need to receive feedback from the outside world. Based on six existing theories, the researchers extracted 14 indicators for the state of consciousness.
According to the reasoning of the researchers, the more the artificial intelligence architecture corresponds to a greater number of indicators, the more likely it is to have awareness. Machine learning expert Eric Elmozzino applied the list to several artificial intelligences with different architectures, such as the Dall-E2 image generator system. Many of the criteria were consistent with the indicators of the recurrent processing theory. A kind of big language model which is the basis of ChatGPT was compatible with the indicators of global work network theory.
Due to the fact that none of the artificial intelligences meet all the criteria, none of them can be considered a strong candidate for consciousness. The researchers are looking to design a broader awareness test that can be applied to micro-organisms, animals and newborn babies. They hope to publish this theory in the coming months. The problem with such projects is that current theories are based on our understanding of human consciousness. However, consciousness can take different forms even in other mammals.