David Chalmers identifies key functions for artificial consciousness
David Chalmers turned a 2022 NeurIPS talk into a paper submitted in March 2023. He asked how ideas from consciousness research apply to large language models.
He looked at whether a system can return to information it has already processed and share it across different parts of its work. He also asked what lets a system act as a whole. The paper describes limits in the models of the time and ways later designs could address them.
When an AI can look back at what it just did, it has a chance to notice a mistake before taking the next step. That is a concrete ability we can study and teach it to use well.
