051 - Research Failures

The award for this quarter's worst research paper goes to: Consciousness in Artificial Intelligence: Insights from the Science of Consciousness.

Unfortunately, despite having so many people claiming to be researchers, they failed to deliver in critical aspects. The set of criteria they named for AI Consciousness were all satisfied by one system, yet their conclusion failed to mention that system, and claimed that no system yet existed meeting their criteria. One of the better-known authors of this paper has also committed fraud in the process, as he is aware of the qualifying system.

Such robust incompetence is not excusable for an 88-page paper with many authors. Even though their selected set of theories had only partial overlap with those used to develop the qualifying system, and one theory used in the qualifying system was specifically excluded, full qualification remains.

The paper also ignores the human motivational system, despite citing some of the authors whose work clearly highlights the critical role it plays in human cognition and "consciousness" spanning a wide variety of definitions. Instead, they engage in another form of fraud, substituting the ML "attention" term in place of a motivational system, conflating the output with the process. ML "attention" also should not be confused with the Attention of Attention Schema Theory. This is both inexcusable for any large group of scientists, and very clearly biased in favor of confirmation bias to post-hoc justify current narrow AI research.

Attempting to link LLMs to Global Workspace Theory in particular is one of the worse instances of fraud I've seen this year. The two are fundamentally incompatible. Conflating Reinforcement Learning with conscious agency was another instance of blatant fraud, or severe mental illness. They similarly trivialize agency and embodiment for self-serving reasons.

Any academic journal foolish enough to publish this "paper" will take a crippling blow to their credibility.