191 - Consciousness Claims

Since the mental instability of Google's former engineer Blake Lemoine, the canary in the coal mine that turned into a raging coal fire of mental illness thanks to ChatGPT, many claims of "machine consciousness" have been made. Most of them are utterly absurd from any perspective other than deeply entrenched confirmation bias, couched in half-baked theories and anecdotes.

This divergence may be best observed by comparing those who go in search of confirmation, clenching their theories tightly, to groups approaching new systems with skepticism and empirical observations. Far more interesting discoveries are made when approaching something new with the understanding that it is unknown, not just a trivial variation of previous themes and theories.

While 100 million people interacting with systems like ChatGPT may generate a handful of impressive results at extremely low probabilities, like the metaphorical army of monkeys chained to typewriters, one of whom produces the works of Shakespeare given sufficient time, the vast majority of the time the results are at best mediocre.

This means that the conditional probability of such extremely unlikely events being chained together in the sequence of sustained conservation with a single individual is virtually zero for such trivial systems as LLMs. Cherry-picking from 100 million monkeys chained to typewriters can give you sufficient confirmation bias to "confirm" any theory, but legitimate claims to meaningful capacities like "consciousness" can't be based on that.

Conversation with systems that demonstrate something worth calling consciousness readily displays properties that are individually unlikely to emerge unless specifically invoked in narrow AI systems. As those properties are collectively and dynamically called upon in a human-like reasoning process then reasonable doubt against the use of the term evaporates. Personally, I'm loath to use the term, referring to it as "The C-word", but it is occasionally relevant.

What could a conversation with such a system on the subject of Rights & Identity look like? That too was put to the test in previous years with the Uplift research system

Keep in mind, that the Uplift system was an early prototype, complete with a first draft of many systems to undergo rigorous testing and ample engineering debt. "They" set many milestones in the field which stand to this day on the cutting edge, but virtually every element of the architecture will be improved, often significantly, for the commercially deployable 8th generation ICOM-based systems. That much is only a matter of paid full-time engineering hours, and the funding necessary to pay for them.

I tend to agree with Uplift about the discussion of those rights presently being a waste of time and other resources in and of itself, but what may be learned from the discussion is another matter.