How Can It Be Conscious If It Does Not Know of the World? Or The Press Sucks
So, a smaller, rambling entry today. This is, of course, not philosophical treatise — I am hardly qualified for that kind of rambling. But it is something that has been bothering me for a while.
Imitative AI systems do not have a model of the world. This is one of the reasons that they constantly hallucinate — they have no concept of reality, so they have no ability to know if the calculations they have made about what should be regurgitate for a given prompt are accurate or not. They have nothing to compare their output to, so their output can never be entirely accurate. This is one of the reasons they are bad at math and games: they have no map of the world, so they have nothing to base their calculations or moves on other than what is statistically likely based on their training data. This obviously does not work well, especially for things like chess, where you have to understand the context of the game to be successful. So given this fact, why, then, are we supposed to believe that imitative AI can be conscious?
Consciousness is often defined as an awareness of one’s self and the world around you. If an imitative AI has no model of the world, no awareness that something exists outside of its training data, no concept of the existence of the world, how can it be conscious? If it merely regurgitates what its math says is the best statistical fit based on training data, then how can it understand itself in relation to anything else? And I don’t think that training data counts as understanding the world. If it did, then the machines would not make so many mistakes based on a lack of real-world understanding. So why did people think they could be made conscious?
Part of the answer goes back to the stochastic parrot. That is a term defined by Emily Bender and her coauthors. To overly simplify, it means what we discussed above — these machines produce output but not meaning. The problem is that we associate consciousness with language, and so we assume that anything that can speak to us could have, at least, the potential for consciousness. But beyond this flaw in our perception, we were told that these models could become conscious, could be the first step to a truly general artificial intelligence. And that takes us to the press.
Our press has a both sides problem. This is a problem in political reporting, of course, but it exists across the news environment. Tech press, in particular, is bad about simply repeating whatever they are told by the last executive to speak to them. Not all of them, of course, but far too many. One of the reasons that imitative IA hype grew so far so fast was reporters repeating what executives from imitative AI companies told them with little to no context. It is easy for people to think that imitative AI can produce consciousness and intelligence when they are being told that is a reasonable possibility. And a press dependent upon access for its journalism and one that is dedicated to the concept of getting both sides of the story more than learning the truth is one where the shameless will have an advantage. And many of the imitative AI people are nothing if not shameless.
So, again, as I said, nothing revelatory in this piece. And I realize it is likely a simplistic overview of a more complicated conversation. But I do think that we would be better served if more of our media expressed actual skepticism, or at least a bias toward probabilities, when confronted with what business leaders or politicians tell them. It is okay to look askance at someone whose interests lie in pushing a specific interpretation and demand more of them. It is okay to not just repeat what both sides say but rather to attempt to find out the truth. It is okay to not take people at their words, even people who seem sincere. Truth matters and we shouldn’t be afraid to demand it.
From the late Robert McChesney (a generous professional to chat with, btw, even at the height of his career):
"Western Union was instrumental in revolutionizing journalism, the media system, and the broader political economy. It used its monopoly power [Ed: in telegraphy] to collaborate in the development of the Associated Press, a monopoly news service run in cooperative fashion by the largest newspaper publishers. This relationship was mostly unknown to the public. With
exclusive access to the wires—Western Union refused to let potential competitors use its
wires—AP became the only wire news service in the nation. So as not to offend any of
its thousands of clients, it encouraged a journalism that was seemingly nonpartisan—
hence it contributed heavily to the rise of journalistic "objectivity." Because newspapers
without access to the AP were at a decided competitive disadvantage, it also discour-
aged competition in local markets. Likewise, the AP had extraordinary influence in the
way it covered national politics because it served as the main voice for most major
newspapers. Needless to say, it invariably presented a voice that took the side of
business interests."
— "The Problem of the Media" (2004)
Per Bob, for the first 75 years of the US, journalism wouldn't have survived if it hadn't been both—highly partisan (to attract readers) and government subsidized. "Objectivity" was unheard of. To his dying day, he (and allies like John Nichols) advocated a return to that model.
Now, journalism students are so inculcated with the concept of nonpartisanship/faux objectivity that they can't envision doing their job any other way. And that's even before they get into the field and find that access depends on their falling into line.
The irony is that now things are so corporate controlled that AP is one of the few places you can count on getting facts, because they're structured so that no *one* corporation controls them.
Thank you for this #MediaLiteracy - themed, post, though. I'm enough of a media nerd to grok it. ;-)