Liability, not regulation.

Free market economics is near and dear to my heart. Essential to that economic model, however, is clear and accurate information. Fraud kills free enterprise nearly as fast as government (in part because it invites government action). False advertising, scamming, and all the rest in the vast playbook of mean deceit, they are all inimical to proper stewardship and to the economic processes which foster that. Unfortunately, ‘Artificial Intelligence’ is right now perpetrating a massive and consistent fraud.

The Fraud

I’ve discussed it before, but ‘AI,’ neural net LLMs in the less common parlance, doesn’t analyze and output information in anything like the human mode. One way to understand this is to ask you to do what the AI does with a text chunk. Follow these steps closely.

1. Your sentence to analyze is, “The lazy cow jumped across a big chasm.” Please provide a paraphrase of the sentence.
2. Now, turn that sentence into a number, a binary code.
3. Run that code through a series of calculations.
   1. Along the way, introduce some extra numbers from your big box of pseudo-random numbers. (This prevents you from getting the same answer every time on more complex questions.)
   2. These calculations are preset; they were established by a series of training sessions in which you tweaked them until the end result was consistently similar to the desired output.
4. Take the consequent binary number and convert it back into letter symbols.
5. Voila!~

Note that nowhere in this process was the information conceptualized. The machine does not consider ‘cow’ or ‘lazy’ or combine them into ‘lazy cow.’ It simply renders the text into a string of numbers, undifferentiated on a conceptual level, and runs it through a preset mathematical formula of incredible complexity (with some RNG to prevent staleness). The AI never ‘knows’ something.

More sophisticated ‘paraphase’ systems, likely, will scan for other sources. But the fundamental process of synthesizing this information is as represented above. The AI literally does not process concepts, and so it has no reason to be accurate in what its output indicates on a conceptual level, except that the calculations are statistically likely to sound probable. AI does lie- because it doesn’t tell the truth. It does present accurate and inaccurate information with the same absolute lack of comprehension; the only reason to trust it is that it’s generally right.

The Danger

Of course, even an apparently well-intentioned or neutral AI can get stuff wrong. I once saw a summary of one of my articles pop up in the little AI answer bubble on my browser (I was looking my own article up to cite it). That summary was half-right and sounded plausible, but it misrepresented what I actually said, and quite substantially.

Worse, we have cases like discussed in this X article. AIs designed to persuade can just generate whatever string of binary is statistically likely to convince somebody (the goal the AI has been trained towards), once that binary is converted into a letter-and-punctuation representation.

Not Regulation; Liability

The X article I cited above was the reason I started writing this article. How do we answer a potential propaganda tool like this? On a personal level, of course, rigorous logic, self-control over emotion, and fact-checking (with reliance on trusted, personal sources) are a good answer. But on a societal level, the study referenced poses a dire threat. How can we mitigate it?

I propose that every AI be required to indicate front-and-center, in every AI interaction, in a way improbable to ignore, that its AI generates false information.

This rule should not, however, be regulation. That tool of government simply invites more government control. No, the tool to force companies to follow this standard should be enhanced liability. They must be monetarily liable for every bit of damage their LLMs cause through presenting inaccurate information as true, unless they have clearly warned the user that the LLM provides false, hallucinated information. Punishment ,not regulation- and the rest, at least the significant players in the game, will self-regulate. And the punishment must loom over the government too, without any shield of sovereign immunity to protect it.

This warning label won’t fix everything; it won’t prevent everybody from being fooled. Some people will be fooled by something, honestly, and if it ain’t AI, it’ll be a scammer on Facebook or X or YouTube or in real life. But the label exerts a force against the abuse of AI and against the deleterious reliance on it which is rotting sections of our culture.

God bless.