AI companies are going on at length about their need for more data to train their AI models so they're more useful than a sack of bricks.
Let's clarify some points
The Internet (of which the World Wide Web is a subset) contains incredible amounts of data.
But there's a problem: the AI models have used it all up:
What's more, the bottleneck can't be solved by training AI models on other AI models:
Thus, Sam Altman and representatives like Nick Clegg now argue that AI companies need access to copyrighted works for no charge:
- Sam Altman says AI is over without copyrighted works
- Nick Clegg says AI is doomed without copyrighted works
But how do they know?
AI companies are drawing a statistical line in the sand based on an assumption that AI capabilities will keep improving based on the data they consume.
The logic works like this:
- My model started off dumb as a brick
- I gave it lots of data and it improved
- Therefore, to improve them more, I need more data.
In other words, because "B" follows "A", therefore "B" is caused by "A".
Faulty reasoning
This is what's known as a propositional fallacy.
Sure, there's some evidence to say that data maketh the AI.
But the companies are always making changes to the algorithms that underlie the systems, and other programming changes. How else do we keep getting new and improved AI models?
Plus, the companies are forced to add so-called "guardrails" to prevent the AI models returning blatantly incorrect, offensive and horrifying results.
Worse still, AI companies openly admit they don't understand how their AI models work.
- Sam Altman admits OpenAi doesn't know how ChatGPT works
- Can we trust AI if we don't know how it works?
So, how exactly will more data make a difference if the companies themselves don't know how their systems work?
Occam's razor
Carl Sagan once paraphrased Occam's razor by stating "...extraordinary claims require extraordinary evidence."
There is not a scrap of proof from any AI company or their representatives that their claims hold water. To the point government agencies are telling AI companies to tell the truth or face sanctions:
- AGI claims don't hold water
- FTC crackdown on deceptive AI claims
- FTC warns tech companies to keep AI claims in check - Archived because deleted from US Federal Government site.
What do AI companies get if copyright is overturned?
First, pretty much all the copyright infringement lawsuits will go away:
Second, they get lots of data that they allege will improve their AI models.
But there's another issue to consider
Show me the money!
AI companies need to make money to survive. But they're not.
Which means their investors, who have poured billions into these companies may not see a penny.
Delaying tactics or success? You be the judge!
Maybe more data will make AI models into electronic god like Sam Altman et al have predicted.
But if the extra data makes no difference, the whole thing collapses. Along with billions in investor funds.