Krystal & Ryan SPAR Over AI Hype w "Enshitification" Author

Doctorow asserts that the AI industry has invested $1.4 trillion globally, with $700 billion in the last year alone, yet generates only $50 billion in gross revenues. He highlights that these assets depreciate quickly (2-3 years) and that each new generation of AI technology is less profitable than the last, making it 'the money losingest thing the human race has ever done.'

The industry has spent $1.4 trillion, with $700 billion in the last year, but only has $50 billion in gross revenues. Assets depreciate every 2-3 years, meaning they need another $700 billion investment every few years, which 'doesn't pencil out' as unit economics worsen.

The speaker argues that the AI bubble is driven by established tech giants like Google, which have saturated their markets (e.g., 90% search market share) and need new narratives (like AI, metaverse, crypto) to convince Wall Street they are still growth stocks, not mature stocks. This allows them to maintain high valuations and use stock for acquisitions and talent.

Companies like Google, with 90% search market share, need to tell stories about growing into 'imaginary markets' like the metaverse, cryptocurrency, and now AI, to avoid being reclassified as mature stocks and suffering a 'massive haircut from the market.'

Doctorow posits that employers are highly motivated to invest in AI not necessarily because it genuinely replaces human jobs effectively, but because it promises to replace 'mouthy workers' with software that 'just does whatever it's told.' This allows bosses to regain perceived control over their workforce and reduce worker power.

Employers are 'extremely interested in replacing mouthy workers' who challenge their ideas with software that 'just does whatever it's told.' The AI sales pitch promises to 'wire the toy steering wheel into the car's drivetrain,' giving bosses a sense of control over their operations and workforce.

Countering claims of AI's emergent intelligence, Doctorow explains that AI models operate by measuring the frequency distribution of words and their relationships. He calls the term 'hallucination' a misnomer, stating it's simply a 'limit to statistical guessing,' not a sign of consciousness or reasoning. He acknowledges that AI has shown statistical guessing to be more powerful than previously thought, but this doesn't equate to human comprehension.

The idea that 'shoveling words into the word guessing program makes it and turn into God is is extremely magical thinking.' AI models measure 'frequency distribution of words and their relationship to other words and phrases.' 'Hallucination' is 'just a limit to statistical guessing,' not comprehension or reasoning.

While AI has shown impressive improvements with scale, Doctorow notes that 'the scale is breaking.' He explains that linear inputs no longer produce linear or better outputs; the returns on investment for adding more training data are decreasing, making each new foundation model less profitable.

It 'used to be that linear inputs to scale produce linear outputs to scale or even better than linear.' Now, 'that number is going down,' meaning 'we're having to put much more in' and 'each new foundation model is is less profitable than the previous ones.'