NVIDIA's Distinction: Is there an AI bubble?

In The Theory of the Leisure Class, Thorstein Veblen coined a number of terms that have become terms of art: conspicuous consumption, conspicuous waste, pecuniary emulation, invidious comparison, and invidious distinction. He introduced the latter term as it pertained to different kinds of employment in a "higher barbarian culture":

...the distinction between exploit and drudgery is an invidious distinction between employments. Those employments which are to be classed as exploit are worthy, honourable, noble; other employments, which do not contain this element of exploit, and especially those which imply subservience or submission, are unworthy, debasing, ignoble.

According to its CEO, Jensen Huang, NVIDIA took its name from the Latin invidia, meaning envy. In a now legendary seven-page memo sent to Wall Street analysts, NVIDIA's investor relations team refuted claims that its "current situation is analogous to historical accounting frauds":

NVIDIA does not resemble historical accounting frauds because NVIDIA’s underlying business is economically sound, our reporting is complete and transparent, and we care about our reputation for integrity,” the memo said. “Unlike Enron, NVIDIA does not use Special Purpose Entities to hide debt and inflate revenue.

Coincidentally, NVIDIA replaced Enron on Standard and Poor's 500 stock index on November 28, 2001, as reported in the Los Angeles Times on November 30:

Enron, which had been a component of the S&P; 500, was removed at the end of trading Thursday and replaced by Nvidia Corp., a Santa Clara, Calif.-based maker of computer-graphics chips. As index funds sold Enron and bought Nvidia, Nvidia jumped $2.25 to $53.61 on Nasdaq.

Indeed, as the memo said, "NVIDIA does not use Special Purpose Entities to hide debt and inflate revenues." NVIDIA openly invests in companies that buy its products and reports that around 58% of its revenue are accounts receivable. I am not an accountant but that doesn't look like accounting fraud to me. It looks like something else. 

My interest in artificial intelligence has more to do with the noun than the adjective. Intelligence is a qualitative term that has been widely misunderstood and misused in the field of education. It all began with a "construct" that was devised to diagnose developmental/intellectual deficits in young children. This was, naturally, meant as a time-saving device since observation of the children would reveal more about the children's abilities but would be time consuming.

As so often happens, a short cut evolved into the standard of practice. Moderate correlations have been found between IQ scores and school grades and occupational success. But correlations are not proof of causation and even those findings may have been skewed by publication bias and poor research design. In other words, the validity of IQ tests as a predictor of performance is questionable. This is not to say that IQ scores are therefore inherently invalid. Only that their meaning and usefulness are not unambiguously clear.

When a term that is already ambiguous is used as a metaphor for another thing that is poorly defined, it descends into murkiness. Large language models perform some tasks that replace intellectual labour. That doesn't mean, however, that they do so intellectually or intelligently. A machine that replaces some functions of skilled labour doesn't thereby acquire any skill. It simply repeats certain motions that have been built into it. Large language models are, by definition, large. So there's that. But it is still just carrying out very complex instructions.

There are four glaring problems with the data center building boom. 1. So far, nobody is making money from selling their AI services. 2. The expectation is that AI will make money by replacing labour. I am not sure anybody has run through the economic demand side implications of that. 3. Energy and water. The U.S. is currently the world's largest producer of petroleum liquids, which depends on a delicate balance of low interest rates to finance investment and prices that are high enough to make it profitable but not so high as to crash the economy. Balancing data center energy requirements with energy supplies may have unfortunate knock-on effects for residential consumers. 4. Security. Protection against AI crime will be an afterthought, as it was with internet crime. I've read that adversarial poetry can defeat current LLM safety protocols. Who knew?

These problems are all in addition to the fact that attributing intelligence to large language models will not make it anything other than magical thinking. Veblen's invidious distinction between exploit and drudgery gives an important insight here. NVIDIA's $4.5 trillion capitalization is based on investors' perception that its future profitability will be based on exploit, not on drudgery. This is why its operations are viewed as "worthy, honourable, noble." NVIDIA's distinction is an invidious distinction.