Not this month, maybe not this year, but in the not-too-distant future the bubble in artificial intelligence (AI) investing will pop.
AI itself will be fine. The replication of human intelligence in machines that think and act like humans is already integral to the way we live.
Prime Minister Mark Carney expects to use AI to enhance productivity in the federal civil service.
Ray Chun, incoming CEO at Toronto-Dominion Bank, said last week that he is counting on AI to produce about $500 million in annual revenue gains and reduce costs by the same amount.
And the 700 million weekly users of OpenAI’s chatbot ChatGPT are using it to create dinner recipes, plan road trips and workout regimens, and for short online answers in internet searches.
Where AI appears to be setting itself up for a fall is the stupendous amounts of money invested in it. The industry expects to spend about $3 trillion (U.S.) by 2028 on new-product development, advanced semiconductors and data centres.
Meta Platforms (Facebook, Instagram) plans an enormous data centre in Louisiana that it says will eventually come close to the size of Manhattan.
Yet the payoff from AI bets is anything but certain.
“Are we in a phase where investors as a whole are overexcited about AI? In my opinion, yes,” Sam Altman, CEO of OpenAI, said in August.
When investors wake up to how the sky-high valuations of AI companies is not matched by hyped productivity and profitability gains for AI customers, revenues and valuations of AI developers Meta, Amazon, Microsoft and Alphabet (Google) and the many AI startups will drop.
But, as in previous bubbles, while the party is underway spending is no object for AI companies.
Mark Zuckerberg, CEO of Meta, said last month that he prefers “misspending a couple of hundred billion dollars” than to fall behind in the race for AI dominance.
AI is the latest investing megatrend.
AI semiconductor maker Nvidia Corp., with a market cap of $4.5 trillion, is the world’s most valuable company.
Three-year-old OpenAI has an estimated value of $500 billion. OpenAI anticipates revenues this year of just $13 billion.
Anthropic, maker of the Claude chatbot, boasts a valuation of $183 billion, but says it will post revenues of just $2 billion to $4 billion this year.
Investors in the AI boom must hope that their entrants in the race include the next Amazon or Google, among the few major survivors of the dot-com bust of 2000-01.
The rest of us must hope that an AI meltdown has minimal impact on the real economy. The dot-com crash caused a recession and a collapse of the S&P 500 by nearly 50 per cent in 2000-2002.
Too many AI applications have yielded underwhelming benefits.
A report this year by S&P Global found that 42 per cent of companies it surveyed that tried AI pilot projects had abandoned most of them.
And researchers at the Massachusetts Institute of Technology reported in August that 95 per cent of the organizations they studied had achieved no return on their AI investments.
The work done by today’s AI tools often requires human beings to correct it. As AI has been pushed to do more complex tasks, it has become more prone to “hallucinations,” or simply making stuff up.
Researchers at Harvard University and Stanford University said this year that many users of AI products for which corporations have paid top dollar produce “workslop” that reduces productivity.
As reported by Bloomberg, the researchers define workslop as “AI generated work content that masquerades as good work but lacks the substance to meaningfully advance a given task.”
Improvements in AI products might narrow the gap between high expectations and disappointing performance.
AI developers are at work on a new generation of “superintelligent” PhD-level AI assistants.
Indeed, the AI market is priced in anticipation of “artificial general intelligence” (AGI), a true replication of the human brain.
But the Holy Grail of AGI has been promised and has failed to materialize since at least the 1970s. AGI’s development costs would be far higher than the fantastic sums now spent on basic AI.
It’s not a given that new AI products will improve as rapidly as what came before. And they are more expensive and take longer to develop than their predecessors.
OpenAI’s GPT-5, the firm’s latest chatbot released in August, took about three years to develop. Reviewers say it is only a modest improvement over previous models.
As AI becomes more robust and users’ expectations of it come down from the clouds, AI’s reliability issues will fade.
But not so the economic and moral issues around AI, including AI’s strain on electric power grids and the potential for massive job loss and anti-social AI capabilities.
It’s too late to put that genie back in the bottle, but not too late to approach AI investments with extreme caution.
