AI-Profits Drought and Historical Lessons

In a 1987 article Times Book ReviewRobert Solow, an economist at MIT winning Nobel, commented: “In productivity statistics, you can see the computer age anywhere.” Despite the increase in computing power and the popularity of personal computers, government data shows that the full output of each worker, a key determinant of wages and living standards, has stalled for more than a decade. As we all know, the “productivity paradox” continued until the 190s and beyond, resulting in a huge and undecisive literary system. Some economists accuse new technologies of mismanagement. Others believe that the economic significance of computers is pale compared to old inventions such as steam engines and electricity. There are also some who blame the measurement errors in the data and believe that once these errors are corrected, the paradox will disappear.

Solow’s articles for nearly four decades, nearly three years since Openai released its Chatgpt Chatbot, we may face a new economic paradox that involves generated artificial intelligence. According to a recent survey conducted by economists at Stanford University, Clemson and World Bank, in June and July, using AI tools, almost half of the workers (45.6% to be exact). However, a new study by a team of researchers linked to the MIT Media Lab says: “Despite $3 billion investing in Genai’s businesses, the report found a surprising result that 95% of organizations are experiencing zero returns.”

The authors of the study examined more than three hundred public AI programs and announcements and interviewed more than fifty company executives. They define successful AI investments as investments that have been deployed externally during the pilot phase and have produced some measurable financial returns or significant growth in productivity six months later. They wrote: “Only 5% of AI pilots extract millions of dollars in value, and the vast majority of pilots are still trapped in the absence of measurable damage.”

The investigation interviews elicited a series of answers, some of which were highly suspicious. “The hype for LinkedIn said everything has changed, but in our operations, there is no basic change,” the chief operating officer of the mid-sized manufacturing company told the researchers. “We are working on some contracts faster, but it’s all changed.” Another respondent commented: “We’ve seen dozens of demos this year. Maybe one or two are really useful. The rest are wrappers or science projects.”

To be sure, the report notes that some companies have successfully invested in AI. For example, it highlights the efficiency created by customized tools designed to be designed for back-end operations, noting that “these early results suggest that even without significant organizational reorganizations, learning-capable systems can bring real value.” The survey also cites some company reports that “companies that improve customer retention and sales conversion through automated promotion and intelligent follow-up systems”, suggesting that AI systems may be useful for marketing.

However, many companies are working to realize the idea of ​​huge returns from another recent survey by multinational Akkodis. After contacting more than two thousand business executives, the company found that the percentage of CEOs in the company’s AI-Implantation strategy fell from 82% in 2024 to 45% this year. The confidence of the company’s CTO has also been reduced, although not that much. The Akkodis investigation said the developments “may reflect disappointing results of previous delays or failures in digital or AI programs, implementation, and concerns about scalability.”

Last week, media coverage of MIT Media Lab Research coincided with high-value stocks related to AI, including Nvidia, Meta and Palantir. Of course, correlation is not causality, and the latest comment from Openai CEO Sam Altman may have played a bigger role in the sell-off, which is certainly inevitable at some point given the recent price increase. During a dinner with reporters, Altman said the valuation was “crazy” and used the term “bubble” in fifteen seconds, CNBC reported.

Still, the MIT study has attracted a lot of attention, and after the initial news coverage of the study, there is a report that media labs with links to many technology companies are quietly restricting access to it. The messages I left behind in the communications office of the organization and the two authors of the report were not blamed.

While the report is more subtle than some news reports, it certainly raises questions about the grand economic narrative of the technological boom since the November 2022 release of Chatgpt. A short version of this narrative is that the entire economic spread of generating AI is for workers, especially knowledge workers, but is very harmful to companies and their shareholders, as it will make a huge leap in productivity and expand profits.

This does not seem to have a possible reason for this, but recalls the suggestion that management failures limited the productivity benefits of computers in the 190s and early nineties. Media Lab research has found that some of the most successful AI investments are made by startups that use highly customized tools in narrow workflow processes. On the other end of the “Genai Divide,” the study noted that “building common tools or trying to develop capabilities internally” was not very successful. The split between success and failure “seems not driven by model quality or regulations, but rather by the methodology,” the report said.

It is conceivable that the novelty and complexity of generating AI may prevent some companies from retreating. A recent study by consulting firm Gartner found that less than half of CEOs believe their CIO is “proficient in AI.” But there is another possible explanation for the disappointing record highlighted in the Media Lab report: For many established businesses, the generated AI, at least in the current avatar, is not all at all. “This is very useful for brainstorming and first drafts, but does not retain knowledge about client preferences or learn from previous edits,” said one respondent from the Media Lab survey. “It repeats the same errors and requires extensive contextual input for each session. For high-risk work, I need a system that accumulates knowledge and improves over time.”

Of course, there are many people who think AI is useful, and there is academic evidence to support this: In 2023, two MIT economists found that contact with ChatGPT allowed participants to complete “professional writing tasks” faster in randomized trials and improved the quality of their writing. In the same year, other research teams identified productivity-enhancing outcomes for computer programmers using Github’s Copilot and customer support agents with access to proprietary AI tools. Researchers at the Media Lab found that many workers use their personal tools at work, such as GPT or Claude. The report calls the phenomenon a “shadow AI economy” and comments that “it is generally better than employer initiatives.” But the problem remains, which senior company executives will certainly ask more frequently: Why aren’t more companies seeing these types of benefits stemming from the bottom line?

Part of the problem may be that the generated AI (though that) has limited applications in many places in the economy. Co-employed about 50 million Americans in leisure and hospitality, retail, construction, real estate, real estate and nursing fields (child attitudes and care people), but they don’t look like direct AI transformation candidates.

Another important thing to note is that adopting AI throughout the economy is likely to be a long process. In Silicon Valley, people like to move quickly and destroy things. But economic history tells us that even the most transformative technologies economists call universal technology cannot leverage the maximum effect until the infrastructure, skills and infrastructure, skills and products that can complement them. This can be a long process. Scottish inventor James Watt invented his cylindrical steam engine in 1769. Thirty years later, most cotton factories in the UK are still powered by water wheels, partly because it is difficult to transport coal in a steam engine. It was not until the early nineteenth century that the development of steam-powered railways changed. Electricity will also spread slowly, without immediately leading to a rapid economic growth in productivity. As Solow pointed out, computer development follows the same pattern. (From 1996 to 2003, the growth of overall economic productivity eventually increased, and many economists attributed the delay effect of information technology.