AI washing: how to avoid the hype and deception

Posted: July 19, 2024

This March, the U.S. Securities and Exchange commission settled charges against two investment advisors for “making false and misleading statements about their purported use of artificial intelligence (AI).”

SEC Chair Gary Gensler said, “We’ve seen time and again that when new technologies come along, they can create buzz from investors as well as false claims by those purporting to use those new technologies. Investment advisers should not mislead the public by saying they are using an AI model when they are not.”

The case against these companies was straightforward. One company claimed it was feeding data from its clients’ social media and banking accounts into an AI to optimize their investment portfolios. But, the company never used any of this data in any way—AI or no AI.

Companies have a lot of incentive to exaggerate how they use AI, a phenomenon known as AI washing. Nearly half of all start-up funding in the U.S. went to AI startups from April through June of this year.

What counts as AI?

But determining what counts as AI washing is difficult, partially because there’s no commonly accepted notion of what exactly AI is or isn’t. Industry has been using technology that people now talk about as “AI” with great success for years. In particular, a variety of techniques for calculating statistics, known collectively as “machine learning,” have been helping companies operate more efficiently by doing things like predictive analytics, scheduling and process optimization, and root-cause analysis. Generally, it’s these machine learning techniques that are now grouped under the general umbrella term “AI.”

Several recent developments have made these older techniques more powerful and applicable to more cases. In particular, a new kind of architecture invented in 2017, called a transformer model, is good at calculating how different items in a sequence—like words, for example—correlate with one another. It’s this technology that has made ChatGPT and other large language models (LLMs) so successful in recent years.

How to avoid AI-washing

Despite these successes, the Head of Global Equity Research at Goldman Sachs, Jim Covello, warns in a June 25 report that we are likely in an AI bubble:

“We estimate that the AI infrastructure buildout will cost over $1Tn in the next several years alone, which includes spending on data centers, utilities, and applications. So, the crucial question is: What $1Tn problem will AI solve?”

While skeptics like Covello worry that companies are investing more in AI than it can pay out, others, like veteran tech investor Roger McNamee, worry that companies are using machine learning products and LLMs in ways that don’t add any value at all. Sure, you can say that such products count as AI—but if the AI doesn’t add any value, they’re engaging in AI washing.

It’s now easy enough for companies to just add a pre-made chatbot interface onto an otherwise standard product without making the product any better. They can also use complex machine learning computations to perform tasks that could be done just as well with less power, less data—and with less cost. 

What’s important is not whether something is called “AI,” or even whether it has a particular computational architecture, but whether that computing power is doing something valuable you couldn’t do otherwise. Time will tell whether or not AI will solve any $1Tn problems. In the meantime, companies should demand a real-world demonstration of how products that claim to use AI will solve their own individual problems before they invest in these products—just as they would with any other new technology.