The world of artificial intelligence is full of players, but not all of them are legitimate. Securities and Exchange Commission Chairman Gary Gensler explained that some people are practicing what he calls “AI washing.” on video This includes “false claims made to investors by those purporting to use those new technologies.”
In the financial sector, the SEC in March ordered two investment advisory firms, Delphia (USA) and Global Predictions, to issue a total of $40. He was fined $10,000. using this technique. (Both companies paid the fine and accepted the SEC's order without admitting or denying the commission's findings.)
This phenomenon is not only being seen in the financial services sector. Analysis firm FactSet examined S&P 500 companies' earnings conference records for the three months ending in mid-March and found that 179 companies used the term “AI,” compared to the five-year average of 73 companies. It turned out to be better than that. While it's unclear which of these companies are being dishonest about deepening AI technology in their operations, experts say adopting the acronym is becoming commonplace across the industry. Masu.
“In fact, incorporating AI into your slide decks, or even just using AI, is not that difficult because you don't have to integrate AI into your business and it's very easy to use from any platform.” said Managing Partner Michael Stewart. At Microsoft's venture arm M12, he focuses on AI, gaming, and deep technology. “But that doesn't provide a sustainable competitive advantage,” he added.
In reality, what AI washing results in is a breakdown in trust between vendors and their consumers, corporate partners, and investors.
“What's really at risk are the people on the back end who don't understand the technology,” said Timothy Bates, professor of innovation and technology practice at the University of Michigan-Flint, who has expertise in AI and other emerging technologies. says Mr. Technical Director at Lenovo and General Motors).
It's not just companies that claim to use AI when they don't have it. Bates also says that so-called button-pushing applications he calls AI-washing. AI learns by ingesting information and receiving various inputs or prompts. “You don't build a database by asking the same button-pushing questions over and over again,” Bates said, adding that third-party applications built on common natural language processing models are added that it is not effective.
For example, if a law firm purchases an AI assistant to replace a human, it will be functional within a few months unless it is trained specifically in the law and is based on its own unbiased database that is actively learning. It may disappear.
A “good” AI litmus test for enterprises
Toby Coulthard, chief product officer at Phrasee, which provides generative AI solutions for enterprise clients (including Sephora and Macy's), says he is wary of any business that uses the term AI broadly.
“AI is a very vague term,” Coulthard said. Instead, he says, companies need to define the type of AI they want to use and specify how they will employ it.
Additionally, Coulthard says it's important to look at when companies first started talking about using AI. “Knowing whether companies were talking about AI before ChatGPT is a good litmus test,” he said. And if they talk about what they won't do with AI and adopt some kind of ethical policy, that's also a positive sign, he says.
“The more we see companies talk about what they do and don't do with AI, the more accurate that will be,” Coulthard said.
Bates says to consider the models the company uses for its AI products. He said: “We are digging deep into the company itself to see if they have generated their own model. [or] Are you completely dependent on a third-party model?'' And if you are relying on a third-party model, you need to identify service level agreements or key performance indicators for the next one to two years, he said. I did. It's marketed as an AI company, so it needs to be maintained, monitored, and tuned to function. ”
Microsoft's VC approach to AI scrutiny
Stewart and M12's other AI-focused partners vet startups using what they call the four D's: data, dividends, distribution, and pleasure.
“If you, as a startup, don’t have access to your customers’ most important data related to how your AI works, other competitors may have access to the same data,” Stewart said.
For the dividend part, it helps identify whether the AI output is the work product itself. In other words, is it contributing to the bottom line? If possible, stakeholders can look at the gross margins of any type of AI business and find that one of the benefits of the technology is very high profit margins. For fully AI companies with limited human intervention, Stewart says gross profit margins of 80-90% are typical.
Although the distribution and pleasure elements of M12's investment analysis process are not directly related to AI wash, these elements play an important role in determining the sustainable lifespan of a startup (young businesses evolve as they mature). (I understand that there is a tendency to
With thousands of AI startups on the market, winners “need strong, stable distribution channels to cut through the noise and ensure they reach customers first,” says M12. says. On the other hand, “Creating a consistently enjoyable user experience is the key to creating an attachment that keeps users coming back for more.”
AI cleaning is just another form of jumping on the technology bandwagon. Unlike previous advances in AI, natural language processing technology is uniquely salable and marketable. Investors and customers therefore have a vested interest in the technology, making obfuscation more likely.
So far, the SEC has primarily focused on investment advisors and broker-dealers, but AI washing is a market-wide concern. “The last thing we want to do with this fund, and involve Microsoft, is use technology that falls outside the guardrails,” Stewart said. “If that's a sham and it's not the AI that's actually creating the magic and we've missed it, then that's something we need to go back and change the whole process of diligently working. .”
Already, M12 has written off the blackboard once in 2022 and developed entirely new guidelines for analyzing AI startups, after discovering that many computer vision companies are more superficial than they claim.
“The hype around AI can attract investors, drive up stock prices, and generate consumer interest, giving these companies a temporary advantage,” Bates said. Regulators are likely to step up oversight and enforcement, he added, as too many resources are being devoted to superficial claims about AI and not enough to concrete advances in the field.