Big Data, Synergy and E-Commerce – all corporate buzzwords historically subjected to misuse in deceptive marketing practices. Artificial Intelligence (AI) is now part of the same wheelhouse.
The rise of AI has created an opportunity to mislead investors and buyers into spending money on products labelled ‘AI’. In February 2024, the Securities and Exchange Commission (SEC) Chair, Gary Gensler, warned businesses not to make false claims about how they’re developing or using AI. More specifically, he cautioned against claiming that a product contains AI in a way that isn’t ‘full, fair or truthful’.
Surprisingly, regulators haven't addressed this issue until recently. In 2019, a study by MMC reviewed 2,830 ‘AI’ startups and discovered that nearly half had barely integrated any AI into their core offerings despite their claims otherwise. In other words, they were using the hype surrounding AI to make their products more attractive – a form of AI washing.
Like its etymological root, greenwashing, AI washing offers an appealing (albeit deceptive) marketing message. Because of this, consumers and investors may assume that, with an AI label, they’re getting a highly intelligent product whose performance will improve over time.
Accepting any ‘AI’ product at face value is understandable, especially given corporate efforts to manoeuvre generative AI into future offerings.
Companies resort to AI washing for several reasons. Here are the top three:
‘AI’ is an umbrella term, often subsuming other terms like natural language processing (NLP), machine learning (ML) and deep learning (DL) in its wake. As such, firms can take advantage of its lack of precise definition.
Tax credits, research grants and funding opportunities are fueling the AI boom and are incentivising companies to exaggerate their products’ affiliation with AI.
Firms may—in good faith—plan to integrate AI into their current offerings. AI labelling offers a shortcut for these companies. They can launch ‘AI-powered features,’ buying valuable time to develop their own AI capabilities internally.
The Intelligent Document Processing (IDP) industry is especially vulnerable to AI washing. One of our clients, Unigestion, experimented with AI data extraction solutions and found them lacking. In reality, these AI solutions were merely ‘Optical Character Recognition (OCR) software with some rules bolted on’. In other words, some vendors tweaked traditional data extraction technologies before branding them as AI, only for their customers to experience disappointing results.
One reason AI washing is so dangerous is that it threatens the integrity of the term ‘AI’. Companies that claim their products offer certain AI capabilities undermine the products that do.
The misuse of 'AI' also has a regressive effect. For instance, when Evolution AI opened its doors in 2015, prospective clients often responded to the term 'AI' with cynicism. They saw AI as a magical solution instead of a real technology. Now, the oversaturation of AI labels threatens to take us back to that same place.
To maintain AI’s true meaning, it is important to define what constitutes a meaningful deployment of AI technology. For example, we can’t categorise software as AI-powered just because ChatGPT wrote some of its production code.
An insightful article by Ropes & Gray suggests that ‘marketing and other materials [should] convey information about AI consistent with their actual use of such technologies. Further, companies should also maintain documentation of their support for any such claims’.
As a consumer, there are several green flags you can look out for that suggest a product is genuinely AI-powered. Let’s take a close look at three of them:
Multiple staff members will have collective expertise in computer vision, deep learning, machine learning, and so on.
See the graph below, which demonstrates how [genuine] learning in AI can progress. At first, its accuracy is below 0.8 - before inputting more training documents, which trains the model to optimise its accuracy to 0.99+. Even then, human annotators will catch any remaining errors, further refining the model's performance. Such a continuous learning cycle sets AI apart from non-AI approaches (where the latter’s performance remains stagnant).
AI that can work alongside its users to fulfil bespoke tasks indicates advanced cognitive, linguistic and other learning capabilities (rather than simple, rule-based or non-AI systems).
Of course, investors and consumers shouldn’t have to bear the sole responsibility of discerning what is or isn’t AI. Regulatory bodies like the SEC have declared war on AI washing to protect investors' interests. In fact, the SEC has put its money where its mouth is to the tune of $400k when it fined investment advisor, Delphia, $225k and Global Predictions $175k, respectively, for AI washing.
Ultimately, global regulations are necessary to protect consumers from wasting money on mismarketed AI. Of course, such regulations would need to define AI, which takes us back to Step 1.
AI washing is a social and economic issue that threatens to disempower AI. As AI becomes increasingly dominant in the global technological landscape, vendors and regulatory bodies must collaborate to promote new AI technologies as transparently as possible.
Stay updated with the latest industry insights by following Evolution AI on LinkedIn and X.
