AI washing: Silicon Valley’s big new lie

Start

“Can you go through all the old pitch decks and replace the word ‘crypto’ with ‘A.I.’?”

This caption, part of a New Yorker cartoon by Benjamin Schwartz, perfectly captures Silicon Valley’s new spirit of AI washing.

AI washing sounds like just another spin cycle, but it’s actually a complex and multifaceted phenomenon. And it’s important for everyone reading this column — technology leaders, marketers, product builders, users, and IT professionals of every stripe — to understand the exaggeration, warped emphases, and outright lying that we all encounter in not only marketing and sales, but also the stories we read based on industry claims.

Understanding AI washing

AI washing is a deceptive marketing practice that overemphasizes the role of artificial intelligence in the product or service being promoted. The phrase is based on “greenwashing,” coined by environmentalist Jay Westerveld in 1986, where consumer products are marketed as environmentally friendly regardless of environmental impact.

Products using old-school algorithms are labeled as “AI-powered,” taking advantage of the absence of a universally agreed upon definition for what AI is and what AI is not. Startups build apps that plug into a publicly available generative AI API and market it as an AI app. Big, bold AI projects that are supposed to showcase technology often rely on people working behind the scenes, because humans are the only way to make the ambitious AI solution work.

Let’s talk more about that last one.

AI: It’s made out of people

Retail giant Amazon rolled out 44 high-tech stores called Amazon Go and Amazon Fresh, which (starting in 2016) used the company’s “Just Walk Out” set of technologies. (I first told you about this initiative in 2017.)

Amazon’s vision: Stores where consumers could walk in, choose their items from shelves, then walk out without encountering a human behind a cash register. Sensors, including cameras, would feed into AI, which could figure out who bought what and charge accordingly — all without any checkout process. It felt like shoplifting, but legal.

The system was powered by advanced computer vision, which watched customers and what they picked up. Sensors in the shelves conveyed the weight of items removed, confirming the kind and number of items detected by the cameras. RFID tagged items also added information to the mix. Advanced machine learning algorithms processed the data from cameras and sensors to identify products and associate them with specific shoppers. Electronic entry and exit gates determined who was entering and leaving and when.

The algorithms were trained on millions of AI-generated images and videos to recognize products, human behavior, and human actions.

For seven years, Amazon has been eager to talk about these components of its Just Walk Out technologies. But the tech giant has been hesitant to discuss the 1,000 or so human beings hired to make it all actually more or less function — and admitted the existence of these employees only after press reports exposed them. Even then, Amazon has obscured the specific role these employees played, saying only that they didn’t review video.

Even with 1,000 employees monitoring and enabling 44 stores (checking three-quarters of orders, according to reports), the technology has been beset by problems, including delayed receipts, mismanaged orders, and high operational costs. 

This year, Amazon has been phasing out Just Walk Out technology from its main stores but still offers it as a service to other companies.

Another big example of humans behind the AI curtain is the world of self-driving cars.

Alphabet’s Waymo (the operation formerly known as Google’s self-driving car initiative) has a NASA-style command center where employees monitor cars through cameras and step in remotely when there’s a problem. (Here’s a fast-motion video I took recently of a ride through San Francisco in a Waymo car.)

General Motors’ Cruise subsidiary admits its self-driving taxis need human assistance on average every 4 to 5 miles, with each remote control session lasting an average of 3 seconds.

Other self-driving companies rely on remote human operators even more. In fact, a German company called Vay straight up uses human operators to drive the cars, but remotely. The company recently rolled out a valet parking service in Las Vegas. The car is remotely driven to you, and you drive it wherever you like. Upon reaching your destination, you just get out and a remote operator will park it for you.

Amazon’s stores and self-driving cars are just two available examples of a phenomenon that’s widespread.

Why AI washing happens

The high-level, high-paid technologists building AI systems believe in AI, and believe it can solve extremely complex problems. Which it can — theoretically. They tell their superiors it can be done. Those leaders tell their board it can be done. Company C-suites tell investors it can be done. And as a company, they tell the public it can be done.

There’s just one small problem: It can’t be done.

Most companies feel some sense of accountability for lofty claims, and so they hide the degree to which the product or service depends on humans behind the curtain making decisions, working through problems, and enabling the “magic” to take place.

The more shameless companies remain undeterred by proof that their AI isn’t quite as capable as they claimed or believed, so they just re-up their claims again and again. Tesla CEO Elon Musk comes to mind.

In October 2016, Musk said Tesla would demonstrate a fully autonomous drive from Los Angeles to New York by the end of 2017.

By April 2017, he predicted that in about two years, drivers would be able to sleep in their vehicle while it drove itself.

In 2018, Musk moved his promise of full Tesla self-driving to be by the end of 2019.

In February 2019, Musk promised full self-driving “this year.”

In 2020, Musk claimed that Tesla would have over 1 million self-driving robotaxis on the road by the end of the year.

Even this year, Musk claimed full self-driving Teslas might happen “later this year.”

It’s not going to happen. Musk is deluding himself and his customers. Musk is the Mr. Clean of AI washing.

The real problem with AI washing

The cumulative effect of AI washing is that it leads both the public and the technology industry astray. It fuels the delusion that AI can do things it cannot do. It makes people think AI is some kind of all-purpose solution to every problem — or a slippery slope into dystopia, depending on one’s worldview.

AI washing incentivizes inferior solutions, focusing on “magic” rather than quality. Claims that your dog-washing hose is “powered by AI” doesn’t mean you end up with a cleaner dog. It just means you have an overpriced hose.

AI washing warps funding. Silicon Valley investment nowadays is totally captured by both actual AI and AI-washing solutions. Even savvy investors may overlook AI-washing exaggeration and lies knowing that the AI story will sell in the marketplace thanks to buyer naiveté.

The biggest problem, however, is not delusional selling by the industry, but self-delusion. Purveyors of AI solutions believe that human help is a badge of shame, when in fact I think human involvement would be received with relief.

People actually want humans involved in their shopping and driving experience.

What we need is more human and less machine. As we speak, AI-generated garbage is flooding the zone with cringy prose and falsehoods, along with weird, sometimes horrifying, images. Google is so eager to replace its search engine with an answer engine that we end up with glue on our pizza.

What the public really wants is a search engine that will point us to human-created content or, at least, a PageRank system that favors the human and labels the AI-generated.

The AI-washing phenomenon is built on delusion. It’s built on the delusion that people want machines creating and controlling everything, which they don’t. It’s based on the delusion that adding AI to something automatically improves it, which it doesn’t. And it’s based on the delusion that employing people represents a failure of technology, which it doesn’t.

Enough delusional AI washing already! Sellers need to tell the truth about AI. And buyers need to demand proof that any AI in the products and services we pay for actually does something useful.

I think I speak for all of us in the technology industry, the technology customer community, and the tech press when I say to Silicon Valley: Stop gaslighting everybody about AI.

Previous Story

Efficient Text and Chat Data Discovery Starts with Information Governance (IG)

Next Story

Will your business apps run on the latest Copilot+ PCs?