What is AI Washing? How bad is it? How can you prove it, and what can companies do about it?

“AI Washing” is a company’s marketing effort to advertise that their products or services contain artificial intelligence, even though this is only weakly the case, if at all.

“As AI accelerates up the Hype Cycle, many software providers are looking to stake their claim in the biggest gold rush in recent years,” says Jim Hare, research vice president at Gartner. “AI offers exciting opportunities, but unfortunately most vendors are focused on the goal of simply developing and marketing an AI-based product, rather than first identifying the need, potential uses and business value to customers.”

Gartner analysts see “AI Washing” as one of the main problems hindering the actual development and adoption of AI in enterprises. How can sales executives trust a vendor?

That is a fascinating topic requiring deeper research. We did the research. Unfortunately, we quickly realised that much of what has been written on the topic so far is very “wishy-washy”. Some articles by renowned editors have raised more questions for us than they have answered.

AI Washing – Easy to Identify, right?

Let’s start with an article from ComputerWeekly.co.uk about AI Washing. They write: “To determine whether a particular product contains AI or not, a good understanding of what artificial intelligence is essential.”

While this is correct, this is where the devil is in the detail: there is still no universally accepted definition of artificial intelligence! There are “attempts at definitions”, “descriptive approximations”, and “AI classifications”. In addition, technology is changing so fast that technologies called “AI” 50 years ago are no longer considered AI by AI specialists today. AI is dynamic.

Therefore, “AI Washing ” is not as easy to identify as “Green Washing”.

Therefore, we wanted to see what ComputerWeekly would write now. And well, they explained the AI classification into “strong” and “weak AI” in the next section. That’s good to know, but it’s not very helpful for detecting AI Washing. The rest of the article is more practical because it mentions questions you can ask providers to uncover “AI Washing”. One example is, “How does the product vendor define AI?” I am sure most vendors have a plausible answer to this question. But if there is no general definition of AI, how can a buyer judge the extent to which AI Washing is taking place? The other questions are also good but no less AI-philosophical.

Now, look at a worrying article on “AI Washing” by PwC, one global leader in the areas of auditing, tax consulting and management consulting, with concrete figures. Here it is written:

“It’s a common scenario. According to a study of 2,830 European startups by London-based MMC Ventures, 40 per cent of those that claimed to be “AI startups” had barely any AI at all.”

That sounds strong. Fortunately, PwC has provided a source for these figures, namely the “State of AI 2019” by MMC-Ventures. But here the figures are described as follows:

“We individually reviewed the activities, focus and funding of 2,830 purported AI startups in the 13 EU countries […]. In approximately 60% of the cases – 1,580 companies – there was evidence of AI material to a company’s value proposition.”

Wait a minute. There is a difference! It doesn’t say that 40% of companies don’t include AI in their products. It says that 60% of companies have concrete evidence of AI… in their own value proposition. That means that for the 40%, no evidence for or against AI has been found.

Watch out, it gets even better. Reading further the same paragraph in the State of AI 2019 report, it turns out: it’s not about AI Washing at all.

These figures above were simply used to analyse how many startups have artificial intelligence at the core of their value proposition. They divide AI start-ups into two “types”:
1. AI is at the core of the value proposition (60%) and
2. AI is not at the core of the value proposition (40%):

“Over time, the distinction between “AI companies” and other software providers will blur and then disappear as AI becomes ubiquitous. Today, however, it is possible to highlight a subset of early-stage software companies where AI is central to their value proposition.”

There is no mention of these 40 % of startups engaging in AI Washing in any way. The assertion in the PwC article is therefore taken out of context and obviously misinterpreted.

And now the biggest questions that were not answered in any of the “Many AI vendors practice AI Washing”-Articles: What procedures, what criteria and what AI definition are used to check “suspicious” companies for AI Washing? How can this be proven? Looking into their software code? And subsequently the question: how much AI is enough to be safe from accusations?

These questions interest us and to which we have not found an answer.

AI Washing: A Real-Life Example

The questions about identifying AI Washing companies are not intended to question the existence of AI Washing or even its danger. We merely want to take a critical look at the extent of AI Washing communicated in the media and are interested in the point at which an AI company can be classified as an “AI-Washing-Company”.

The fact that there are companies that want to ride the AI train without having any significant AI behind them shows this real-life example of the startup “Engineer.ai” (now builder.ai):

It claims to use artificial intelligence technologies to largely automate the development of mobile apps. However, several current and former employees say the company exaggerates its AI capabilities to attract customers and investors. The company claimed its AI tools are “human-assisted”, whethever that means.

Internal staff and employees know exactly how much artificial intelligence is in the product and “how much” intelligence is advertised. Engineer.ai was likely rightly categorised as “AI Washing” in this case, according to reliable sources.

But we doubt it is so easy for outsiders to prove AI Washing. Especially since AI allows some room for interpretation, proving it will probably end in long AI-philosophical discussions.

What can companies that want to introduce AI systems do concretely?


Dealing with AI Washing: There is a Consistent Advice

Even though we did not find a concrete approach and criteria in any article on how to identify “AI-washing-companies”, there is one unifying piece of advice:

Whether AI or non-AI, the focus should always be on solving a specific problem in the company. Let’s say you want to speed up a particular process, and you are looking for a technical solution to do so. Then you should first and foremost assess how well the software in question can solve your problem. Companies should avoid trying to do “anything with AI”.

Let’s take the example of “acting data-driven in sales”. You have a lot of unused sales data. You want to identify hidden sales opportunities and give your sales team concrete indications of churn risks, pricing potential or cross-selling opportunities. Then you have two options:

1. an AI-based predictive sales software that uses machine learning to discover opportunities.

2. or rule-based manual analysis using, for example, targeted filtering with Excel or even business intelligence solutions to discover leads yourself.

Option two is sufficient for a manageable data set. When should you avoid using machine learning? When the rule-based system is working great for your data set. Machine learning, an AI technique, will bring you the most on a huge data set.

You’re the judge. Try to solve your current challenges with an AI software and contemplate its benefits. In the end, AI or not AI, that is the question, “whether ’tis nobler in the mind to suffer, [] or to take arms against a sea of troubles.” – Shakespeare.


Further Read:

ComputerWeekly (2020): AI Washing.

PWC (2020): How to be an AI leader, not an AI washout

The State of AI 2019: Divergence, Chapter 7. MMC.

The Wallstreet Journal (2019): AI Startup Boom Raises Questions of Exaggerated Tech Savvy.

Gartner Newsroom (2017): Gartner Says AI Technologies Will Be in Almost Every New Software Product by 2020