• AI-washing is the modern-day business slur.
• The technology of AI opaque enough to hide many inconvenient truths.
• Announcing a non-AI-driven service may be a future differentiator.

Given the huge amounts of hype around AI, it’s little wonder that many companies have been accused of ‘AI-washing’. It’s a process akin to green-washing, where organisations make specious statements about their good works with regards to the environment. AI-washing involves companies making claims about their products’ use of AI that are at least economical with the truth, but in many cases are outright lies.

There are significant pressures placed on companies to produce AI-driven products, and together, they add up to good reasons for AI-washing. Apart from the need to join the bandwagon (“everyone’s talking about AI and we need to be a vendor that everyone’s talking about”), there is a consensus that AI makes things better. Without AI, organisations cannot improve their products, especially in the ways competitors seem to be doing.

Luckily, companies can lean into a general ignorance around the subject. As science-fiction writer Arthur C. Clarke once said, “Any significantly advanced technology is indistinguishable from magic.” To most users, the basics of how the internet works are unknowable mysteries – AI is a hundred times more inexplicable, so might as well be termed magic.

By stating their membership of a group of organisations that practise magic, companies elevate their offerings into the same realm as incantations: spells that have an effect, but mere mortals cannot hope to understand the ethereal forces that create it.

In boardrooms, the practical realities of AI’s capabilities are ignored when discussion turns to new product features and markets. The act of ‘getting on the AI train’ makes the train a destination as of itself, not because the train is going anywhere worthwhile, but because people say everyone’s on the train.

In many cases, those tasked with ensuring that the organisation gets on the train will simply take the easiest path: at present that’s often a ‘smart’ chatbot that gives personalised garbage answers to customer or employee queries, rather than traditionally-created garbage answers. But it’s enough to gain a seat on the train, and that’s what matters.

AI can’t iteratively improve

As is often the case with new technology, there is a divide between the promise and the reality of AI. That gives companies the green light to deliver features and upgrades to their products that claim, falsely, to be ‘AI-driven’. When the new versions ‘with AI power’ fail to deliver any real benefits, no one is particularly surprised nor disappointed. Genuine AI technologies produce empirically average results^* which can’t be trusted to be accurate. Why should anything feigning AI be any better?

In the US, the Federal Trade Commission is bringing proceedings against dozens of companies claiming falsely to use AI in their offerings to the detriment of users and investors. Investors spending billions of dollars should perhaps know better than to back companies without going through due diligence processes. The only sympathy owed around investors’ losses is due to the fact that it’s often other people’s money being bet with.

Quantifying users’ losses is more difficult. By handing over a sum of money to a self-proclaimed Wizard Of The AI Arcana, users can and should expect a spell to be cast. But even the most learned Mage (we could call them ‘OpenAI’, for example) can only spell-cast a standard, grey-brown, rather average rabbit from a hat. The charlatan mage also takes users’ money, mutters incantations, and produces a similar rabbit from a similar hat. Who really can tell the difference? (Although closer inspection may reveal that the ‘real’ wizard’s rabbit has three heads and a tail that melts.)

There are historic reasons why getting on the AI train early should prove beneficial, apparently. Companies selling goods or services that didn’t get on the ‘internet train’ early were in many cases the losers to those that did, like the independent bookseller vs. Amazon, for instance. Software development is iterative, and improves over time, so early-adopters are thought to be wise. The same must be true for AI, is it was true for cloud, NFTs, and blockchain. Or were those technologies over-hyped?

In the same way that good magicians never reveal how their tricks are done, AI wizards rely on ignorance of some hard facts based on the irritatingly-immutable laws of physics:

• AIs need bodies of data to learn from, and given that they already have combed the sum total of available digital human knowledge in the form of the internet, AIs have hit a finite limit on what they can learn from. Sam Altman cries when he realises there are no more internets left to conquer.
• The more content produced by AI that’s published on the internet, the worse the answers AI gives to subsequent queries.
• Queries and requests to AI cost more to answer than users pay. Additionally, questions posed to AIs are very often made up of several, cumulative queries, making their operators even more financial loss than, for example, search engine queries.
• If we assume the sum total of human output on the internet forms an ^*average level of accuracy, dependability, and intelligence, undiscerning AIs combing all available data will produce answers of average accuracy, dependability, and intelligence.
• At present, hallucinations (AKA, mistakes) dog a non-trivial percentage of AIs’ responses, giving the game away that they make mistakes. Big AI vendors are working hard at present to remove this telltale sign of fallibility by engineering checks and balances after results to queries have been inferred.

In the future, organisations may wish to differentiate themselves from the average, fallible, and inaccurate results their AI-powered competitors offer customers, employees, and decision-makers. Perhaps there will come a day when companies proudly declare that their software is based on real code, written by real people? In that possible future, companies will be accused of AI-washing when their services are AI-driven under the hood, and more trustworthy businesses don’t bother with the underwhelming technology we once called artificial intelligence.