58) AI Washing

Behind-the-scenes building Vambrace AI, a company on a mission to figure out its mission. Please pardon the stream-of-consciousness style. Subscribe to follow along or visit the site here:

(typos are to make sure you’re paying attention)

Introductory Remarks

Dear Vambracers —

In last week’s post, Adoption, we examined AI’s value chain, from the earth to actual end-use, and I claimed that the adoption piece of the value chain was the most important and is currently one of the most difficult parts of the equation to solve. I am of course very openly biased since I work on the adoption component of the AI supply chain, but that’s neither here nor there. Continuing on!

AI Washing

In today’s post, I wanted to discuss the concept of “AI washing” which is pretty much just claiming the use and/or presence of AI for some business activity in an overstated manner. It’s effectively the exact same as “greenwashing” just applied to the use of artificial intelligence within business. As AI has risen to the fore of the broader business discourse, there are more instances of AI washing than ever before.

Why does it happen?

To deconstruct the motivation to AI wash, we have to understand incentives. In general, I’d contend that the public markets dictate the language that companies use to market their business decisions. Specifically, there’s a premium associated with companies perceived by the public markets as being “ai-forward”—and that premium extends to the private markets as well.

A good historical and well-known example is what happened with WeWork, where Adam Neumann worked really hard to convince people that WeWork—which was fundamentally a real estate company—was actually a technology company. It’s much easier to tell a growth story for a technology company than for a real estate company, and a “growth story” drives a richer premium.

In a way, the public markets sort of help us understand, at any given point in time, what people think about the future of some business organism, on some longer-term horizon—maybe 3-5 years. And right now there’s a huge advantage to being perceived as AI-forward in terms of competitive position, which will put companies in a better position to capture value and dominate in an AI-first future. Honestly it’s all pretty straightforward.

Given these incentives, companies have started to really lean in on AI, due in not-insignificant part to be perceived as being AI-forward and reap the valuation benefits associated with being AI-forward. I do think that this perception battle drives a fair bit of the interest in AI, in addition to the actual business value that can be possible from AI tools and automations more broadly.

Behavior that AI washing promotes

The potential downside, however, is that people start to overly AI-ify things even when it might not actually make any sense. I’ve put a silly example below of this dangerous behavior.

In effect, AI is very good at predicting the next word. We know that to be true. But, in many cases within business and computing, a deterministic approach is actually sufficient to transform data within some business function. And what I think people are starting to feel inclined to do is the above, where you just put a deterministically-suitable question or task into AI and then claim the use of AI. This isn’t computationally efficient and it also introduces randomness, which could lead to poor outcomes—because AI could at some point say that 2+2=5.

What’s next?

There are, then, legitimate questions about how and where to use AI vs deterministic approaches within some business automation. And I think we’re all kind of figuring that out. I definitely think that relying on deterministic regular expression capture is a pretty crude tool for some categorization or term-identification stuff. But I also think that AI can be grandiloquent and verbose for some of the same work, so there’s a balance there. And then how do you implement deterministic filtering as an anti-hallucination measure or something?

I just think there are a lot of questions that are still being asked about the value of Ai within some business automation, and I still think we are pretty far from an AI-only approach to some business automation. Because eventually, assuming continued model improvements, we theoretically will be able to give an entire set of production data to some model and ask it anything about that data, and it will give us accurate, context-rich results. But that’s really challenging right now without robust ontological support and dataset-specific context.

I do think real organizational gains are being made in terms of the personal use of AI for higher output, from an IC perspective. But organizationally, there is a lot of difficult work ahead to figure out how to string together these deterministic and AI-enabled systems that effectively complete some critical data transformation for a business. Because ultimately really all a business is is one super complex system of data transformations. And right now the data transformation agents are largely human, but that doesn’t have to be the case. And I think with a lot of hard work it won’t be the case.

Looking Forward

A little bit meandering today, and I feel like I have more thoughts too, but the only point I really want to make is that the line is blurring between what is and isn’t “AI” within some business automation. And it’s important to objectively and critically assess the individual (or organization) making any AI-use claim, and to think about the motivational dynamics that might underpin their claim.

None of this is to say there aren’t legitimate AI use cases that exist, or that there won’t be an abundance of legitimate AI use cases in the future. But I just think we’re in a weird gray area that permits many actors to over-state the role of AI. And that’s okay! But it’s our duty to be aware of it.

Thanks and have a great week!

Sincerely,

Luke