Why people say AI does not work — what ails visual inspection, and the need to explain it honestly

What this article covers

AI visual inspection drew a great deal of attention over a few short years — and also no shortage of people saying it did not work as expected, that AI was no use. It is true that the technology was still developing then, and that putting it to work on a factory floor was difficult. But more than that, exaggerated claims, expectations pitched too high, and an accumulation of small misunderstandings shook people’s trust in AI itself.

This article looks at where that mistrust came from, in the structure of the industry, and at why the industry needs evangelists: people who explain the technology correctly. It then asks what AI can contribute to the problems Japanese manufacturing faces — too few people, high-mix low-volume production, sharpening competition as production goes global — and how OUEN approaches that.

It is a slightly sentimental piece. It is written honestly, out of a conviction we hold: that AI can make Japanese manufacturing strong again, and that doing so takes the ability to explain it truthfully.

A spreading mistrust of AI

After the AI boom around 2018, solutions using AI spread quickly through visual inspection too. Startups appeared one after another, trade fairs filled with AI visual inspection, and case studies were presented everywhere. Behind the enthusiasm, though, plenty of people were saying they had bought it and never got it working, that in the end they were inspecting by eye again. Talking to customers on site, we meet more and more people who have concluded that AI does not work, or that it works but only within narrow limits.

We know better. AI is not magic, but used properly it can certainly make Japanese manufacturing stronger. And behind the misunderstanding that AI does not work — beyond whatever the technology of the day could not yet do — we think the real problem was how it was described.

From around 2018, when AI drew everyone’s attention, many companies went for fast growth on outside money. Racing towards a listing or towards profitability, they hired heavily to support the scale, expanded sales teams and pushed hard on volume. To let anyone sell it, conversations and materials were turned into templates, and the only conclusion those templates could reach was that any inspection can be done with AI.

So projects that would have been better served by rule-based image processing, and projects whose conditions made AI a poor fit, got bought anyway, the novelty of AI carrying the investment decision. Then operation began, the results on the floor were not what had been imagined, and the verdict spread: AI does not work.

Meanwhile the market filled with exaggerated success stories, and the difficulties of actually running the thing became hard to talk about. Disappointment accumulated — bought it, does not do what we hoped — and trust in AI itself gave way. That was not a limit of the technology. It was mistrust produced by a lack of honesty.

A new technology spreads only on a correct understanding

What the industry needs now is people who explain it correctly: evangelists. An evangelist is not someone who sells their company’s product, but someone who explains so that the value of the technology is understood.

In the diffusion of innovations, the theory put forward by the sociologist Everett Rogers, a new technology reaches society through innovators, then early adopters, the early majority, the late majority and laggards. Early adopters matter particularly: they not only take the technology up early but share the experience around them, carrying it to the majority. For a technology to reach society at large, that sideways spread by early adopters is essential.

But a correct understanding does not spread merely because people who have used something pass it on. Mixed with exaggerated expectation or wrong information, it can damage trust instead. What is needed is people with sound knowledge who are clear about what it can and cannot do, and who say so honestly.

That is the role we have in mind for an evangelist. Not selling the technology but growing an understanding of it. Reducing the misunderstandings and spreading a fair assessment. That patient accumulation is what builds trust in AI visual inspection, and in turn lifts the competitiveness of Japanese manufacturing as a whole.

AI is one of the cards manufacturing has to play

Of course, belonging to a company makes complete neutrality hard. That is exactly why it matters not to exaggerate, and to say the difficulties out loud. Our work is to convey a correct expectation rather than a convenient dream. In a field as unfinished as AI visual inspection, explaining honestly how far it goes instead of hiding where it stops — that is the first step towards AI earning trust again.

AI can make Japanese manufacturing stronger. Not only by raising accuracy and speed, but by carrying the experience of skilled people into something digital, so that stable quality and efficiency can be had together. Japanese manufacturing today has a stack of problems to get past: a shrinking working population, high-mix low-volume production as customer demands diversify, and price competition as production goes global. AI can help with those. Supplementing the experience and judgement people have carried, so that a limited workforce can still hold quality steady — that practical approach is, we think, where the value of AI actually lies.

At the same time, AI is still a developing technology. There are things it cannot do today. We do not take that gloomily: what matters is turning those limits into the seed of the next piece of development. Because we develop our AI ourselves, we can take a reason something could not be done on a factory floor and convert it directly into a research topic, then give it back as technology. That is one of our greatest strengths.

For the potential to be realised, though, trust has to develop as much as the technology does. For people to believe in AI again, we have to keep explaining honestly ourselves. That is bound up with where the name OUEN came from.

We try not to exaggerate what AI can do, not to be afraid of what it cannot, and to walk alongside the floor. A new technology spreads only by being tried and being understood. So: explain more honestly than anyone, describe it more carefully than anyone, and talk with more factories than anyone. That activity is what evangelism means to us.

Spreading a technology is not simply selling a product. It is accumulating trust. For AI to make Japanese manufacturing strong again, we mean to hold on to the ability to explain it truthfully, and to win back the industry’s trust and its expectations.