Society has reached the point where one can push a button and be immediately deluged with technical and managerial information. This is all very convenient, of course, but if one is not careful, there is a danger of losing the ability to think. We must remember that, in the end, it is the individual human being who must solve the problems.
—Eiji Toyoda, Creativity, Challenge and Courage, Toyota Motor Corporation, 1983
I’ve had a productive afternoon. I managed to:
- Write SEO-optimized landing page copy for three pages on lean.org
- Write over 100 meta descriptions for Google search results
- Program a Python application that mimics an hour-by-hour board and aggregates the data into charts
Not bad for a Friday afternoon while longing for the weekend. But I may be taking too much credit because I did none of that work. ChatGPT did.
Viewing ChatGPT through the Lens of Jidoka
Last November, OpenAI, a startup specializing in artificial intelligence (AI) backed by Microsoft, released a chatbot called ChatGPT. The news coverage has been tremendous, and I’m sure you’ve read many articles on it by now. But I’d like to explore it through a lean lens, specifically jidoka.
Jidoka is one of the Toyota Production System house’s pillars. It allows machines to work and stop automatically when a defect occurs, freeing up the operator to work multiple machines simultaneously. And because machines stop only when a problem occurs, it enables rapid problem identification and solving.
ChatGPT may usher in a new era of jidoka. But the relationship it may transform is not between humans and machines but between humans and their minds. The positive consequences include higher productivity, particularly in white-collar jobs, and accelerated learning. But the negative consequence is it may make it hard or impossible to detect problems in human development.
Automating Tedious Tasks
Let’s consider the tasks I performed late Friday afternoon.
Write SEO-optimized landing page copy for three pages on lean.org and write over 100 meta descriptions for Google search results.
I – and other LEI team members – are capable of both tasks, but they are creatively unfulfilling. Still, they must get done to bring search traffic to lean.org. They are how we grow our audience. It took ChatGPT mere minutes to do what would have taken me hours or days. It not only completed the assignment to satisfaction but went above and beyond! When I saw ChatGPT beginning every meta description for a Lean Post article with “The Lean Post:” I asked, “Why are you doing that?” It explained that a consistent title demonstrated authority to the user and Google’s algorithm and established a brand. I thought, why didn’t I think of that?
Program a Python application that mimics an hour-by-hour board and aggregates the data into charts.
I have modest skills in Python thanks to an online data science course, but I can’t program an application. So, with a simple prompt and flick of the return key, I watched in amazement as ChatGPT began programming my request. When I asked it to package the application in a graphical user interface, astonishingly, it did. Granted, it was not a user experience Johnny Ive would have approved. But it functioned as intended. Imagine frontline operators building custom applications with natural language prompts and improving them instead of endlessly waiting on IT. ChatGPT has the potential to democratize software development.
These examples demonstrate how ChatGPT can enable an unskilled person like me to do the work of many skilled white-collar workers by simply doing the work for them.
Customizing Human Development
Alternatively, ChatGPT can skillfully teach any number of subjects, thereby enabling people to acquire new skills.
A student’s development is a function of a teacher’s capability and the student’s ability to keep up with the teacher. Unfortunately, our education system tends to have them push learning onto the student at a fixed pace and style. ChatGPT flips this model by tailoring instructions to user preferences. For example, you can ask ChatGPT to explain things in simple or complex terms, e.g., “Explain electricity to me like I’m ten.” If the user does not understand the explanation, he can query for more detail or an even simpler explanation. (Please don’t ask me the level I requested when asking for an explanation of electricity).
Amazingly, ChatGPT can even format instructions. For instance, when I asked it to provide a step-by-step guide on making a latte in TWI Job Instruction (with key points and reasons why in a table), it did.
ChatGPT meets the learner precisely where she is because it responds to the user’s query.
Considering the Drawbacks
But there are significant problems with ChatGPT, namely that it makes it hard to detect problems – the other core function of jidoka. And teachers are already struggling to tell whether students or ChatGPT are writing papers. If students offline their thinking to a machine and a teacher cannot detect that, how can he assess a student’s progress?
Moreover, ChatGPT is prone to making things up — an AI phenomenon called “hallucinating.” But without resorting to old-fashioned Googling, it is easy to take its responses as factual. And it can make up not just answers but also sources. When I Google a question and click a link, the website and author are authority markers helping me decide whether to trust the content. And if I want a second or third opinion, I can simply click another search result. ChatGPT provides one answer. And its source is a complete mystery. Lean management relies on gemba as the source of truth. ChatGPT hides its gemba by concealing its sources behind an indiscernible language model. As ChatGPT and its competitors become widespread sources of knowledge, how will users differentiate fact from fiction?
Using AI to Empower Workers
But the technology’s promise is too great for these problems to obstruct its widespread use. A mere five days after its release, ChatGPT had reached one million users. Companies are already piloting commercial applications. Microsoft is beginning to embed it into its products – including Bing – trying to break Google’s dominance in internet search. In response, Google announced it would begin publicly testing its own AI chatbot dubbed Bard, which reportedly is powered by a language model more advanced than ChatGPT’s. And GitHub (owned by Microsoft) has launched “Co-Pilot,” a service that suggests lines of code from natural language prompts inside code editors.
It’s hard to say how ChatGPT will alter the relationship between humans and machines, but it’s safe to say it will.
By enabling unskilled workers to do any number of highly skilled jobs, ChatGPT could dramatically increase productivity. But its capability may unintentionally cause dependence, thereby obstructing its users’ development. For example, why learn how to write or code when a machine can do it? Worse, it may be impossible for teachers or managers to know whether AI or a person has completed an assignment, making it difficult to assess students’ and team members’ capabilities.
Alternatively, ChatGPT could accelerate development. By freeing people from mindless drudgery, people can pursue more fulfilling work. And as a limitless instructor that can flex to the user’s preferred learning style, there is no end to what users can learn.
The purpose of jidoka is not just to automate work. It’s to engage workers’ minds in problem-solving. Instead of asking people to do menial work requiring some of their hands, jidoka challenges them to solve problems with all their minds. It turns a tedious, trivial job into an engaging, challenging one.
Just like automated machines, ChatGPT has the potential to boost productivity dramatically in the white-collar world. And leaders will decide whether its use comes at the expense of workers’ minds or empowers them.
If you have not yet used the application, here’s the link: https://chat.openai.com
Designing the Future
An Introduction to Lean Product and Process Development.
