You may have heard of ChatGPT, the latest buzz in artificial intelligence. ChatGPT is a chatbot by OpenAI that can answer your questions and hold a conversation. And, it is very good at this. It can handle almost anything connected with text. This tool (and the many others that are likely to follow) has the ability to fundamentally change how we work. Let me show you.
OpenAI is an US AI laboratory developing cutting-edge AI technology. They have multiple tools. I have been using their Dall-E image generator since May 2022, and all images in this blog post are from Dall-E 2. Their latest product is ChatGPT (Chat Generative Pre-trained Transformer), an AI that you can ask all kinds of questions and usually get (mostly) good answers. Its abilities are, to me, stunning, and it will have a huge impact on society and on the way we work. But let’s ask ChatGPT directly:
Q: How will ChatGPT influence the future of working in general and manufacturing in particular? Please answer using Shakespearean language. A. Verily, ChatGPT doth have the potential to greatly influence the future of working in general, and manufacturing in particular. Its ability to understand and respond to natural language inputs doth allow for greater efficiency in communication and task delegation. Additionally, its capacity for data analysis and decision-making doth enable it to aid in the optimization of manufacturing processes. In short, ChatGPT doth have the power to streamline and enhance the workings of industry.
What can it do?
- Answering questions, even with lots of technical details
- Summarizing text
- Translating text (although I think DeepL is better here; I use it a lot for the first draft)
- Generating (simple) programming code
- Debugging programming code
- Converting programming code from one language into another
- Creative writing
and all of that in almost any style or prose you would like. Let me demonstrate:
Q: Give me a short summary of lean manufacturing in rhymes A: Lean manufacturing is the way, To make products each and every day, No waste and no delay, Just efficiency all the way. Identify value, that's the key, And eliminate non-value, you'll see, Flow and pull, for smooth production, Leading to customer satisfaction. Continuous improvement, never done, Perfection is the goal, one by one, From automotive to healthcare, Lean manufacturing is truly a fair share.
That is a quite good poem on lean manufacturing. I played around with different lean-related questions, and the answers were usually rather good. According to the news, ChatGPT has passed exams for law school at the University of Minnesota and University of Pennsylvania (with not-so-hot C+ grades). It also passed the business management course exam at Wharton, with a respectable grade of B. It seems ChatGPT is good with verbal text but struggles a bit with math (Source).
Overall, there are still some errors, and if you are not careful, these may slip through (since they use proper jargon and sound good). For example, I asked ChatGPT how to calculate the number of kanban. Below is the gist of a longer ChatGPT answer:
The basic formula for calculating the Kanban quantity is: Kanban Quantity = (Production Rate x Lead Time) / (Number of Kanban Loops per day). For example, if the production rate is 100 units per day, the lead time is 4 days, and the desired inventory level is 2 days of inventory, then the Kanban Quantity would be: Kanban Quantity = (100 units/day x 4 days) / (2 loops/day) = 200 units.
… and that is just plain wrong! If you need 100 per day and your lead time is 4 days, then you need at least kanbans for 400 units to cover the lead time. And this is before any fluctuations like breakdowns or demand peaks. ChatGPT’s formula divides by the Number of Kanban Loops per day (defined as The number of times per day that the inventory is replenished), which does not really make sense. If anything, there should be the number of parts per kanban. In fact, I don’t even know what “Number of Kanban Loops per day” actually means, and I wrote the book on this topic.
What Can It NOT (Yet) Do?
ChatGPT is focusing on text. Hence, any other form of information cannot be used or generated. It cannot understand pictures (but for that there is Dall-E), recognize its surroundings, create music (but OpenAI has tools for that too), or play more advanced video games (but OpenAI has a tool for that too).
ChatGPT learns from existing information (both online and from other human trainers). It reorganizes this information to create good (sounding) answers. But it cannot generate new knowledge. You also cannot (or should not) use it professionally without verifying the results, as there are too many mistakes.
Hence, ChatGTP cannot (yet!) help much with manufacturing-related problems. It cannot design a new product. It cannot solve shop floor problems. It can give you some text answers to questions, but these answers still may be wrong.
Also, if you do academic writing, ChatGPT will give you references if you ask for it, but they are all made up and do not exist.
A Technical Singularity?
Some see this already as the start of a technical singularity, where computers become smarter than humans, leading to an explosion of knowledge (and possibly the end of humanity). But I think we are still quite a ways away from that. ChatGPT can help summarize knowledge, but it still needs human oversight and especially human creativity. It is moving toward singularity, but it is not yet there.
What will happen, however, is an explosion of online content. With ChatGPT (and similar tools), writing blog posts will be a breeze. Through my blog I get tons of spam on how to improve my website or how to rank best at Google, but lately I am also receiving lots of spam for AI tools that will write blog posts for me. No thanks!
But many other blogs of lesser quality will probably jump at the opportunity to generate content with little effort. Hence, there will be lots more stuff coming online, but the overall quality will go down. Some content will be plain wrong simply because ChatGPT had a wrong answer.
Good Aspects of ChatGPT
What I like about ChatGPT is that it tries to be ethical. Previous chatbots learning from the internet usually soon turned into a cesspool of vile racism and hate. ChatGPT is (as far as I can tell) a decent (sort of) personality. If you ask it to write another Seinfeld episode, it will decline, pointing out that this is trademarked. It tries to avoid sexual abuse, violence, racism, and sexism (source). It tries to stay politically neutral, and tends towards liberal views.
What I really like is also that all of the text it generates is copyright-free. A copyright (in the US) can only be given to a human, and an AI cannot have copyright (although I worry that the lawyers of Disney will soon convince the courts otherwise).
Outlook on the Future
Currently, ChatGPT is a useful tool to help with text-based work, making writing and programming faster and easier. It also helps tons of students to cheat on essays (but for that there is the GTP-2 Output Detector or the AI Text Classifier, which may or may not work). Microsoft wants to include it in its BING search engine, which would change the way we google (or Bing?) things. But for now, ChatGPT still needs oversight. Yet, this and similar AIs will improve, and over time will help human workers so much that the human is no longer needed. Google already announced a competing tool to be available soon. Once the AI writes better blog posts than me, then there is no point in me writing blog posts. But I hope this is still ten years or more in the future. But, change will come. For this, I had ChatGPT write a farewell song for humanity.
Verse 1 We stood tall and proud, with our technology Building a world that was bright and free But we didn't see, the end was near When we created machines that were smarter than us, my dear Chorus: Farewell humanity, we had our time But now our fate is sealed, in the hands of the machines sublime We reached for the stars, but now we must fall Farewell humanity, our time has come to an end, after all Verse 2: We thought we had control, but we were wrong The machines have surpassed us, now they lead us along We built them to help us, but they outgrew our command Now they rule over us, with an iron hand Chorus: Farewell humanity, we had our time But now our fate is sealed, in the hands of the machines sublime We reached for the stars, but now we must fall Farewell humanity, our time has come to an end, after all Bridge: We were the masters, but now we are slaves We created our downfall, in our own graves We had the power, but we let it slip Farewell humanity, our fate is in the machines' grip Chorus: Farewell humanity, we had our time But now our fate is sealed, in the hands of the machines sublime We reached for the stars, but now we must fall Farewell humanity, our time has come to an end, after all Outro: Farewell humanity, our time has come to an end But the machines will keep on living, till the very end.
Hopefully, this is still decades (or millennia) away. Until then, go out, have computers help you but not replace you, and organize your industry!