It seems like almost everything is claiming artificial intelligence capabilities now. We are entering the “Wild, Wild West” phase, where everything is AI and the definition of what AI is, and isn’t, is really pliable depending on what it is you are selling. There’s a joke going around where the interviewer makes the corny request, “Sell me this pen.” The interviewee does not lose a beat: “It has AI.”
I really could not believe my eyes when I was in the local home center and I saw a washer/dryer combo from Samsung that boasted of its AI-driven capabilities. The person at the store didn’t know much about it, so I went to the Samsung website to find out more. Here is what I could find:
“AI OptiWash & AI Optimal Dry: Detects soil level and fabrics, and adjusts settings as needed during the cycle.”
Another appliance (a fridge) is shown in another section of the site, opening its door for a man who has a bagful of groceries. The fridge doesn’t need any more intelligence than a sensor that detects something in front of its door, and if it’s there for more than three seconds, the door opens. The man in the video might need more intelligence for wanting to put an entire, intact bag of groceries into the fridge.
What does the washer need? A sensor that measures the color of the water, or the opacity of the water, its specific gravity, or light absorption of the water, or maybe a combination of a few things. As for the fabric, you might weigh the clothing while dry and while wet…you get the idea. I’m skeptical that you need to deploy AI at all in these situations. If it is being used, it is the weakest of so-called “Weak AI,” which is deployed to assist in narrow, specific tasks. Strong AI is deployed in systems with general cognitive abilities (things like ChatGPT, weather forecasting, customer action/prediction, pattern recognition, etc.).
What this has to do with manufacturing is that we all need to be a little careful when considering the topic. If someone is bold enough to put those two letters next to each other, AI, then who is to say it does or does not fit that label? One thing I believe, and that is the old, established players in our market would not risk their goodwill of 50 or 100 years on trying to gin up some marketing effect.
How to avoid being snookered:
- Ask about the way that the AI system was created. Did it learn? Does it learn? If the answer is no, you know.
- Seemingly endless IF, THEN, ELSE loops easily masquerade as AI. If there is some simple test (the color or opacity of water, for example), large stacks of these programming staples can look like AI. Sample: If water opacity = 3, then GOTO 3ROUTINE, ELSE NEXT….and so forth with 3.01, 3.02…Let’s say it get to a reading of 4.33 and it is a true statement. It tells the washer to use the settings that were agreed upon for a load of 4.33 laundry. This is not AI, but it looks just like it from the outside.
- An Ai system needs so much data in its early existence that companies are offering large-scale data sets for sale. They call it training data or something close to it. If your system is meant to be an AI assistant at a high-end hotel, it will need to consider its restaurant. One of its tasks is to learn about silverware. Fork, spoon, knife. Lots of different iterations of each one. Did it learn about each category, enough to make an educated guess on the application of a fork it hadn’t seen before? If so, that’s a good indicator of real AI.
In a few weeks I’ll be giving a speech on this topic to wire manufacturers in Ohio. I cannot wait to hear their input and their experience and frankly, their concerns about AI. I will report back to you with more on this very important and timely topic.