As a retired software architect and director of software engineering, I worked with AI on and off for decades. It wasn’t until the last 5 years or so that systems have gotten fast enough to really make an impact with AI systems. But an old saying still applies - Garbage in = Garbage out. AI systems rely on knowledge from experts in the field. You don’t go survey 1,000 random people for input on how to build nuclear reactor. You get input from educated people in the field.
One AI system I had a lot of input in was when I was working as a software consultant almost 30 years ago. A team of 5 software engineers designed and built an automated warehouse system - nowhere near the sophisticated systems used today by companies like Walmart or Amazon. We gathered knowledge on how the warehouse worked from the people that worked there. It became clear early on that there were 2-3 people who had extensive knowledge on how everything worked. We used their knowledge to build the AI based inventory control system for the warehouse. The system was actually built in 5 months…and then another 5 months to workout all the kinks (flaws). As of 2 years ago the system was still in place (that’s an awful long time for a piece of software to run).
Back to cars. Ford recently laid off many of their engineers to be replaced with AI. The problem is Ford laid off too many of the WRONG engineers. Many of the engineers laid off were top engineers with decades of experience. The AI system Ford built was based on faulty knowledge from the wrong people (Garbage in = Garbage out). Ford is now hiring those engineers back to train junior engineers and for better input into the AI system.
The auto industry has made similar mistakes before that they did not learn from.
During a downturn they would offer early retirement packages to the senior engineers only to find the loss in very specific knowledge bit them in the rear after they hired new grads at a fraction of the pay. Same mistake only with AI.
I just shake my head and chuckle when I read about such things.
Strange the folks in charge never realized this. They talk about reputational issues like it’s just another line item to manage, but for some of us who are not up on all the product detail, we are simply afraid to buy a ford product. Add GM to that list now and Chrysler (a rose by any other name) has been left in the dust for years.
While I blame EPA for forcing conversion to turbo 4 cyl and unreliable transmissions, there is no reason they couldn’t beef up bearings and other engines to handle the job. So now reputations destroyed and millions of engines on recall. Duh.
Absolutely right Mike. If you work in this industry, you don’t need hindsight to see the failings happening in real time. It is obvious to the people in the trenches for example.
One issue I have with this kind of reporting and it’s a nitpick, is everything is lumped into this category of AI now when often what’s being described is a small subset of what encompasses AI. These are often describing expert systems with inference engines and not really what a modern AI system can/should be able to do. The main difference is that a true AI system can adapt its programming to new data. What they likely have is a rules based logic system that can identify a possible solution based on the outcome of the rules logic that has been programmed into it. A real AI system would eventually figure out it is wrong and develop new algorithmic logic and subsequent solutions based on those incorrect outcomes. It learns. The human experts are no longer required then.