AI news is like bungee jumping.
Someone once accused me of being an adrenaline junky. I don’t agree, but I have enjoyed a few bungee jumps. It’s wrongly named: bungee jumping is less jumping and more bouncing. First you jump off something high – which I really like. Then you get to the bottom, experience whiplash, and come hurling back up. And then down again. And up again, and down again, and so on.
The AI bungee bounce
Here is one of the best examples of the AI bungee bounce. Companies went from token maxxing to token minimizing. Known, in corporate speak, as outcome maxxing. A case of “strategic re-alignment”, or serious whiplash.
Gartner predicts another bungee bounce. It predicts that 50% of companies that cut staff due to AI will have to rehire those roles by 2027. Thanks to corporate speak, this will have a different explanation and new position titles.
According to Gartner: “AI simply isn’t mature enough to fully replace the expertise, empathy, and judgment that human agents provide.”
We all need a reminder that AI doesn’t think. It is an LLM, a predictive engine. We laugh when AI claims there is a “b” in “Canada”. But that is the result of its nature.
The human condition
Humans understand. Humans think. Humans also microwave metal, walk into walls while texting, and argue with themselves. Can AI deal with the strange and silly? I have no idea.
Every tech support person has stories about strange problems they’ve encountered. Many of these are strange because of the way the customer experiences and describes them. And so these need a strange and unusual response. A very human response.
I had fun finding and reading some of these stories. These stories aren’t about outdated tech. They’re about the human ability to break things in ways engineers never imagined. If you are tired of the endless AI bungee news, you might enjoy two of these stories.
The 500 mile email
(This makes the rounds every few years, so you may know the story.)
Trey Harris shared this experience from his time running the MIT campus email system. The chairman of the statistics department complained that they couldn’t send email further than 500 miles.
You can imagine Harris’s disbelief. Email doesn’t work that way. It’s not a carrier pigeon. But the stats department had, of course, collected statistics to prove it. (There are some users you shouldn’t argue with.)
After much testing and investigation, Harris found the problem. When a consultant patched the server, a certain timeout was set to zero. The result was that the time to connect to a
remote host on a nearby network was governed by the speed of light distance to the destination. A zero timeout would abort a connect call in slightly over three milliseconds. Or just over 500 miles.
For the full explanation, you can read the full story.
The car that hated vanilla ice cream
A customer wrote to Pontiac to complain about the behaviour of his car. Every night he went out to buy ice cream for his family for dessert. If he bought any flavour except vanilla, the car started without problems. But if he bought vanilla ice cream, the car wouldn’t start.
Pontiac was skeptical, but sent an engineer out to check. He tested this for three nights, and it was true. The engineer made more trips to the ice cream shop and recorded data about each trip.
At last the engineer discovered that it took less time to buy vanilla, because of the layout of the store. This led to the answer. The extra time to buy other flavours allowed the engine to cool down enough to start. When the customer bought vanilla ice cream, the engine was still too hot for the vapor lock to dissipate. This shut down the engine.
I wonder if an AI support agent would have solved these problems. Perhaps it would confidently explain that cars are allergic to vanilla, and you should only buy chocolate ice cream.
I’d love to hear your thoughts.