AI firms believe that no matter how much ultra-sensitive data enterprises share with them, everything would be so much better if they simply shared more. The idea of “too much data” is as anathema to most AI vendors as profitability is to OpenAI. 

OpenAI itself is pushing one of the more audacious plans to collect virtually all of an enterprise’s data. (I can’t envision how that move might end poorly.)

The recurring theme in most of these pitches is that these are smart systems that can figure out in a split second what would take human employees three months. The problem is that interactions with most of these systems today prove that they are anything but smart. 

This is not solely about data/analysis reliability, although that is an issue. These models are often trained on data that is outdated or of low reliability, — and let’s not get started on the never-ending hallucination issues. Then there are problems with these systems accurately understanding the intent behind queries as well as improperly analyzing data. And the models have limited abilities to detect when a user is using the wrong prompts for the information they truly need.

Nope, this is something else. Vendors like to pitch these smart capabilities as being akin to having the most brilliant administrative assistant on the planet. But when people actually interact with these tools in real-world situations, they’re not seeing the “smart.”

It’s not just because of the complexity of Al technology (though it’s almost unfathomably complex). Even the simplest tools come up short.

Let’s look at some consumer devices to make the point. How about Amazon’s Ring video doorbell system? The company touts its Smart Video Search and claims that “using Ring IQ, your eligible cameras can identify people, vehicles, and packages and will only alert you when those things are spotted.”

Really? My settings allow me to tell it to only alert me when a human is detected. So, why am I awakened at least twice a week at roughly 2 AM when the same spider chooses to walk across the doorbell camera? Or when it feels the need to alert me when it starts to rain hard? Last year, I swear that it would routinely alert when — gasp — the sun set. (Ring eventually fixed that one.) 

But my favorite examples come from my iPhone. Better than a human administrative assistant? Envision this: You’re aggressively prepping for a 2 p.m.  in-person meeting with your largest client. Your assistant and three other people are going with you, so you all get into the car and drive off.

When you’re about 10 minutes away from the destination, your assistant interrupts a conversation and says, “Sorry to disturb you, boss. But I need to remind you that you have a 2 p.m. meeting with the client.” You’d look at the assistant as though they were crazy. “You know that I know that. We’re all in the car together literally driving to the meeting.”

And yet, I get that routinely from my phone. Data points: I have the appointment in my phone’s calendar, complete with the precise address. The phone knows my exact location and sees that I am heading to the meeting. I am literally using Apple Maps to give me directions so it knows I am heading to the location of the meeting. 

It has access to plenty of information to know that a reminder — which splashes up on the screen, blocking my ability to see the live directions — is unnecessary.

Consider another scenario. It’s election night. Your assistant barges into a meeting and announces: “Sorry to interrupt, boss, but you wanted to know about that election. Smith won.” You thank your assistant and continue with the meeting.

Two minutes later, he comes in again, interrupts your discussion and tells you that Smith won. “You already told me that two minutes ago.” Yes, the assistant replies, “but that was my telling you that AP called it for Smith. I am now telling you that Reuters called it for Smith.” Is the outcome any different? No. And for the next 40 minutes, he pops in every two minutes telling you Smith won, according to a lengthy list of various news media.

No assistant would do that. Yet my iPhone does. 

Then there’s my Apple Watch. I have it programmed to display the time and my upcoming appointments. And yet, throughout the day when I look at it for a quick update on my next meeting, it has decided to display almost anything else. Sometimes it’s a weather report, other times something entirely random. Heaven forbid I am playing music or watching a video. It then moves to remote control settings.

All I want it to show me is the time and my calendar until I tell it to show something else. My low-tech Timex watch can handle that.

Until these companies start leveraging all of the data they already have — and start analyzing it intelligently — they should rethink asking for enterprise data crown jewels. If the small things aren’t working right, you can bet the larger, more complex systems won’t work right either.