In the Early Days of AI

I am not a luddite. In fact, I am building AI tools to manage digital advertising. But I just have to say, we need to be a little less trusting when it comes to these early days in this technology.

Stay with me here because what seems to be trivial illustrates the potential for catastrophic results. Here’s a case in point.

I’m doing some research on my next car. I asked Google’ Gemini the following question:

—Doug: “What is the optimal lease deal on a Hyundai Ioniq 5 XRT?”

I received a pretty comprehensive response, which included the following statement

—Gemini: “Ensure the $7,500 EV lease bonus (Pass-through / HMF incentive) is fully applied to capitalize cost reduction.”

There is a big problem with that conclusion – the lease bonus does not exist. So, my response was:

—Doug: “You are out of date on your information. There is no federal $7,500 EV tax credit”

To which Gemini replied:

—Gemini “You are entirely correct, and I apologize for that oversight. The federal $7,500 EV tax credit was repealed under the One Big Beautiful Bill Act (OBBBA) and officially ended for new purchases after September 30, 2025”

I was a little incredulous at this oversight, so I asked Gemini

—Doug: “Why didn’t you know this?”

Google’s rationale was:

—Gemini “I slipped back into an outdated baseline because my training history is heavily saturated with years of data detailing the original 2032 expiration date. When analyzing lease dynamics, my system initially defaulted to that historical “leasing loophole” framework instead of strictly locking into the current September 2026 reality.”

Here’s my issue with all this. This is a pretty poor level of analytics. If I applied this current level of technology to create the optimal media plan based on historical results, I don’t think my advertising client would be happy with the results.

Here’s my bigger issue with all this. If a government agency relied on this level of analytics, would it be any surprise if they used the resulting data to arrive at faulty policies, or misguided objectives?

Just saying. It’s early days here. Be vigilant in accepting conclusions at face value.