Wrong answer
I would still ask AI about a trouble code because it can explain what it means and help me understand the possible causes.
Text size
I would still ask AI about a trouble code because it can explain what it means and help me understand the possible causes.
A car dealer recently had to honor a trade-in offer made by its chatbot.
Zack Giacomelli contacted BMW Toronto to sell his 2021 BMW and spoke online with Quinn, which he believed was a person. Quinn offered him C$27,162.79 for the car.
The dealer later told him that Quinn was an AI chatbot and lowered the amount to C$20,000.
BMW Toronto blamed the error on a mix-up involving the balance Giacomelli still owed on the car, then honored the original offer after a news outlet asked about the case.
The dealer eventually said human employees would make future offers.
The chatbot made a mistake of about C$7,000. It was fortunate for the owner that the error went in his favor.
A wrong answer about a vehicle repair may end differently.
If the same thing happened here in the Philippines, the dispute would probably move to Facebook and most likely go viral.
I could imagine a car owner posting screenshots of the offer, while the buy-and-sell outfit insists that its chatbot made a mistake.
Then the comments would soon be filled with accusations, jokes, memes and people who were not part of the transaction but had already decided who should pay.
In AI, there is such a thing as AI-as-a-judge, where one AI model evaluates the answer produced by another.
On Facebook, especially in the Philippines, anyone can become the judge, even without reading the full exchange.
That is the risk when a business allows AI to speak on its behalf. The chatbot may generate the answer, but the customer will still hold the seller responsible for it.
I worked in semiconductors for many years and have worked with AI since 2022. I now use AI as a coding assistant and build deterministic systems that coordinate AI agents without relying on unverified LLM output, so I understand why their answers can sound convincing.
A scanner or an On-Board Diagnostics (OBD) may retrieve the code P0301, which means the vehicle detected a misfire in cylinder one. AI can translate that code and suggest common causes.
A driver may type, “My car shakes and shows P0301 on the scanner,” on his AI chatbot without mentioning that the engine was washed the previous day, or he may forget that a replacement coil was installed last month.
Arguably, the chatbot can also ask follow-up questions, but it still depends on what the owner notices and reports. The mechanic can inspect the vehicle and check the work already done.
A technician can see the spliced wire and the connector held together by tape, while AI will not know they exist unless the owner finds them first and sends a clear photo.
I would still ask AI about a trouble code because it can explain what it means and help me understand the possible causes.
That gives me enough information to ask the mechanic why a test or replacement is needed.
The cost of a wrong oxygen sensor or ignition coil may be manageable.
However, advice involving the brakes or steering deserves no such experiment because a chatbot cannot test the vehicle after the repair.
AI can tell me what P0301 means.
I still want the mechanic to show me what caused it.
