AI Hallucinations

AI Hallucinations

There are three things to look for when trying to determine if AI has created an image.

Part of my job is to acquire images for magazines, videos and websites. I was recently sent links to artificial-intelligence (AI)-generated content of parts and technicians. I searched for pictures of brake calipers. Here are two examples (below):

It is easy to see that these brake calipers would never work. These are AI “hallucinations” of what it thinks a brake caliper should look like, according to other websites, not how they operate. They create a composite image of how a brake caliper should look for all vehicles. 

One of the worst AI images is titled “Technician Inspecting Wheel Bearing” (at the bottom). How many rotors are on that axle? Why is he resting on a clutch disc with a dual-mass flywheel?

Over the past 20 years, I have acquired the ability to look at an image and identify the car or truck by the layout under the hood or other components. With an AI image, it is almost impossible to tell what they are working on because it is a composite of several photos that AI has scraped from other websites. 

The one good thing about these AI images is that they usually make the techs look good. The majority of techs in these images are buff and have trimmed beards. But, the look on their face is the typical 1,000-yard stare and they’re not wearing safety glasses.

So, how can you tell if an image is AI?

There are three things to look for when trying to determine if AI has created an image:

1. It doesn’t look right. If your gut tells you it is AI, it probably is.

2. Look at the hands. AI has a problem rendering hands and fingers. The fingers might appear to all be the same length, and the thumbs might not be shown at all.

3. Blurry Areas: You may notice that some AI images have a shallow depth of field. This can be seen as blurry areas of the image. Often, in the blurry areas, you will see AI artifacts where it is using a part of an actual image that has been poorly altered. 

While the resulting images can be comical at times, they make the serious point that generative AI is not ready for prime time. Yes, AI can make some jobs more manageable and possibly eliminate some jobs in the future. But when it comes to work that requires some degree of accuracy or hands-on experience, AI is not there yet and probably never will be.

You May Also Like

Quality Triumphs!

If a brand of parts has a great reputation and hasn’t caused a comeback, stick with that brand.

Last month, I was trying to diagnose a P2096 for the downstream oxygen sensor reading too lean. The cure was a new downstream oxygen sensor. But, when the new sensor was installed, I started to get a P0141 for the oxygen sensor heater circuit. It ended up being a defective part and the heater circuit having less resistance than the original. 

Being Happy Gets In The Way of Being Successful

Bryce Kenny says his greatest satisfaction comes from helping others to find the courage to chase their dreams.

Do OEM Service Bays Offer Opposition or opportunity?

With great power, of course, comes great responsibility.

It Ain’t Bragging If You Can Back It Up

It can be difficult to sit and read about yourself.

Putting Yourself First For Safety

Policies and procedures are only as good as those following them.

Other Posts

ChatGPT: Coming to a Mercedes Near You

The automaker is adding ChatGPT to the “Hey Mercedes” MBUX voice assistant.

Why Is NHTSA Involved With RTR?

A closer look into NHTSA’s involvement with RTR.

How Effective Are Non-Competes for Shops and Technicians?

NCAs restrict workers, hinder innovation, and impact employment options.

Is Your Shop Being Robbed Right Now?

How prepared are you for the bad guys?