In Mantel Group’s latest blog post, they dove into the experience of unexpected decisions driven by AI. 

The TLDR; of this is when Anna (an event coordinator) asked OpenAI’s ChatGPT to pair wines with meals, it suggested a wine well above her budget. Highlighting how generative AI – doesn’t always follow human reasoning.

Now, I’m more of a whisky drinker myself so I was intrigued to see what AI would suggest for me. To my surprise, the complete opposite. I gave the model a prompt of “Produce pairings of specific whisky brands for a set corporate dinner menu with a max of $100 per bottle”. 

I thought the model would generate some pairings like Starwood, Suntory Toki, Glenfiddich, even a classic Monkey Shoulder. Instead, it assumed I was budget conscious and gave me the option of (genuinely) bottom of the barrel choices, $30 dollar bottles such as Bulleit Bourbon and Wild Turkey Rye to “fit within my limit”. 

I then asked AI to remove the monetary limit to the prompt and lo and behold it gave me a 16 Year Old Single Malt Scotch Whisky, now we’re getting somewhere.

However, given this exchange it’s clear there’s not a single tastebud in its LLM’s and wouldn’t know what good whisky looked like if it walked into a Dan Murphy’s. 

It’s interesting how AI makes its own individual assumptions and personalised decisions rather than seeking clarification on preferences to deliver a more nuanced response.

Natalie Rouse mentions “Generative AI models are not always as smart as you expect and don’t reason in the same way as humans. This highlights why humans in the process where AI is making business-critical decisions is essential”.

Moral of the story? AI is still learning to perfect its tastebuds and is definitely not taking over from Whisky sommeliers any time soon. 

——

You can dive into the original article here and hear from AI expert Natalie Rouse’s on why human oversight is essential in AI-driven decisions, and how AI is reshaping business landscapes.