|
AI makes it dangerously easy for me to think that I am something I am not. A scientist, a doctor, a lawyer, a trip planner, an AI implementation expert…
But as we built our internal AI hub, I used a thought exercise with the team to highlight AI strengths and weaknesses so we could maximize its effectiveness.
You are standing in front of a 50 story office tower. I ask you to go find a specific red pen that is somewhere in the building. Good luck, right? You methodically start covering every corner of each floor. Each closet. Maintenance room. Restrooms. Office. Desk drawers. You clear a floor, and move up to the next floor.
This is brutal. It’s inefficient and time consuming. You are wandering around hoping to stumble across this stupid red pen. How long will that take? 150 hours?
Even my LA bestie Warren with his Yale/Harvard degrees can probably figure out that the building is a metaphor for structured data and the person is a metaphor for AI, but it takes a state school education to understand why that matters.
The red pen example is deterministic. “I need one single thing and it can’t be wrong.” For all its awesomeness, AI is worse at this than the more traditional ways we’ve always done this. Without clean, structured data, AI wanders the hallways hoping to find the pen. And when it returns, it will tell you with total confidence this is the pen you wanted.
Have you ever asked AI the exact same question and gotten two slightly different answers? It went and grabbed a blue pen. Or a stapler. Maybe even a ham sandwich out of the fridge in suite # 2707. If you get this wrong, you lose.
Let’s start over again but ask a different question. “How many red pens do you think are inside this building?”
Probabilistic.
This is AI’s wheelhouse. |
 |
|
The key is to point AI where it adds the most value. Too many of our teammates just dumped all their files and spreadsheets into AI and started asking questions. The answer seemed right so they assumed it was right.
When you ask AI to search the entire 50 story office building, there’s a good chance it brings back a red pen that you’re pretty sure is the right one…but is it?
Plus, cost is eventually going to be an issue with AI and we were spending tokens on the wrong thing. Asking AI to search the entire office building is a waste of money. We were being lazy…and AI encourages that! That’s how they make money!
Don’t spin your AI wheels (and waste your money) on deterministic outcomes. Create the structured data. Invest in mapping so it doesn’t need to scour a 50 story building for a red pen. Structure data and AI are a partnership, it’s our job to pair them effectively.
Let’s revisit the original question, except this time I include detailed instructions, “Go find a specific red pen that has a piece of tape wrapped around it. It is on the 17th floor, suite 1707. Take a left off the elevator and then a right down the hallway. The office will be halfway down the hall on the left. You will find the red pen in the top left drawer.”
How long does that take? 3 minutes? That’s a 3,000x improvement over the 150 hours.
Deterministic - old fashioned structured data
Probabilistic - AI leverages the structured data |
 |
|
PS - remember when I suggested it might take 150 hours to find that pen? Guess where that came from? |
 |
|
|
 |
