Eleven Questions for the Show Floor
An AI field guide for HITEC week
By now you know the why (links below, if not). Models guess, the data underneath is the real product, and "the user is logged in" is four different things in a trench coat.
So here's the payoff: one card you can use between the espresso bar and the demo theater. Ask these at every booth. Not to trip anyone up. The good vendors will have good answers, and this is how you find them faster.
How accurate is it? How did you measure it, and whose data was used?
Who's running this in production today, and can I call them?
When it's wrong, how do I know? Who's accountable and who hears about it first?
What's the blast radius? What's the most harm this can do without a human in the loop?
Show me how I'd review what it actually did. Live, in the demo.
When it acts in our systems, whose access is it using? Can we revoke or reduce its capabilities?
Where does our data go, what's done with it, and how long do you keep it?
What would be the total cost to deploy across my portfolio, not the showpiece? All in: integration, deployment, and staff trained?
What does it do when it isn't sure? Pause? Guess? Act anyway? Is the bias toward action or passivity?
In two years, when the technology has moved on, what do we still own?
Does any of my data train your system, or your model provider's system?
Be curious. Be cautious. Be thoughtful. Take the demos, enjoy the espresso, and watch which booths lean into the hard questions. Those are your shortlist.
The rest is kabuki.
The why, if you missed it:
- Why models guess — on probabilistic systems
- The data is the product — why the foundation matters more than the model
- Four things in a trench coat — what "the user is logged in" actually means
- The Human Division — replacing undocumented tribal knowledge