The Sommelier of the Digital Age
Why better answers still begin with better questions.
5 min read
A friend of mine is a top sommelier.
Yesterday he made a mistake.
He left me alone with his bookshelf.
I found a beautiful book called Gourmet Food & Wine Pairing Around the Globe and immediately opened the FAQ section.
I was expecting the important questions.
I have $50, a first date, and no personality. What should I order?
Or:
I just got promoted. What do I bring to dinner that says successful but not insufferable?
Instead:
How long should I chill wine?
How do I know if a bottle is corked?
I stared at the page.
Humanity has spent roughly eight thousand years making wine and this is where we landed.
Temperature.
I had the sudden urge to remove the page for public safety.
Then I realized the book wasn’t the problem.
We have become very good at getting answers to very boring questions.
And AI is about to make us spectacularly better at it.
Answers Are Cheap Now
For most of history, knowing things was impressive.
You had to read books. Find experts. Remember facts. Occasionally travel somewhere inconvenient.
Now I can ask AI to explain quantum mechanics while standing in line for coffee.
This should have made humanity unbearably intelligent.
So far, we mostly use it to rewrite emails.
Progress is complicated.
AI has made answers astonishingly cheap.
Which means the expensive part is moving somewhere else:
What should I ask?
Because there is an enormous difference between:
Tell me about Burgundy.
and:
Why does Burgundy make otherwise reasonable adults discuss dirt for three hours?
Same AI.
Very different evening.
The quality of the answer still depends surprisingly often on the quality of the curiosity that produced it.
Wine Figured This Out Centuries Ago
Ask a sommelier:
What is the best wine?
and watch a small part of them die.
Best for what?
Dinner?
A date?
Tuesday?
Your billionaire father-in-law?
A picnic where someone forgot the corkscrew?
Context changes everything.
A $1,000 bottle can be completely wrong.
A $30 bottle can save the evening.
Good sommeliers understand something we are only beginning to understand about AI:
The best answer does not exist independently of the situation.
The food matters.
The person matters.
The budget matters.
The mood matters.
Sometimes even the weather matters.
And sometimes a guest says:
I have fifteen minutes for lunch. Make it unforgettable.
I once had exactly that request.
Now that is a good prompt.
Constraint. Intention. Context.
Something to work with.
Nobody asked me for “the best wine.”
They asked me to create an experience.
Prompt Engineering Has a Branding Problem
This is why I find the phrase prompt engineering slightly funny.
It sounds like someone should be wearing safety goggles.
Much of it is simply the ancient human skill of knowing what you want and being able to explain it.
Which, judging by meetings, marriages and restaurant orders, was never particularly widespread.
Ask AI:
Write something about wine.
It will.
Unfortunately.
Ask:
Explain why intelligent people confuse expensive wine with good wine. No jargon. Give me one example that would annoy a wine collector.
Now we have somewhere to go.
The AI did not suddenly become smarter.
You became less vague.
This may be one of the more uncomfortable things AI is teaching us.
Sometimes the disappointing answer isn’t evidence of a stupid machine.
Sometimes we asked a lazy question.
Terrible news for everyone involved.
Knowledge Is Not Judgment
A mediocre sommelier can memorize 5,000 wines and still ruin your dinner.
A brilliant one may know exactly when the legendary bottle should remain downstairs.
Knowledge matters.
But knowledge without judgment is just a very expensive inventory.
AI now gives us access to an absurd amount of knowledge.
So the competitive advantage shifts.
Not toward remembering more.
Toward selecting better.
What matters?
What doesn’t?
What is technically correct but completely uninteresting?
What assumption is missing?
When should you challenge the answer?
When should you stop asking?
And when should you ignore the machine entirely and open the bottle?
That last skill may become increasingly valuable.
Taste Is Selection Under Abundance
Taste sounds decorative.
It isn’t.
Taste is selection under abundance.
When there are ten bottles, choosing is manageable.
When there are ten thousand, judgment becomes valuable.
AI is doing the same thing to ideas.
We are entering a world in which producing another answer, paragraph, image, strategy, presentation or opinion costs almost nothing.
The problem will not be making more.
God knows we have enough PowerPoints already.
The problem will be knowing what deserves to exist.
Which answer is worth keeping.
Which idea deserves another hour.
Which connection is interesting.
Which beautifully written paragraph should still be deleted.
That is taste.
And AI cannot outsource it for us without creating a rather obvious circular problem.
Four Empty Glasses
When I looked up from the book, there were four empty wine glasses in front of me.
I was supposed to be polishing them.
Instead I had somehow turned them into a theory about artificial intelligence.
This happens more often than it should.
But perhaps the glasses had the right idea.
An empty glass is not a problem.
It is a question.
What belongs here?
The same may be true of AI.
We keep obsessing over how intelligent the machine is.
I suspect the more interesting question is what happens to our intelligence when answers become unlimited.
Because the future probably won’t belong to the people who can generate the most.
It may belong to the people who know what is worth asking for.
Before you fill the glass, it helps to know what you’re thirsty for.