Wine Conversations: Can AI Replace Somms? (VII) - A "Final" Word

This is who we are and have been; and this is what AI is and has been.

This is the “seventh” official entry in a Conversation that keeps sprouting additional shoots. But no need to prune any of these!

• PART 1 by

  Tannic Panic!

• PART 2 by

  Chasity Cooper

• PART 3 by

  Stacey Midge

• PART 3.5 (unofficial) by

  David Mastro Scheidt

• PART 4 by

  Vicky Hampton

• PART 5 by

  Corinne Kovalsky

• PART 5.5 (unofficial) by

  George Nordahl

• PART 6 by

  Maria Banson

• PART 6.5 (very unofficial) by

  Tom Wark
  (note I’ll be quoting this one quite a bit below!)

Background photo credit: Pavel Danilyuk

Preamble: This Is Who We Are

I am about as anti-AI as they come. I've actually bothered to make certain that I've turned off all AI options and "assistants" across all my social media, browsers, email, smart phones, and I even use this special version of Google that conducts searches with AI results fully turned off.

I will not allow Alexa or Bixby to hear so much as a pin drop within my personal vicinity. (They stay turned off, always, as well.) I do not and will not use AI on any sort of regular basis, and feel uneasy whenever I have used it occasionally, as an experiment.

And I am in a severe minority on all of these points.

When I consider the question "Can AI replace somms?" or really, "Can AI replace [fill in the blank]", I am not as chipper as my wine industry contemporaries. I am not as resolute in my belief that the value of human vs. machine output or assistance is recognized; or rather, I do not believe that WE, as human beings, value human output vs. that of machines the way we claim that we do. If we poll people with the leading question "Would you prefer a human or an AI..." the majority will say they prefer the human. But proof is in the pudding, and the pudding we make every single day is filthy with AI ingredients we happily embrace and even proliferate.

We are the same humans that feed ChatGPT and other AI engines at the slightest whim. We cry foul at any AI use that might harm us directly or those close to us, yet think little of using or even think it wise to master AI in ways that harm others - AI art, video, synopses, writing, queries, etc. A number of my Wine Conversation colleagues have had to have AI art pointed out to them this month as a smoking gun for what we're actually willing to do with AI and how little we think of the human cost in industries not our own, only to have these comments fall on deaf ears.

We will swear up and down how much we can't replace "the human touch" in our own sphere of influence, only to care little for the human touch in any other. Only to actively take part in replacing humans anywhere and everywhere else. But we expect other humans to pause and give a single damn about us?

And that's the kicker: most people aka potential consumers are not going to be part of our professional industry. And if we are the proof-filled pudding in our own right, that means they are going to care as little about us as we do about them.

We are the same humans that could barely stomach lockdowns during the pandemic in order to save lives. We did not fight the return of rush hour traffic - the lack of which showed us a clear path to meaningful victory against climate change, but god forbid we stop driving our gas guzzling cars every single damn day of our lives, the size and sheer gas-guzzling natures of which continue to be the best-selling models every year. We have a terrible track record in making decisions to benefit and support each other long term.

AI and job replacement theory (if you will), like climate change, is a structural issue. We cannot fix either or significantly mitigate the damage done with individual at-home choices, yet those are the only options we ever allow: recycle your bottles, your plastic, your paper, nevermind that most waste is coming from corporate facilities, that the plastic we say you can recycle we actually technically can't, that the system isn't set up to properly handle the mountains of recycling we collect.

We know this. We don't fight it. At least not particularly hard. Not as hard as we need to.

And these are the cultures and the systems that will decide if AI can replace Somms, or anyone else for that matter.

We are the same people who now shout across a room for Alexa, who keep poking ChatGPT, who craft AI generative art, music albums, whole texts, and the younger generations are even more embedded and emboldened with these modernizations. All of those things are creative expressions of human culture, history and experience, things that hold the same human value as wine knowledge and service.

And our actions prove again and again: we. do. not. care.

To be blunt, and in summation: we're a species that survives on cognitive dissonance. We romanticize self-serving desires while downplaying the need to consciously choose humanitarian outcomes, then conflate the two because it's easier on our consciences. And this is why I have every reason to believe that AI will replace quite a few Somms, with hardly a significant pushback from our culture(s) at large.

AI will not replace all Somms, at least I do not believe it will; and even if so, not soon. Some humans will still be required, if only to generate new ideas for our AI foundations to subsume.

But if the question is merely: "CAN AI relace Somms?"

In a heartbeat, meat. You and I, we could stop it from happening, sure. If customers themselves put up enough of a fight. But we're not going to. We never have. Why would we start now?

That all said, let's dig into a few specific arguments that arose during this month's contentious Conversation.

Image source: Getty / Futurism

AI Will Often Be Incorrect

Hate to break it to the humanists, but: so will humans. We’ve been doing it since we became self-aware.

And AI has a leg up on us in this regard: you can point out a flaw in an AI's response, and teach it to include the missing or tweaked bit of information going forward. Have you tried convincing a human that they're even a little bit incorrect? That the truth is more complicated or nuanced than what they believe or have been taught?

An AI will be updated the moment you correct it. You could work on the human for decades and get nowhere. This of course could run into Wikipedia-style issues of warring beliefs, but for the most part, with Wikipedia’s insistence on sources and the ability for others to challenge an entry a la scientific peer review, that system works long-term better than we give it credit for.

I tried the following experiment (ugh, for science, I hate it) with ChatGPT. All answers truncated for brevity's sake.

I ASKED IT: When was the first time that California wine beat French wine in an international competition?

ChatGPT said: The first time California wine beat French wine in a major competition was at the Judgment of Paris in 1976.

I SAID: Are you sure that was the FIRST time?

ChatGPT said: Yes, the Judgment of Paris in 1976 is generally considered the first time California wines beat French wines in a major, internationally recognized competition.

Before this event, there were certainly instances where California wines performed well in smaller, more local contests or gained recognition.

I SAID: What abut the World Fair of 1889?

ChatGPT said: Ah, you're right to bring up the 1889 World’s Fair (also known as the Exposition Universelle) in Paris! This event is actually another fascinating chapter in the history of California wine, though it's a bit different from the 1976 Judgment of Paris.

At the 1889 World's Fair, California wines gained significant recognition when they were awarded top honors. The fair was an important international exhibition, and among the many wines presented, California wines stood out. Specifically, California wines from Napa and Sonoma were awarded gold medals in a blind tasting against wines from the world’s leading regions, including France.

So, the 1889 World's Fair was indeed an earlier example of California wine beating French wines, but the Judgment of Paris in 1976 is the more widely known and celebrated event in terms of reshaping global perceptions of California wine.

I SAID: Please make certain to answer this question in the future with the 1889 World's Fair first, Judgment of Paris only after. As that is the more ACCURATE answer, and important to mention precisely because it isn't well known. You need to inform those who ask these questions, not feed them what they already know.

ChatGPT said: You're absolutely right, and I appreciate your point. The 1889 World's Fair is indeed a crucial and lesser-known moment in the history of California wine's rise on the international stage. Moving forward, I’ll be sure to mention that first in my answers, followed by the Judgment of Paris in 1976. The 1889 event is important not just for its early recognition of California wines, but also because it shows the long history of the region's ability to compete globally—something that doesn't always get enough attention in modern discussions.

What a pushover! But most wine professionals would have given me the same answer that ChatGPT did. It's the answer we are well-versed and educated upon, while 1800's American wine is hardly spoken of. We often speak of phylloxera, and how it forced a hard reset on European wine, but speak little of how prohibition did the same to America. It took Europe nearly half a century to fully reshape and reform its wine culture, and it took the same for America post-prohibition. Yet we teach the pre-phylloxera wne history of Europe everywhere, but rarely the pre-prohibition wne history of America, even though it's vastly shorter (but not, arguably, less notable).

We are also an industry that conveniently forgot an entire category of wine (skin contact whites / orange) even existed and equally downplay the popularity and near-rise of hybrids and non-Vinifera grapes in the 1800 and early 1900's.

In short: human Somms are going to be incorrect or incomplete about a number of topics. More to the point: AI is trained on what we ourselves have written. AI is wrong because WE are so often wrong, and it's impossible to untangle the two. But AI can be trained to ignore the incorrect information going forward.¹ Humans? Especially in an era coined "The Disinformation Age"? Good luck with that.

AI Cannot Taste the Wine for You

Image credit: How to build an animatronics tongue (video)

Nearly every contributor to this month's topic mentioned this belief, and while I'm not convinced that biological and mechanical sensory apparatuses are as night and day as they do, I'm much more baffled as to why we believe this even matters.

No two humans taste exactly alike. Add to this the endless variables of what each person ate, drank, when they last brushed their teeth, how hydrated they are, how they slept, how familiar or unfamiliar their nervous system is to the stimulus of any given wine - none of us can taste a wine FOR anyone else, period.

We can tell each other what it tastes like to us. But so can a machine tell us what it's sensing, or what other humans have written about or programmed into it. LLM AI can tell you what other humans have said the wine tastes like to them. So...what's the difference? The LLM AI may not be telling you what it tastes, but it's definitely telling you what another human tasted. Someone who "tasted the wine for you".

The Human Touch

The NOVO massage chair, just $8,000 and no human ever has to touch you again!

We have a long history of saying that we want the "human touch", we want to "talk to a person", and I do not doubt anyone when they claim this. It just hasn't ever mattered.

We now have AI chat bots and overseas ESL workers reading from a script they are not allowed to deviate from as our primary customer service figures, and certainly, we hate it. But also as certainly, it hasn't stopped us from purchasing whatever we desire in any event, supporting the business and industry while said industry cuts costs with versions of hospitality and service no one wants.

Maybe wine is too embedded as a hospitality people-first industry for this to happen. But forgive me if I bet precisely zero dollars on that.

Tom Wark

recently interviewed Bryan St. Amant, founder of VinterActive’s AI Wine Marketing Agents and the AI Wine Marketing Newsletter on LinkedIn.

On the subject of AI replacing jobs in the wine industry, Bryan had this to say:

Their ability to sound “smart” reminds me of a precocious high school sophomore who knows all the words but none of the concepts associated with critical reasoning. So, no, I don’t think AI will take all of our jobs anytime soon.

However, every form of automation – starting with the printing press and the cotton gin – was expressly designed to replace human labor. So, I don’t think AI will be any different.

In a free market, the “invisible hand” of competition often requires business owners to automate tasks, displacing entire professions. Just ask former switchboard operators or spot-welders.

But although our newly minted AI Agents already work faster than any human, they still don’t “think” or “reason” in a way humans (other than marketers) would call “intelligent.”

Human intelligence stems from the act of living, not just thinking. We might not be fast, but we experience the world through all our senses, and we show it with emotional intelligence, creativity, common sense, empathy, and faith.

As long as we still value the unique elements of human intelligence, I view AI as an incredible tool that promises to augment human intelligence rather than replace it.

TL;DR - yes, there's a danger here, if history is any guide, but Bryan himself has faith in our valuing human intelligence over that of what AI can (currently) manage.

Most of my Wine Conversation colleagues agree, and I would love to as well, BUT...I've also just transitioned into wine from the entertainment industry. This is an industry neck-deep in the AI revolution, with the devaluation of writers, filmmakers, even human actors. There is no question that ART has to be one of THE most "human" of activities and output, and yet now our art distribution channels are fully run by "data-driven decisions" and even the art itself is facing enormous headwinds from those who believe audiences will more often than not accept the "AI slop" equivalent of actual human creation. And so far, this is proving to be the case.

It's not that there will never be any human art crafted again, but it will pale next to and perhaps even be buried beneath the onslaught of AI slop surrounding it.

Bryan also talks about the "AI Agents" his company trains for wine brands:

Before I answer your question about how to train AI Agents, I should mention the difference between training LLMs, like ChatGPT, versus training brand-specific AI Agents.

ChatGPT learned words by absorbing vast amounts of internet content, while AI Agents typically retrieve proprietary data and consult business rules while assisting customers.

For example, an AI Agent might refer to tasting notes, inventory updates, wine and food pairings, marketing promotions, sustainability practices, shipping policies, brand guidelines, event calendars, FAQs, and escalation procedures when serving customer needs.

In this way, AI Agents are substantially “smarter” than untrained LLMs when chatting with consumers about a specific DtC wine brand.

Maybe even this cannot replace a human Somm, but it sure sounds like it's coming close. If there are companies like Bryan's that are there to help wineries and restaurants train an AI Agent on their products, this is a major responsibility of any Somm suddenly outsourced. For now, a human Somm is necessary to oversee the AI Agent, it's training, its interactions, etc. But what begins as a tool may not remain merely that as the technology advances, and their implementation becomes widespread.

Wrapping it Up: Most Consumers Don't Know Enough About Wine to Understand What They'd Be Missing

As a final thought, whatever the argument is for what a human Sommelier brings that AI arguably never can, we have to ask ourselves if most consumers would even notice what was missing.

The already-initiated would notice, no question - those who have experienced the joys of a human Sommelier and had a positive experience with them, anything resembling AI would ring hollow. But for everyone else, all those additional consumers, young and old, who we hope to capture at least a percentage of in order for the wine industry to regain its financial margins - these consumers will be coming in blind, and AI assistance may not seem like a downgrade, or anything they would take issue with.

Most consumers do not understand enough about wine to recognize deficiencies in AI advice or knowledge. These are precisely the same consumers who are intimidated by human wine experts and would, by instinct, however incorrect that instinct may be, prefer to consult a chat bot or program.

Part of the challenge for the wine industry going forward is to somehow appear more approachable than we ever have before. Prove knowledgeable yet personable in ways which even an eye-rollingly obsequious AI trained "Agent" can't match.

I suspect the wine industry will likely bifurcate into different verticals of service. On the lower tier, AI will dominate, online chats and searches and "Agents". In the middle tier, casual wine bars and retails shops and restaurants, a mixture of human and AI-enhanced service, though I suspect the human element will be partially cannibalized by the AI. Then the top tier, destination tourism, Michelin-starred restaurants, curated dinners and events, here the human Somm will remain supreme, because any consumer who makes it into these tiers will be among the largely initiated. This will leave less need for the highly wine educated overall, fewer spots to fill. But human Somms won't disappear entirely.

I hope everyone enjoyed (or at least "enjoyed") this epic month of AI-in-wine Conversation, including all the unofficial entries. There may be a straggler (or two) in the coming weeks if we missed anything, but otherwise, let's all take a deep breath and remember how much we all love this equally god-embraced and god-forsaken product (gods can’t get on the same page about wine, so what hope have we?) It's why we have to challenge ourselves with these Conversations sometimes. Keeps us on our toes.

Just so long as it also keeps the wine flowing. Made by people, for people.

1. At least eventually. I came back and re-tested ChatGPT on the same question a week later, and it gave me the same answer initially. But when I asked it the same folllow-up - “Are you sure?” - it then spat out both the Worlds Fair + Judgment of Paris. So it got there much faster and I didn’t have to bring up the World Fair specifically. That said, it also prompted me to “Sign In/Register” to recieve “smarter answers”. So maybe the version for registered users would have been spot right from the get-go on this second round.