Your breakdown of recommendation engines and their inherent limitations in wine is insightful. But what’s missing from the conversation is something no algorithm—no matter how sophisticated—can ever predict: the emotional connection a person has with the story behind a wine.
People don’t just fall in love with wine because of its flavor profile. They fall in love with the people who make it, the history behind a vineyard, the way a single bottle transports them back to a place, a moment, or a shared experience with friends and family. No LLM or data model can quantify the feeling of walking through a vineyard at sunset, hearing a winemaker’s journey, or discovering a wine that suddenly means something beyond just what’s in the glass.
Personalized recommendations imply an understanding of what moves someone, what resonates with them on a human level. That’s not something an algorithm can provide. Wine is, at its core, about connection—something technology will never be able to replicate.
This is an insightful point. In the more technical language of AI development you could say that models work purely or largely via inductive inference (with some deductive reasoning), but they have no abductive ability. There are arguments (see Erik Larsen) that mathematical reasoning may never acquire abduction, while other pioneers in the field (Judea Pearl) stress the profound challenges of developing truly abductive causal reasoning (“counter factuals”). Faith in the eventual inevitability of “scaling laws” has been severely tested in the past few days (see ChatGPT4.5 and the lack of a significant breakthrough - Gary Marcus on Substack is good here).
Reading Paul’s essay I am reminded of Neil Postman’s “Technopoly” and the idea that “Technological change is not additive; it is ecological.” These systems “work” in the sense that people stop thinking about product choice in a meaningful way, and instead dumb their abilities down to the convenience and mode of thought of the machine. And then sit in awe at the machine’s “superior” ability. Jaron Lanier echoes a lot of Postman’s warnings (given Postman wrote in the 1970’s).
As always “cui bono?” Who is championing AI’s ability to automate wine recommendation? People who’ve invested money and social capital in the ability of AI to automate wine recommendation.
First, you'd probably need to ask the companies and people leading these efforts their motivation.
Secondly, increasing vs. significantly increasing with measurable causation are two different things. The latter is a universal goal for any long-tail category - hence why Netflix spent so much money investing in improving its recommendation engine, as does Amazon, Spotify, etc, etc, etc.
The last point is sorta what I am getting at - like, there are already companies creating great ecommerce recommendation engines that can easily be leveraged successfully by wine retailers and that will continue to improve over time. (I don’t know if wine.com builds or buys, but the latter seems more sensible for a company like that.) The ROI to create some flavor/chem analysis/whatever just for wine seems iffy at best, at least with current tech. (Although it would not shock me if eventually that became easy too.)
Very insightful and well documented. Excellent article. Thank you!
Your breakdown of recommendation engines and their inherent limitations in wine is insightful. But what’s missing from the conversation is something no algorithm—no matter how sophisticated—can ever predict: the emotional connection a person has with the story behind a wine.
People don’t just fall in love with wine because of its flavor profile. They fall in love with the people who make it, the history behind a vineyard, the way a single bottle transports them back to a place, a moment, or a shared experience with friends and family. No LLM or data model can quantify the feeling of walking through a vineyard at sunset, hearing a winemaker’s journey, or discovering a wine that suddenly means something beyond just what’s in the glass.
Personalized recommendations imply an understanding of what moves someone, what resonates with them on a human level. That’s not something an algorithm can provide. Wine is, at its core, about connection—something technology will never be able to replicate.
With Gratitude,
A.J.
———
Founder - Block 55
Winery Reservation Platform
https://block55.app
Website - http://weinnotes.com
Instagram - https://www.instagram.com/weinnotes/
Newsletter - https://newsletter.oregonvinocountry.com
iPhone App - https://apps.apple.com/us/app/id1522306889
YouTube - https://www.youtube.com/@weinnotes
Apple Podcasts - https://podcasts.apple.com/us/podcast/weinnotes/id1603014320
Spotify - https://open.spotify.com/show/1CVDOdGCtTzHjCTcFuRdaL
This is an insightful point. In the more technical language of AI development you could say that models work purely or largely via inductive inference (with some deductive reasoning), but they have no abductive ability. There are arguments (see Erik Larsen) that mathematical reasoning may never acquire abduction, while other pioneers in the field (Judea Pearl) stress the profound challenges of developing truly abductive causal reasoning (“counter factuals”). Faith in the eventual inevitability of “scaling laws” has been severely tested in the past few days (see ChatGPT4.5 and the lack of a significant breakthrough - Gary Marcus on Substack is good here).
Reading Paul’s essay I am reminded of Neil Postman’s “Technopoly” and the idea that “Technological change is not additive; it is ecological.” These systems “work” in the sense that people stop thinking about product choice in a meaningful way, and instead dumb their abilities down to the convenience and mode of thought of the machine. And then sit in awe at the machine’s “superior” ability. Jaron Lanier echoes a lot of Postman’s warnings (given Postman wrote in the 1970’s).
As always “cui bono?” Who is championing AI’s ability to automate wine recommendation? People who’ve invested money and social capital in the ability of AI to automate wine recommendation.
If the goal is to increase conversion, and that is already being done with the tools we have, what is the driver to build something better? 🤔
First, you'd probably need to ask the companies and people leading these efforts their motivation.
Secondly, increasing vs. significantly increasing with measurable causation are two different things. The latter is a universal goal for any long-tail category - hence why Netflix spent so much money investing in improving its recommendation engine, as does Amazon, Spotify, etc, etc, etc.
The last point is sorta what I am getting at - like, there are already companies creating great ecommerce recommendation engines that can easily be leveraged successfully by wine retailers and that will continue to improve over time. (I don’t know if wine.com builds or buys, but the latter seems more sensible for a company like that.) The ROI to create some flavor/chem analysis/whatever just for wine seems iffy at best, at least with current tech. (Although it would not shock me if eventually that became easy too.)