AI is not new – it has been with us for a while–first coined at the Dartmouth Workshop in 1956. While AI is not new it is a hot topic especially in the past two years as noted in graph below:
I am not in the AI ‘scare camp’ as being the ‘end of everything’ but I am much more so in the ‘careful, judicious, and realistic’ camp. I have seen demonstrations that were compelling and I do use and test AI products more than I ever have before. And I do see cautionary notes about potential inaccuracy in AI output; as well as an abundance of caution by those who use it for teaching.
So when I am speaking of wine and AI, it is not just from reporting on the marketing outputs of a wine producer but the totality of operations of a wine producer. Wine producers don’t have just PR, media, marketing and influencer functions; the core of business activities go into areas less covered by many wine journalists.
A wine producer additional business activities include:
- HR
- Finance
- IT
- Regulatory compliance
- Supply chain
- Logistics
- Viticulture & Viniculture management
While AI is a hot topic and even with news stories (as shown below) it seems all we have to do is ‘get to it’ and start using AI right away. But there is much more to the subject of AI and wine than just an immediate adoption.
“…AI will be a game changer”
“AI & Wine: An Unlikely Pairing for a Perfect Vintage”
“How AI will Revolutionize the Future of Wine”
There seems to be no issues or downside with using AI in a wine business by the articles glowing titles. But there is a sense of “if something sounds too good to be true it probably is.” And yes there are plenty of stories about the downside of AI as well. AI as a ‘Game Changer’ and ‘Revolutionizer’ is way too early to see if AI will be this leap forward and worth the vast expense they promise to be.
The questions I have are more practical and I see AI’s shortcomings that are apparent already. The use of AI-generated scripts or edited videos are apparent on YouTube today. I have experimented and found the realities of AI to be quite limiting, inefficient, and low quality output. So to use HR performance management parlance AI so far “Partially meet Expectation.”
There are some fantastic YouTube shorts: there are YouTube producers who craft compelling ‘shorts’ and do so from scratch and then there is equally a large number that have been created with the aid of Opus Pro, Clips AI, Qlip.AI, Vidyo.ai etc.
The repurposing of video content is to take long YouTube videos (longer than a minute) into easy to consume one-minute clips. Generally, each video repurposed will take a long video format and create one-to-many one minute videos (edit, caption and score). Sounds like a superb use of content and to maximize exposure of videos but you’ll see the reality is different as I described my experience below.
My experience:
I trialed one AI video product and had no opinion of that particular product prior to testing. Here is what I did: I had a 90-minute trial (which means the total time of videos submitting for repurposing) and used up the entire amount for two different video types – 1) wine and 2) business. I wanted to see if either type had a higher success rate.
Here are the results:
- I received an output of 51 videos clips and evenly divided between wine and business topics
- I received only 5 usable video clips: 2 wine videos and 3 business videos
- Usability rate was under 10%
- The AI product scored it in 90+ range meaning a high rate of success and viralability (a higher than average click and total watch time rates) of the 5 published videos
- The performance of the YouTube shorts performed approximately 50% less than non-AI shorts (generally YouTube shorts perform best upon release and do not behave as evergreen content)
- Given the amount of time I had to use to review each video for quality and usability and even additional edit work I had no time savings whatsoever. AI has a hard time to distinguish “Palate”, “Palette” and “Pallet” and almost never picks the right word for brand names, wine varieties, place names and more
While I am judicious about videos that I upload to my channels I am not so sure other content producers are; we have seen an intense growth rate in YouTube shorts. Some of the content is so incoherent that I am not sure the point of those videos. YouTube is aware of this practice of AI generated content; and the question is what will they do when this content type will be so dominant that it renders the category “shorts” meaningless i.e. reduced click/view rates. There is still a rich reward for AI sharpened content through algorithmic recognition and highlighting this content in search and next video recommendations.
I am not optimistic that YouTube’s algorithm will stop highlighting AI-generated content anytime soon. So if content is not AI generated the closest thing to mimic the algorithmic experience is captioning and super tight and fast paced editing.
I am using software to help transcribe for captioning and it is completed in the 70% range of accuracy meaning I am missing one quarter of content accuracy; it is very time consuming to view each captioned frame for accuracy (and make updates/edits). So I am doing captions as sites like YouTube, Instagram and so fourth do captioning but it is hard to read and not compelling and also has issues with accurate captioning as it relates again to brand, variety and geography points and more. So many people might read the captions as a way of consuming video versus listening and watching. I have not seen any time savings whatsoever from AI as it relates to my video production.
***
But this is not where the story ends…
AI’s potential pervasiveness in the wine business has not yet had a profound impact. And as I stated above there is so much that AI could do in any organization.
There are a number perspectives to keep in mind:
Accuracy: I hear all the time AI “is learning” and improving all the time in terms of accuracy. But I do find accuracy to be a concern. While I have expertise in several areas I can quickly pinpoint problems with an AI generated “answers” and know that it is not accurate or comprehensive. I have had to ask many permutations of my query to hone in on the most accurate answer. AI systems are only as useful as the input is accurate and can be utilized for a potential approach or even solution.
Expense: AI is not cheap; both in actual costs and cost to review output. For actual costs will, of course, depend on the SaaS AI product you will be utilizing – one model is to bill in terms of minutes purchased (the new long distance calling rates). Because minutes are the business subscription model, the costs are quickly consumed and the resulting work may not even be at a 50% usable rate (I suspect that number is half of that perhaps in the 10-25% range). Also, the cost of inspecting work or having someone that works for your organization costs money in terms of their salary and benefits–and ultimately it is now more work than was present before AI. AI will not be equivalent to the old fangled objects like a VCR or even a microwave that came down in price after mass adoption. Technology expenses, at least, in the business sector has doubled in 10 years to $567 billion dollars.1
Energy: AI is expensive because it takes a considerable amount of energy to complete an AI query which can use between 0.0017 and 0.0026 KWh of electricity; compare with a Google query which is 0.0003 kWh (note this is a reported number by Google itself). If a wine producer is working towards reduced-to-zero emissions how does this fit in with the producer’s overall plans responding to earth friendly commitments?
Explainability The ability to understand and interpret the predictability ascertained by AI models. How does AI determine one potential approach or answer over another? It is hard to formulate or even understand how that was determined. Also, is there enough of a data model for a wine business to make the best use of AI and does AI provide the best possible solutions now and ongoing?
Reliability: I have engaged software providers and of course their SLA time shows they are up 99% of the time, however, reality can be quite different. Try to prove this can be difficult and providers will often say they are abiding by their SLA. I was using Open AI on September 26, 2023 and there were issues when I was using ChatGPT:
There are many other issues that AI needs to address to be considered a trusted and viable solution for any enterprise large or small. Below is a graphic I created to look at the large landscape of issues:
And these are top level problems and there are many more than can be very specific to a wine producer. Perhaps not enough data for an AI product to give meaningful or even realistic responses to an issue.
Think of an HR ATS (applicant tracking system) and overfitting. Ahh overfitted (a very AI term) that describes how narrow a search might be for a specific open FTE position and candidate pool. While the intent is to get a qualified candidate a whole pool of candidates might be eliminated where they might be completely applicable. Overfitted ATS might identify a talent pool that is lacking in diversity and inclusion, poor experience by the applicant community, potential ethical and legal liability and increase in employee turnover amongst other issues.
And there are many business needs a wine producer has and if engaging AI might not increase efficiency and add an overall business expenses without increasing profitability or being able to get more work done. A perpetual motion machine has been tried many times before and all have failed. AI like any other system or software does not take care of itself or maintain itself i.e. there are no perpetual motion aspects to AI – it requires considerable maintance plan.
The rise of AI does not necessarily mean we have a revolution before us and perhaps a system-by-system potential. Instead of a revolution it will be an evolution in some AI application in a wine business but not be a universal and total AI deployment in all aspects of any business.
I remain in the “wait and see” sideline to see if AI will be all that it promises to be. Tempting for a wine business to engage AI but note that the benefits are much less than a promised “revolution.” and like any other innovation will need to be managed
My eyes have been widened by my recent testing of AI for video production and less-than-excited as it certainty hasn’t lived up to the hype. And ChatGPT, the AI darling can be helpful but I do have to do numerous queries to get to an accurate answer. I am very curious about ChatGPT‘s explainability and my repeated refining and repeated queries gets to a decent output but it takes considerable effort: how or will this ever be improved?
Answers and approaches take great effort and human knowledge: sifting gold from sand. I would urge testing of potential solutions to see how it might benefit your business before fully deploying an AI solution in your business or endeavor.
Thank you,
James, JamesTheWineGuy
1Beyond Silicon Valley, Spending on Technology Is Resilient, New York Times,
Updated Feb. 19, 2023
© 2023 James Melendez / JamesTheWineGuy— All Rights Reserved – for my original content, drawings, art work, graphics, photographs, logo, brand name, rating, wine taxonomy, and all designs of JamesTheWineGuy. JamesTheWineGuy is also on Facebook, Twitter and most major social medias.
James Melendez - JamesTheWineGuy 