Ep. 2062 A discussion on how Artificial Intelligence | wine2wine Business Forum 2023

Who wants to be the next Italian wine Ambassador? Join an exclusive network of four hundred Italian wine ambassadors across forty eight countries. Vineetly International Academy is coming to Chicago on October nineteenth is twenty first. And Walmatikazakhstan from November sixteenth to eighteenth. Don't miss out. Register now at Vineetri dot com. Official media partner, the Italian One podcast is delighted to present a series of interviews and highlights from the twenty twenty three one to one business form, featuring Italian wine producers and bringing together some of the most influential voices in the sector to discuss the hottest top fix facing the industry today. Don't forget to tune in every Thursday at three PM, or visit the Italian wine podcast dot com for more information. So welcome to this very special section. First of all, thank you, Rajin, to prepare a very wonderful presentation. It's a very interesting subject. And just to introduce you, I I do something that I really learned yesterday. Sorry if I just did it only yesterday, but I was preparing a presentation for yesterday night and my little daughter that she's eleven and she's double. So she's speaking Italian English on the same way as I didn't. I don't do it. And, I asked her to say, please, can you help me to do a translation about my speech? And she say, okay, dad, but I want to do something. I translate for you, you're Italian. And after that, we put your Italian speech on the GBT and we try to see if they do better than what they can do it. And I was really, really impressive about how much the two documents they were very, very close. This is for sure something that we need to fix. But I really learned that we have an incredible amount of opportunities in front of us. It is just to use. I think that is a matter of our flexibility that we need to train every day in order to improve actually the opportunities that we have in front of us. It's just a matter of be open. And I in the last experience of the last seven, eight years I got some chance of understanding what does it mean really to open your idea, try to do new things. So please tell to us what you think about these incredible new opportunities that you have in front of us. So by the way, sorry. I'm Andrea Ardi. I became recently master of wine. So, and I want just to introduce briefly myself. I'm a CEO of Angelini Wines and Estates. I have a background as an agronomist. And after that, I started to make wines and started to sell. So I covered different position between production, sales and marketing in Angelini Wines and Estates. And since I am, if I come from verona, welcome again in verona, and I enjoy for you. I hope for you a very lovely stay around the town and especially by Patricia Lara. Thank you for coming. Thank you very much, Andrea, for the introduction. Hi, everyone. Looks like we have a full house today. Fantastic. My name is Regina Lee, m w, and I'm going to talk to you today about artificial intelligence and how it will change the way we work in the wine business. So you're probably wondering why am I qualified to talk to you today about AI, and I'll admit I'm absolutely not qualified to do so. I am not an early adopter to technology at all. So my phone is always three years out of date. This is a Samsung Galaxy S twenty one, I think. I don't even know. I'm not on social media a whole lot. Sorry, TikTokers. I am on Facebook. I go occasionally on it, and, you know, it still says I'm in a relationship in my status even though I've been married for thirteen years. So again, I'm not an early adopter to technology. I'm not passionate about technology in itself. However, I am passionate about the wine industry, and I'm committed to find ways for us to evolve and grow while we're facing some existential crises. I'm talking about lower consumption, higher costs, dealing with major geopolitical issues that might affect us, in terms of transportation and production, as well as climate change. So we're facing some real challenges, and we need to evolve and adapt to the times. And I believe AI can help us do so. Just a quick introduction to myself. I joined the wine industry in twenty ten, and I worked in a variety of different roles. I worked at WSTT, spreading wine education in Asia Pacific, And I spent the majority of my wine industry career in import and distribution. I worked for many years at Liberty wines, that wonderful Italian wine specialist importer based in the UK, and currently, I'm the managing director of a boutique imported distributor called indigo wine, and we work with small scale artisanal producers using low intervention methods. But all to say that I've had a variety of different experiences in sales, marketing, managing operations, finance, etcetera. And even before I joined the wine industry, I was in management consulting for a number of years. And when I put on my management consulting hat, and I look at the wine industry, I see that the wine industry is rife with inefficiencies. Right? We are an industry that is full of passionate people, intelligent, and wonderful, great communicators, but we're not often technocrats. Right? We don't have a lot of those technical skills to help us be more efficient at our jobs. So this is where I think AI can help us because when we're facing these challenges, as I mentioned before, being profitable, managing our cost base, and being more efficient will be more important than ever. So AI has an enormous potential to help us, but we must be cautious. So in this talk, I'm going to firstly talk a little bit about AI, what it is, and how it works, then I'm going to talk about how it will be applied to the wine industry or my views on what we can do immediately. I'll talk a little bit about the risks and the issues with AI. And then, lastly, I'll conclude with my call to action and what we can do tomorrow or even today after this talk on how we can harness AI for our good. Okay. So let's start by talking about AI and particularly generative AI, which is the focus of this presentation. So firstly, generative AI creates content. It doesn't just rank it. So we're familiar with using, of course, Google, Amazon, Netflix, so these algorithms rank what we would most likely want to watch or buy or consume based on our preferences, our search results, our search histories, and what we bought or consumed before. The AI is fundamentally different because it creates that content that we're consuming, not just ranks it. Secondly, it learns, but how does it do so? Essentially, what we what generative AI is based on is deep learning and large learning language models. And what this means is that programmers feed AI large training data sets, and AI learns to see patterns in these data sets and apply these patterns to other data sets. And so when you prompt AI by asking it a question or asking it to generate something, it will scan these datasets, look for what is statistically going to be the correct answer, the answer that you will most likely want and feed that back to you. And it continually learns from this. So it's continually evolving. And the analogy that a lot of people use is that if you are teaching a young child, I've got two sons, you know, five years old and two years old, And if you teach them things in repetition, for example, I, you know, sing nursery rhymes to my sons, and they learn to to these songs, they learn the words. And so if I prompt them by singing twinkle twinkle little, They will finish my sentence by saying star because that is the most likely word that's I'm going to want to have them say when I prompt them. So large language models learn from this. You feed them data. They spit out a response. Based on what it thinks that you're going to want to know. So that's how it learns. And then the third thing, which is very transformative for us, is that it speaks naturally. So we don't see things as kind of a ranking system, again, a list of you know, movies to watch or websites to click through. It speaks to us as if it was a person responding back. And this is where some of the risks and the scary things about AI kind of come creeping in. So how do we apply AI to the wine industry? And I think this is a very interesting question because wine and AI are very different. In fact, opposites, right? You know, wine is complex. It's variable. It's based on sensory experience. It speaks to what we love as human beings. The AI on the other hand is based on having things that are predictable, repetitive, and automatic. So how do we marry these two things together? Well, I think the answer is is that we use AI to automate things that we do in our daily lives and work that are manual, repetitive, and can be predictable to make our daily work more efficient so that we can spend more time on the other side what makes wine beautiful that complexity, the variability, and speaking of the stories, and the winemakers, and the consumers of the product. So I want to start by looking at some immediate areas that we can explore within the wine industry. And I'll talk about each of these things very briefly. The first is marketing, and I think there's a huge opportunity here, that we can harness pretty quickly. I'll give you an example I have a very good friend, his name is Kevin Chung. He works at a AI startup called writer AI, and they generate marketing content. One of their biggest clients is L'oreal. And you can imagine that L'riel has thousands and thousands of products across beauty, makeup, personal health, etcetera. And it sells to hundreds and hundreds of different channels. You know, various different kinds of retailers, supermarket chains, pharmacies, makeup artists directly, etcetera. So you can imagine how much time their marketing team spends on creating marketing content, taking product data, and formulating marketing content to sell across the different channels that L'oreo has. So their marketing team spends hours doing this, and writer AI has come in and developed, a language model that takes L'riel's house writing style, and it has been fed, L'oreal marketing material to get a sense for the tone of writing, the words, and how the marketing materials are tailored to different channels, and it consumes all the product data, those thousands and thousands of products, lots of different data points, and it automatically spits out marketing content tailored for all its channels. So we used to take hours for the marketing team every day now takes minutes. There's a lot of applications to this for us and wine. You know, think of all the fact sheets, website content, blogs, etcetera, etcetera, or priceless that we have to produce. This could be automated very, very easily, very soon. Second, forecasting, and I'm here. I'm talking about forecasting of demand planning, production, supply, profitability forecast, etcetera. So pretty soon there will be AI tools. In fact, Microsoft is developing quite a lot of suites of AI tools for its own products like Excel, Power BI, etcetera, to help us automate some of these very laborious tasks and quantitative modeling. So my operations manager at Indigo spends hours crunching data. She takes downloads data from our sales, puts it in an Excel spreadsheet, does a few pivot tables, does V lookups to look at how she can crunch the numbers to get a forecast of how much we need to ship of every single one of our products. But pretty soon, through Microsoft, she will be able to ask Excel, can you produce a forecast model for, let's say, the next twelve months of each of our SKUs and how much I need to ship of each product based on demand and what we've sold for that product thirty six months to the present. So all this can be very, very effective for us when we're doing these sort of administrative and quantitative heavily modeling tasks. It has other applications, as you can imagine, if there's some winemakers in this room, think about forecasting of bottling quantities, shipment quantities, looking at how workflows can be better managed in terms of when you rack, when you bottle, how much to prepare for shipments, etc. Another area that is rapidly changing now with AI is customer service. So I think all of us I've probably used a website with chat bots that run customer service for us. I use this, this closed shopping, secondhand closed shopping website called Vastair collective, and they've got chat bots So if I need to return something, I can just ask the chat bot how do I return it? If I have an issue with the products that I buy, here's how I need to kind of go through the processes to get that fixed. So chat bots are an extremely easy way to harness AI because is you can feed it prompts for what you think your customers might have questions about, and you can feed it the answers to those prompts, and chatbots can very quickly help customers in a very kind of natural way as if a real human is talking to them and helping resolve their issues. Next, communication, and here's where it gets particularly exciting. So, AI is fantastic at translation. Andrea mentioned earlier about using chat GPT to translate a speech from Italian to English. And just out of curiosity, I wonder how many people have used chat GPT k. Good. The majority. And have you used it for the translation? Yeah. Exactly. It's pretty scary how good it can be, and it's constantly learning. And so there are a lot of different startups now trying to take natural language, the tone, the grammar, the specific vocabulary that, we all use in our our native tongues and feed them through AI models so that translation can be even more accurate hear it, going forward. And what's also very scary is that there's a there's a new AI startup called humane AI, and they've developed a little device that's about this big. And you can pin it to your clothes. And basically, you can ask it to translate what you're saying in real time, in your own voice. So it learns your voice cadence, it feeds it through an algorithm, and then it translates it through audio directly in real time to whomever you're speaking to. And this is really exciting for us because we all know in the international wine industry that language can be a huge barrier You know, we settle for less than fluency. We lose a lot of the understanding and nuance sometimes when we, when we speak to each other because we don't have that capability. And and I think this is a huge opportunity for us in the wine industry. We really, really need to engage in how we can use these translation tools effectively. And, the other thing I'll mention is creative design. So I talked about how, large language models use text and words to generate responses, but I think we all know now that AI is particularly adept at creating imagery as well as text. So I used Dali. I don't know if anyone's used Dali. But it's Chad GBT's image generator. And I asked, Dali to generate an advertisement for a wine tasting event in the style of Pablo Picasso. And this is what it created for me within seconds. So, for pretty good, wine tasting AdWords at this style of late era Pablo Picasso. And if I were to do this myself, firstly, I'd have to be a lot more creative than I am now. And I'd have to know different design tools, like Adobe indesign, Photoshop, etcetera, to kind of create this. But I I did this myself in seconds. I think what's also interesting if you notice the detail of it is that it's misspelled wine tasting in all of these, So you can kind of see the fact that although AI is extremely powerful, it's not perfect. It's not spelled wine tasting correctly. But, you know, pretty good. And, if you wanted to take a step further, if you did have a little bit of design skills and Adobe subscriptions, you can you there's a very similar thing in Adobe, they call it Adobe Firefly, which you can use AI to generate imagery, and then edit it very easily. So I could go in and and change the spelling of of those words if I needed to change the colors, etcetera. So, I mean, I you know, we spend money on graphic designers. At Liberty, we had two full time graphic designers producing, wine labels, priceless covers, event adverts. And I'm sorry to say that, their job might be in question. I don't know. Don't quote me on that. But it's pretty powerful what you can do with AI. So, what now are the risks and challenges with AI? So as this quote says, you know, the real danger is not that computers will begin to think like men, but that men will begin to think like computers. And I think the thing is with AI, it feels like magic when it works well. But like magic, if you don't know how it works, you can't tell what is real and what is false. So here's a picture of two Asian American female wine professionals. Which one is real and which one is not? Well, I can say the one on the left is me with different glasses and you I've got a new prescription. And that was before I had kids. So I was looking a bit perkier. But the one on the right is AI generated. And it's pretty scary because again, I use Dally to create her, and I just put in a prompt, which is picture of Asian American female wine professional, and that's what Dali created. I think there are two really scary things about this. The first scary thing is that she looks very real. And then the second scary thing is that she is a compilation of pictures of real humans in the internet maybe even including me. I don't know. Maybe there's some of her, me and her, which is also kind of scary. And this leads us to questions about the limits of what we should have AI do and whether our judgments and using AI should be applied at all costs. So, again, now let me talk about the risks in more of a summary way. So the first risk is misinformation. So because AI and large language models, are created by content that it is fed. This content is created by humans, and it can be rife with mistruth, misinformation, lies, and frankly quite evil content. So we're only as good as the content that is fed into it. The other kind of parallel to this is what we call hallucination, and that's when an AI, let's say, chat GPT, for example, is convinced it's given you a correct answer, even though that answer is definitely false. But it believes that it has given you the true true answer, because it has scanned all the data and spat out what it believes is statistically the correct answer for you. So this kind of comes back to that question of how much do we trust AI to give us the correct answer. The other issue that goes alongside of this is data privacy. As I mentioned before, if my image can be manipulated into other images, who owns that other image? Who owns the other woman's image? Where does that live? How is that being used? There's a lot of questions about data privacy and what we're putting in there and what these AI models are feeding off of as well. Third, biases, as I mentioned, The fact that AI is feeding off existing content means that it can be prone to certain biases. It can be prone to discriminatory content, reflecting kind of content that it's combed off of, the internet as well. So we must be aware of inadvertent biases that AI is creating for us. Intellectual property protection, and here's a huge area of concern for people is if you're asking AI to generate all this new content, let's say I wanted to create a children's nursery rhyme, twinkle twinkle little star in the style of William Shakespeare. What AI produces lives on, but who owns that? You know? Who who does it belong to? And lastly, although not comprehensively, we have the real question of if we use AI and we use AI more, What is the purpose of us human beings? What jobs are going to be at risk? And how do we measure education and human knowledge? Is that still going to be important? And, you know, this is I'm gonna show you, talk about a new example about this, which is quite scary as well, but, there was a paper written in March twenty twenty three and basically these researchers fed chat GPT for a lot of exam questions, and they fed these questions from the US, legal bar association and, you know, entry exam, They fed it graduate school admissions, exam questions, etcetera. And what it they did was that they fed it through chat, GPT, all these questions, chat GPT gave them answers, and they marked these answers as if chat GPT took the exam like a real student. Interestingly, they also fed chat GPT questions from the Court of Master Sommelier exam. So how did Chat GPT do? So here are the results. So chat GPT four ranked in percentile in wine exams. So the introductory, Sommelier's theory exam, it ranked ninety two in the ninety second percentile. So it was better than ninety one, students ranked from zero to one hundred on the exam. For the certified sommelier theory exam, it ranks in the eighty sixth percentile, and the advanced, sommelier theory exam, it ranked seventy seventh percentile. So essentially, Chat GPT was better than three quarters of the students that took the advanced sommelier theory exam. So, Andre, I'm not sure why we did the MW, to be honest. Just kidding. Just kidding. But, you know, so it really kind of kicks up a question of what are exams for, what are certifications and knowledge for, and we really have to radically change what we expect in the wine industry in terms of how we judge and grade this sort of knowledge. Yes. Good question. I don't know if everyone heard, but she asked Chad GPT to not evaluate or taste the wine. So what do you think about that, and will it be able to do that? This is interesting because I did ask a friend of mine, his name is Gus Zoo. He works at, Harv eighty one. Essentially a a chemical analysis company based in Napa Valley servicing the wine industry. And I asked him whether he thought AI will take over in terms of tasting. Basically, if you fed AI, the chemical composition of wine, will it be able to predict what the flavors and sensory experience would be like of that wine potentially so that you can, you know, back engineer a wine, chemically, that might be rated highly in a score, for example. And he told me this, which is that maybe it could do that, but wine as we know is extremely complicated. There's so many variables. In fact, one chemical compound, even if they isolated that, can be perceived very differently on the nose and then on the palate, depending on the entire matrices of other chemical compounds of that wine. It also, of course, depends on the person tasting it. You know, what is their, background, what is their history of sensory perception? Are they sensitive to things like Oak or reduction, etcetera? It's extremely difficult to actually feed AI, these sorts of sensory analysis, maybe in the future, but it would involve ex extremely large data sets to do so. So we'll see. But not now, I think, would be the answer that Gus would say. AI regulation. So these these are a few different, AI sort of regulatory frameworks in place to kind of safeguard the risks. Here, it's important to say that although governments are acting, governments are slow to respond to this. And in fact, AI is being developed by private companies around the world. So really, you know, how are we going to coordinate the policing of AI? You know, it's always going to be two steps back from what, private companies and AI startups are doing now. There's a lot to be aware of. Okay. So I'll wrap up by saying, you know, what is the call to action that I would like us in the wine industry to take on now? Firstly, it's to take the time to learn and experiment with AI. Literally, you know, use chat GPT think about your daily lives as well and think about if there's other if there's things that I'm doing that's repetitive that can be automated, just see if there might be an AI tool in development that might help you with this. Be it graphic design, marketing materials, translation, quantitative analysis, etcetera. Secondly, it's to share knowledge amongst each other. I think we, in the wine industry, are very collaborative. So we need to work together and share what we've learned and help each other out. And lastly, and most importantly, we need as an industry to collectively uphold truth and authenticity. Authenticity in our industry is essential. So we must make sure that we're promoting that, in the content that's being generated and always question whether content you see is authentic or not. And I think, ironically, the advent and increase in the use of AI might make things like person to person interactions, meetings, tastings, trips, visits, etcetera, so much more important. Here's a few AI tools that I would suggest that you can go back and try immediately. And, you know, I think the first four are probably the most useful and applicable to us today. So I do recommend that you go out and try these websites and see what's going on. I think particularly Microsoft co pilot, so basically they're developing an AI assistant to help you with the usual Microsoft products that we all know and love, excel, PowerPoint, Word, Outlook, Outlook, etcetera. I'll end by saying that while we use these AI tools and start engaging in AI, we must never forget what our ultimate goal should be. And I believe that it it is to embrace and help grow that sense of beautiful messy complexity that wine is and that what we have within our human souls. So I'll end this presentation, by sharing one of, my favorite poems and a few lines from that by the great humanist American poet, Walt Whitman. And he's and he, when talking about what is special about humanity and about the lives we lead, talks about the immense complexity that we have within us, which is also reflected, I think, in wine. And he says, do I contradict myself very well then? I contradict myself. I am large. I contain multitudes. Thank you very much. Do we have any questions? Thank you very much, Rajine. That was fascinating and terrifying in equal parts, I think. It's it's a comment really, not a question. I thought the the the the the Dow E pick casso, wine tasting posters. Did you say that you were surprised that, that the spelling that it was misspelled? Because, what what I found most most interesting was because the misspelling of the wine tasting was deliberate, wasn't it? You know, it's an interesting one because I think what it's done is that it's mistakenly popped in what I prompted it to say, which is wine tasting. But scrambled the words and sort of imagery that it expected I be to want. But it wasn't it wasn't it doing wasn't it spelling wine tasting in the style of Picasso? It was definitely, but then it was also adding like random letters within within the tasting as well. But but that showed a sense of humor and it's I I find that I find that absolutely terrifying because it would have taken me, you know, you could have done that that that design in the style of Picasso, couldn't you? But you never probably would have thought of actually jumbling up the letters Yeah. That's a good point. In the style of Picasso. That's a good point. Yeah, me. Sure. So anyway, thanks very much. That's great. No. You're welcome. You're welcome. Any other questions or comments or any examples of someone using AI in an interesting way? Oh, question and comment? Just a quick follow-up to the last question, because I was also thrilled with the image, and the picasso image, and totally agree on the fun, misspelling of the words. But did you then prompt it and correct it? Because that's what I find when I've experimented with image making, how great that you can then. So did you next did you have a next step where you said asked it to do it again and and spell it correctly? Just curious about, what you did next and what what you saw. Yeah. So I did prompt it to do something else as well. And I was actually I would say in real life, what I would do is download that image because you can easily download the images. And then pop it into Adobe Firefly and then sort of start editing it. But then, you know, I think the interesting thing about chat, GBT, Dolly, and all these AI tools is that I went down very quickly a rabbit hole of doing other stuff. So I was like, oh, Picasso. Oh, can I try this with Matisse? I wonder what, like, Bauhaus design would look like in a wide tasting advert. So, like, I think my attention span quickly moved on. But, yes, like, I think the the idea then in a, in a sort of normal way is to, to try to correct it and try to get it. Well, actually you spelled it incorrectly. I didn't do it in that instance. But I think, you know, if I were to use it effectively for a wine tasting advert, I would ask it to correct it. And and probably use Adobe Firefly to edit it. I have a question. It's a question as a manager. Today, many times in our work when you're talking about wine, we are referring to a very specific sense of place. And many times, the sense of place is the sense of intuition. So in the day where everybody, it will became very annihilistic. So we will be very objective. How in a group of people can we support or we can express our sense of intuition? That's a very good question. How do we promote our sense of intuition? I think this goes back to questioning what is the purpose of our human selves? That was a very existential question. But, like, going back to, you know, would we even want AI to start analyzing wines and trying to predict what a wine would smell like based on chemical composition? That's not really the point of wine, you know. And I think the more we use AI to automate certain parts of our lives, I feel personally that we're going to love and cherish the things that make us human like the intuition like talking about wine in a very personal sort of intuitive way. So I think it's more of a philosophical question, Andrea. What do what do you think? Oh, yeah. Absolutely. But, you know, we if we discuss about specific subject that, are not related to testing, but, for example, are related to cells that it became or which direction we should to take for a brand in terms of positioning and other things like that. I think that this sense of, being very objective, it will improve our capability of having a sense of intuition that it will be much more precise than what we will get today. So this is my my point of view. The problem is probably in between relationship between the people and the members of the team that they will be probably considerate in a different way than before. So one of the issue on the major concern that I I think that we will have to take is how we can, with with instruments like this, how we can take we can keep people in charge or, alive, in the position that they should have. Yes. And let me be clear. A lot of AI startups are not talking about AI takeover per se. But, right, rather AI assistance. So AI will help you do most of the job, but a human still has to make the judgments in the final call. So even with demand planning and forecasting, AI can generate a demand planning model about how much wine I need to ship. But still a human needs to kind of go in and correct it and and judge whether that's correct. Right. That's it. I think. Thank you so much for coming. Thank you for coming. Thank you for your attention. Thank you, Regina. Lovely speech. Listen to the Italian wine podcast wherever you get your podcasts. We're on SoundCloud, Apple Podcasts, Spotify, Himalaya FM, and more. Don't forget to subscribe and rate the show. If you enjoy listening, please consider donating through Italian wine podcast dot com. Any amount helps cover equipment, production, and publication costs. Until next time, Cheaching.