In a thought-provoking session entitled The promise and the perils of Generative AI in wine at last year’s wine2wine Vinitaly Business Forum, wine journalist Felicity Carter explained how Generative AI is destroying company productivity.
Felicity Carter
Generative AI is a type of artificial intelligence that creates content (including text, images, code, audio, and video) by learning patterns from massive datasets. Generative models such as ChatGPT and DALL-E generate content based on user prompts.
While Felicity acknowledges that such tools can be useful for performing certain tasks, Generative AI, she argues, can be destructive in the context of wine-related content. There are a number of reasons for this, which are considered in more detail below, as well as some of the areas where Felicity suggests AI can actually be useful.
In her presentation, Felicity refers to a recent study published in the Harvard Business Review (you can view a video summary here) which found that while Generative AI promises efficiency, its current implementation is causing significant, unexpected, and often “invisible” losses in productivity. Rather than reducing workload, AI is frequently intensifying it, leading to burnout and, in some cases, lower-quality output. In fact, Harvard researchers have identified a phenomenon termed “workslop”—AI-generated content that appears polished but lacks the substance, accuracy or context required to advance projects. The Harvard research estimates that a 10,000-person organization can incur an estimated $9 million in lost annual productivity due to this “workslop tax”.
As a researcher, AI technology is making Felicity’s life miserable. Generative AI is so simple to use that people are just churning out content and leaving colleagues to deal with it. As a consequence, we are spending more time reviewing, correcting, and managing AI-generated output. AI has, Felicity explains, doubled or tripled her workload because people are handing her material they think is well written but is actually never quite right. Furthermore, using AI is negatively impacting how colleagues perceive each other, with someone using AI to generate content being perceived as less competent or creative in the workplace.
Another problem that Felicity identifies is the banality of AI generated content. With the increased use of ChatGPT to generate everything from LinkedIn posts to emails, everyone sounds the same. Worse, people are even responding to AI slop with more AI slop of their own!
It is, according to Felicity, impossible to hide AI-generated content. We instinctively recognise it, and that’s because of the way it is generated.
LLM (Large Language Model) is a type of AI designed to understand, generate, and process human language by analyzing massive datasets. But, as Felicity explains, LLM generally only has access to information that it finds on the internet. According to AI itself, it prioritizes plausibility over veracity and “can produce highly convincing, grammatically perfect text that is factually incorrect, biased, or nonsensical”. LLM-generated content often provides the illusion of knowledge without genuine understanding. Furthermore, in an effort to please, LLMs are often trained to prioritize agreement with the user over accuracy, confirming false beliefs to appear more helpful. Unlike in a real boardroom or office, there is no pushback on bad ideas.
LLMs are now programmed to be risk averse and appeal to the widest audience, making output bland and banal. There is no tension or grit, so generative AI can result in you telling the blandest version of your story. This is a particular problem for the wine sector, as wineries are all fundamentally telling the same story. One effective way to get around the shortcomings of AI is to treat it like your enemy, to ask it to be critical.
Despite its major shortcomings, AI is, Felicity observes, excellent for some things, such as automation, resolving tightly defined problems, offering consumer insights, sorting and analysing products and portfolios, and scenario simulation. Just don’t ask it to generate the content for your winery. Some things still require a human touch!
Listen to the Full Session:
