We covered some of the core principles of how AI is trained and learns in our AI Primer for the Food Industry post. From there we saw that AI models, (the technology that powers chatbots like ChatGPT), were trained with an enormous amount of data, including essentially all the information that exists on the internet.
This means models “know” quite a lot of things. For example, take my favourite quote from Shakespeare:
“This above all - to thine own self be true,
And it must follow, as the night the day,
Thou canst not then be false to any man.”
Type this into any AI chat and the AI will know instantly that this is from Act 1, scene 3 of Shakespeare’s Hamlet. That’s because there is a lot of text from Shakespeare on the internet (Google gives back 47,000,000 results when you search “Hamlet”).
This is why AI can feel so impressive. Interacting with it gives us the feeling that it knows things, maybe even that it understands. But much of that is an illusion. AI doesn’t understand or knows things, it calculates. That’s why it’s easy to identify a specific Shakespeare quote, it’s just pattern matching something that it has a lot of information on.
But that’s not enough if you are running a company in the food space, or any space. Because knowing a lot of things is not the same as knowing the right things. And when it comes to a specific food distribution or manufacturing company, the right things to know about the business are not available on the internet, unless something has gone horribly wrong.
These are things like how one customer can describe a product they want and knowing that it corresponds to a specific product that the company offers, even though the two things don’t exactly match. Or, it’s knowing that a customer prefers products to be delivered in a specific way.
Let’s take a specific example we’ve seen at Burnt. Do you know what the difference between tortelloni or tortellini pasta is (for my friends who are slightly dyslexic like me, one ends with “oni” and the other with “ini”)? Not many people do, including, it turns out, some of the people ordering it for their restaurant, excluding Italians of course, who I imagine will get offended at the very idea that these can be confused.
Knowing the right thing in this case is to understand that a specific client actually means tortelloni when they write tortellini in their order. This is knowledge that gets built up over years of relationships with customers. No matter how hard you try, you won’t be able to share this level of understanding to ChatGPT or Copilot, unless you provide it with this information each time, which then mostly defeats the purpose of using AI for this process in the first place.
Here we touch on why general AI like ChatGPT is limited to being a useful support tool, and not an integral part of business processes driving real efficiency and accuracy improvements. Unless AI knows the things that sit inside the heads of people in the team who’ve worked in the company for years, combined with the information sitting across all the different systems that those team members use to inform their daily work, then AI will never be an effective solution to meaningfully improve how businesses operate.
And this is where we can draw an important distinction between generalised AI and specialized AI like Burnt’s agents for the food industry.
Once you get AI to understand tribal knowledge from the team, and give it access to data across all the relevant company systems, it becomes an incredibly powerful tool. So powerful that it can become an integral part of a food company’s daily operations, significantly outperforming human team members in terms of accuracy and output on specific tasks.
This is what good specialized AI solutions do. They bring together the ability to remember past actions, for example how each customer orders, or that a restaurant has a demand spike on Thursdays when there’s a baseball game going on. It then learns, using the memory to adapt how the system acts in the future, building up context of the business, customers and suppliers to continuously improve and adapt over time. The outcome is a solution that can still quote from Hamlet if asked, but can do so much more by performing actual work that uses the right context to perform the correct action each time.
We repeat and will continue to repeat that we don’t believe the value of AI is to replace humans, especially in a relationships driven industry like the food sector. Instead, its true power is to enhance the work that humans do by replacing the manual and repetitive work that is needed to keep the business running, something that AI with the right information and structure can do very well, so that humans can focus on the high value work that grows the company.
In this context, if you asked a customer success representative or purchasing manager whether they’d prefer to have an AI solution that knows Shakespeare, or one that understands each customer, each supplier, their preferences, habits and full history with the company, we already know how they will respond.