Andy Thomas-Emans: AI in the labels world – threat or opportunity?

Anyone who has engaged with ChatGPT or any of the other Large Language Models (LLM) out there cannot have failed to be impressed with their capabilities to turn natural language inquiries into intelligent and actionable responses.

Of course, the accuracy and usefulness of the ChatGTP response are directly related to the information it has been trained on. And this is why the idea of incorporating your own custom knowledge base into ChatGTP is an interesting one. To be able to train an LLM with your company’s own data would allow an almost limitless set of natural language queries to be asked without the need for programming knowledge or knowledge of specific software. Let’s call it LabelChatGPT. The possibilities are endless.

For example, an in-house design team at a converter could generate multiple design variations for a customer’s new packaging campaign within the capabilities of the equipment available, and as the equipment list expands new treatments can be added.

If the LLM ‘understands’ the goals of the customer’s marketing strategy, understands the competitive landscape and is given a set of design parameters to work with, it could come up with a series of variants, learning what it needs to change as it gets closer to the ideal label design.

What about modeling different job scheduling scenarios? ‘LabelChatGPT, here are my five most important clients who need to have their jobs finished by Thursday. Give me five different scenarios which make the most efficient use of pre-press, press and finishing resources while allowing me to fit in three short notice short run jobs involving variable text and foil finishing.’

Or equipment maintenance schedules. ‘LabelChatGPT, write me a document containing the sequence of job maintenance for every piece of equipment in the plant, taking account of peak workload, holiday periods of key staff and availability of external and internal engineering resources.’

Warehousing and stock levels? ‘LabelChatGPT calculate when I need to reorder my top 10 label stocks given the jobs scheduled for the next month and average makeready and wastage rates per press?’

Of course, many of these queries can be answered at the moment by interrogating a management information system (MIS). The difference is that the MIS will require training in a specific software package and the ability to request and integrate information from different silos – warehouse management data, material consumption data, equipment configuration data, press running data and so on.

The ability to use natural language requests and for the LabelChatGPT to accurately parse that to actionable scenarios which can be endlessly refined, queried and subject to numerous ‘what if’ questioning - in real time - could be a game changer in the way we interact with information and with the physical plant around us. This is particularly the case as we start to gather more and more data automatically from sensors around the plant and at key nodes of our operations: from machines, from environmental sensors, from materials and consumable workflows through the plant. How will we make sense of this data and not be overwhelmed by it? How will we make it useful and useable?

Dedicated AI tools like ‘LabelChatGPT’ would be one possible answer. This is all hypothetical of course but well within the bounds of current technology.

The most important initial task would be to organize the information which the LabelChatGPT will use and to keep it current. This would give converters with well-organized data sets a key advantage, and anyone with a comprehensive MIS probably has the data backbone required.

Indeed, going forward it could well be the MIS suppliers who best know the label industry who will lead the way in integrating LabelChatGPT into their MIS programs. It would be a natural language front end to an intelligent assistant which uses the information dynamically updated in the MIS as a basis for its decisions and recommendations.

Custom data sets are already being integrated in LLMs by a wide range of companies, from financial institutions to car manufacturers.

Thoughtful commentators on LLMs are stressing that rather than being a threat to human creativity, they can act as force multipliers. Humans will make higher-level decisions using intelligent assistants to analyze and integrate complex data sets and deliver clear and actionable recommendations.

To read more of Andy Thomas-Emans’ columns, visit www.labelsandlabeling.com