Lego plans to expand its AI toolbox, with human oversight and no full automation
Lego plans to add artificial intelligence to its corporate toolbox, but cautiously and always with humans in the loop. Morten Lundsgaard Degn, Vice President for Digital Technology at Lego, said so at Ingeniøren's Digital Tech Summit in Copenhagen. He described AI as a permanent part of the company's journey, as Lego grows quickly and now has 34,000 employees in 40 countries, with almost everything driven by digital technology.
Degn said AI is a tool and will not solve problems for people. Lego is considering how to add it to its toolbox, where it does not exist today, and will apply it case by case where it makes the most sense. The company looks at frameworks, data and integration, because scaling AI would otherwise become very expensive. He noted that many employees use AI at home and can be disappointed by what is allowed at work.
Top management has been clear that AI is not released without human oversight, and Lego does not fully automate things. Degn tied this to the brand and to children, saying people have seen terrible examples of what AI can be used for and that this will not happen under the Lego brand. Lego is careful but experiments a lot in order to learn. Security and compliance are central considerations.
Lego is also upgrading its factories, which, because of growth and past design criteria, are not necessarily built for automation, digitalisation and AI. Management has approved investment of many millions across six, soon seven, large factories to build a foundation for scaling AI-driven automation.
Two years ago, CEO Niels B. Christiansen told Avisen Danmark that Lego used AI for sales forecasting, product recommendations and a customer-service chatbot. He said ChatGPT was the only general tool allowed, under rules that forbid sharing sensitive Lego material, to protect factory secrets.
A reader commenting on the piece questioned whether Lego has considered the risk of sending business secrets to US services, or running an open-weight model itself.