The lion, the steam train and the LLM

Last week, I was fortunate to go on a journey on a steam train, from one side of England to the other and back again. All along the route, people smiled and waved and took photos as the train chugged past: it was a special sight.

From a purely technical point of view, this happy reaction may seem to make no sense. Steam trains are slower, less reliable and more polluting than their electric alternatives. Yet a steam train has charisma: it has a presence and an energy which an electric train, no matter how sleek or modern, can’t match.

There is an analogue in the world of nature: some animals are deignated charismatic megafauna: lions, tigers, elephants, rhinos and so on. You can probably guess the top ten animals in this category: they are the animals that appear in film and stories, in advertising and environmental campaigns. They have a status as symbols: their images mean courage, ferocity, strength and other abstract concepts which may have little to do with their actual attributes and behaviours.

In the world of physical and mechanical engineering, our equivalents of charismatic megafauna are things like steam engines, bridges and skyscrapers. We have charismatic megafauna in the world of enterprise technology too: the mobile app, the ERP and CRM systems, and the multi-year deal with the cloud provider. 2023 has been dominated by the arrival of a new big beast, in the form of generative AI. These are the things which draw attention, investment and resources.

But do they draw too much attention? There is a danger that the charismatic megafauna in any field skew our focus. If we try to manage the environment in a way which is best only for lions and elephants, we may disadvantage many other species. If we optimised our railways for the romance of steam, then we wouldn’t get anywhere - which is why a trip on a steam train is now a rare treat. And if we do nothing but pursue the biggest beasts of enterprise technology, then we risk neglecting the less glamourous fundamentals.

I believe that we can draw a lesson from environmental campaigners. Such campaigners understand that charismatic megafauna cannot survive in isolation: they live as part of an ecosystem, and the whole ecosystem needs preservation if the biggest animals are to survive. Indeed, the biggest animals, often at the top of the food chain, are the most dependent on the complete ecosystem. Campaigners are willing to use the innate appeal of the biggest and most charismatic animals to make the case for the whole ecosystem.

It is the same with physical engineering and technology. My ride on the steam train was dependent on special arrangements to take on fuel and water: the ecosystem that supports steam locomotion does not exist any more. Similarly, my mobile app is dependent on back end systems, the data centre that hosts them, the network that connects them, and invisible services such as DNS. Users of the mobile app my be unware of these essential components, but we can use its charismatic nature as an entry point to explain this ecosystem to sponsors.

This is particularly the case with generative AI. This year has been an exciting period of experimentation and exploration. However, at times, it may have felt distracting: that sponsors and stakeholders were chasing generative AI solutions while neglecting the fundamentals. I do not think that we have to be distracted, though. As technologists, part of our job is to explain how technology works, and to make the case for doing the right thing. If we explain properly how generative AI works, then we also have to explain how it is dependent on data, on deployment pipelines, on integration with other services, on security, on monitoring and management - and so on and so on.

The enterprise technology landscape is dominated by charismatic megafauna: not all of them survive or persist, but more will always arrive. They will generate attention and draw excitement: our job is to connect that attention and excitement to the whole ecosystem, and build a technology environment that works for everybody.

(Views in this article are my own.)