By PAUL VON BÜNAU
The concept of Artificial Intelligence had a massive impact over the past several years. Marvelous tales about its prospective achievements abound, just as much as sceptical questions about whether (or when) the machines will finally substitute the humans on this planet. As a result of this heated discourse, leaders across many industries feel an urgent need to somehow incorporate AI technologies into their workflows, fearing to miss out on an important innovation step.
Great FOMO, even greater expectations
The reality is different. Most Artificial Intelligence initiatives fail. Not because of technical challenges, but because of flaws on a conceptual level. The insights derived from data analysis are meaningless or tell us things we already know. Proofs-of-concepts never make it to production. AI experts cannot explain their results or over-optimize irrelevant criteria. And these are just a few of the common patterns of failure. But why is that?
Today, most of our AI-experts come from academic disciplines such as computer science, mathematics, physics and engineering. There, students typically learn to attack limited, well-defined problems using quantitative and computational tools – and thus come to the field with a specialist’s mindset firmly established in their brains. Budding AI-experts are used to working within an abstract domain which admits only isolated, clear-cut tasks. Only the world isn’t like that.
The world is messy, ambiguous, volatile and largely unknown. And even though many mathematicians, scientists and physicists master AI methodology on a technical level, they lack the conceptual understanding that is crucial to find creative solutions to actual problems that are VUCA and often wicked.
Enter the architect
This is, essentially, what architecture as a meta-discipline is about: sorting out the complexities involved in any technical project. The notion of the architect is prevalent in various contexts, be it technical construction, information technology or business strategy. Architects are recognized for their creative and technical abilities, but their role and skill set is much wider. When planning a project, they will look into multiple social, political and environmental factors. They will try to understand how their project will support people in their daily tasks. They will look into which construction materials should be used. How much money the project could cost. How the project integrates into larger structures. Which environmental and social factors must be considered. And so on.
This holistic perspective has been common sense in systems engineering for a while. When software engineering matured in the 1990s, it became apparent that it didn’t suffice to just write the code. Engineers needed to take into account multiple external factors. This insight prompted Hasso Plattner of SAP to found an institute bearing his name in Potsdam. Twenty years later, the notion of a systems architect is now well established in IT projects around the world.
In Artificial Intelligence, we are only beginning to understand this necessity. We, too, need to take a holistic perspective on our projects. This is even more imperative in AI than it ever was in systems engineering, as AI is even more interconnected and complex. Whereas the majority of software systems tend to process data in a series of merely logistic tasks (such as storing, retrieving, filtering, aggregating, displaying data), Artificial Intelligence systems aim at something more ambitious: automating cognitive work. This is much more difficult to specify, as the automation of cognitive tasks is almost always directly linked to other matters, whereas simple “information logistics” mostly isn’t.
The making of an AI expert
In order to deal with these challenges, we need AI architects. We need people who can take a multifaceted perspective and implement AI projects in a meaningful way. Our projects must be useful for the organisations we work with. We need to build trust in our models. We must convince clients that our analytics will not just deliver flimsy predictions, but will actually help transform the way they go about their business. We must help them along the daunting journey of organizational change.
But where should these AI architects come from? Young graduates who come to us with a science degree are typically not prepared for the broad scope of our tasks. Should we teach them to adopt an architect’s holistic approach? Yes, that is one way to go. We need to extend our AI-related curriculum to put more emphasis on understanding maths and algorithms on a conceptual level.
Or should we take another avenue? Should we hire all-rounders from different disciplines and educate them to be AI architects? Philosophers, musicologist and art history graduates as the future AI architects? Emphatically: yes! It is not only possible but also fruitful to discuss AI without being able to actually code it. Therefore, we must find ways to teach brilliant minds from other backgrounds to become AI-architects and introduce their respective world views to our field.
Both routes are promising, and both routes must be pursued.The classical architect’s curriculum should be an inspiration for how to educate AI architects. Every architect must have had her stint at the concrete mixer; every architect must – at least once – have calculated the statics of a supporting structure, before calling him- or herself an architect. It is not sufficient to learn the basic tools: you must understand your craft in order to make meaningful use of it. Budding AI architects must also take sprints in real-business contexts. They must learn relevant side-topics such as business and project management.
Digitisation has evolved to a level where mathematics and algorithms are permeating all areas of life. We need the AI architect now more than ever, both for implementing successful projects and for persuading the broader public that AI is useful and valuable.