About briefing. Q&A with Silvio Carta, Head of Design, University of Hertfordshire

In the past decade, digital technologies such as CAD and BIM have had a huge impact on the work of architects and engineers. Now, generative design is emerging as the next advance. Computational design applications, powered by artificial intelligence (AI) and machine learning (ML), seem capable of independently producing, optimising and testing architectural designs, based on parameters coming from e.g. regulations, zoning plans and the client's brief.?For me, the question is: how will such technologies impact the practice of briefing? Do we need to change the way we write briefs? Will we at all need briefs? To find some answers, I talked to computational design expert Silvio Carta . Silvio is Head of Design at the 英国赫特福德大学 and he has just published an authoritative book on this topic called 'Machine Learning and the City: Applications in Architecture and Urban Design' (Wiley, 2022). ?

JVM: You have worked extensively with computational design. What is the role of the client’s brief in such processes?

SC: When approaching design challenges through computational methods, one of the first phases is the translation of the client’s requirements into elements that can be clearly identified, quantified and eventually computed. Quite often, initial briefs include general information that describes the intended outcomes through qualitative characteristics (for example, “the new building aims at becoming a reference point for the users and passers-by…”). In such cases, computational designers may need to reinterpret the initial brief by posing new questions or presenting the requirements through a different lens. In our example, we need to clearly identify what ‘becoming a reference point’ may mean and how it can be measured and assessed. One approach involves for example the introduction of metrics that help designers (and clients) to appreciate and measure the success of the project on that particular criterion. Our ‘becoming reference point’ can be measured, for instance, in terms of changing the perception of people who pass by the building. If we find a reliable method of canvassing people’s opinion before and after the introduction of the new building, that gives us a robust way to measure the success of the project, and therefore quantify how well the design solution responded to the client’s brief. On the other hand, those briefs that contain clearly articulated requests that are measurable (for example, building performance, CO2 emissions, energy consumption, etc.) are—generally speaking—easier to deal with via computational processes. ?????

JVM: What makes a good brief, from a data perspective??

SC: In order to be addressed through computational methods, briefs need to be described in discrete and measurable entities. If briefs contain data as the basis for their requirements, this makes the job of the computational designer easier. A good design brief would include clear and measurable outcomes and ambitions (e.g. improve a particular aspect by a defined percentage). Clear, quantifiable targets help designers to better understand the expectations and requirements of the client, but also to find the appropriate method to meet those targets.

"Clear, quantifiable targets help designers to better understand the expectations and requirements of the client, but also to find the appropriate method to meet those targets."

JvM: Can an algorithm create a design solution on the basis of a brief—taking over activities that would normally be done by an architect?

SC: Let’s suppose the client has produced a very detailed design brief with measurable outcomes and assessment metrics (exact room sizes, access and control measurement, light and temperature requirements, ranges of intended walking distances between rooms, etc.). In this case a well-designed generative approach could theoretically be used to translate the clients’ requirements into objectives and fitness functions for a ‘solver’ (an algorithm that generates and evaluates design solutions) to accurately respond to the brief. If it is true that such an algorithm could automate the development of design solutions, it is also worth noting that we still need a designer to develop and calibrate the algorithm that works with that particular brief. In other words, a bespoke algorithm can, in principle, produce a design solution that perfectly fits the design brief. We already have numerous examples of such approaches. One of the most common is perhaps Galapagos, the solver developed by David Rutten for Grasshopper (now fully integrated into Robert McNeel & Associates’ Rhinoceros), where the designer needs to define a number of variables that need to be optimized against a fitness function (that tells us how close to the optimal solution the variables are in an iterative process). In this case the designer is effectively completing and refining the design brief with clear instructions and measurable factors. ???

A fully autonomous and independent algorithm that can simply take the design brief as an input and return a fully-fledged design solution as an outcome is far from being reality. To achieve this, we need to wait until an Artificial General Intelligence (AGI) is developed. There is a lot of speculation on this topic, but it is perhaps useful to consider the prediction by Google’s futurist Ray Kurzweil that the singularity (where a truly self-learning intelligence surpasses human intelligence) would occur by 2045. However, we are currently quite far from such an advanced level of artificial intelligence. For now, designers still play a crucial role in interpreting the client’s brief and translating it into something that a computer can understand and compute. ???????????

"A fully autonomous and independent algorithm that can simply take the design brief as an input and return a fully-fledged design solution as an outcome is far from being reality."

JVM: How do you see the future of briefing in relation to advances in machine learning and artificial intelligence? Will we, for example, still need briefs?

SC: The short answer is yes, we will need clients’ briefs in the future. Moreover, these briefs will need to be increasingly precise in defining the requirements for the project. Analytical and design methods based on ML and AI are—generally speaking—very sophisticated and powerful. As such tools tend to be characterized by a high degree of precision, it is important that clients are ready to describe their requests in accurate terms to make the most of these advanced techniques. I suppose a good analogy would be that of the type of light that can be used to indicate a direction. A traditional brief could be represented with a torch (or even a lantern), where a client casts light on a dark pathway to show designers the direction they want to take with their project. The torch would provide a wide-angle light that provides a good general overview of the direction while illuminating only a few feet of the pathway. Designers can see the general direction of travel, but they need to physically walk into the pathway in order to see what is next. On the other hand, a brief designed with AI/ML tools in mind, would be like a laser, with a very narrow-focus, yet deep, light. This laser illuminates a very specific direction in the pathway showing designers the full extent of the clients’ intention and ambitions with that project. Unlike the torch, the laser won’t show the context (i.e. the wide-angle view), so the brief needs to be as accurate as possible for designers to be able to fully satisfy the given requirements. ???

"A traditional brief could be represented with a torch (or even a lantern), where a client casts light on a dark pathway to show designers the direction they want to take with their project. On the other hand, a brief designed with AI/ML tools in mind, would be like a laser, with a very narrow-focus, yet deep, light."???????????????

JVM: Any recommendations for construction clients?

SC: Yes, I would recommend having a clear overview of what tools and services are already available on the market to help investors, developers and building companies to automate some of the key processes. Examples include Delve by Sidewalk Labs (Part of Google) , Finch , Labit Group ’s Turing, Spacemaker AI , and TestFit among many others. All these companies are developing very interesting tools to help building companies and developers with the first stages of design, more specifically with space planning, feasibility studies and initial studies. Such tools can help clients to both better define their own requirements and aspirations for their projects and be in a more informed position when they commence initial conversations with the design teams. ?

Many thanks to Silvio for sharing his expertise on this exciting and challenging topic. For those who want to know more about computational design: Silvio’s latest book can be found here.