I recently attended the Chicago Humanities Festival at Illinois Tech to hear “Redefining Design in the Era of AI” with John Maeda, formerly with MIT Media Lab and now Vice President of Design and Artificial Intelligence at the Microsoft Corporation. Anijo Matthew, dean of the Institute of Design, facilitated the session.

The session’s premise was to explore AI’s transformative potential in design. It illustrated “how AI can directly address functional needs, potentially rendering traditional forms obsolete. Through this lens, these AI experts envision a future of design where AI not only shapes form but potentially eliminates it. This new paradigm, where function dictates existence without the constraints of physical form, is truly intriguing.”

John Maeda’s contributions to the field of design and technology have been positive, as he has been ahead of many trends. His five books, including the insightful ‘Laws of Simplicity’ (a personal favorite) and his recent ‘How To Speak Machine’, are testaments to his deep understanding and innovative thinking. His annual ‘Design in Tech Report’ is a valuable resource that identifies trends and predicts the future of technology deployment and use.

His career has been defined as exploring how creativity and design can interplay with STEM and bringing a human-centered perspective on shaping, deploying, and using technology. He is also a big proponent of “computational design,” which focuses on designers and other stakeholders intentionally designing computer algorithms, simulations, and data analysis for greater human agency. His perspective is that digital technology transforms humanity and needs a critical human eye. It is better if designers learn about computation and shape it rather than remaining consumers of software and platforms.

The Rise of AI Anthropomorphization and the Black Box Problem

His current Design in Tech report and what he used in the Chicago Humanities presentation is called “Design ag(ai)nst AI” – a playful way of collaborating and combating the mindless embrace of deploying AI. John stated that machine learning has focused on creating expert systems for decades due to the many limitations of computing power and software programming. From the 1990s to 2021, John called this “AI Nuclear Winter” period when AI was on the outer edges of utility and visibility.

AI has reemerged from its winter into a blazing summer. Like all technologies on the Garner Hype Cycle, AI’s rapid mainstreaming of AI chat platforms and AI image platforms and integration as microservices that power many other digital platforms exploded in 2023. Reflecting on this explosion, John believes we may be creating a false god unquestioningly accepting AI outputs without questioning these outputs using critical thinking skills. An analogy I thought about was the ancient oracles, like Pythia, the Oracle of Delphi, who was the interface between the human and god world and delivered ” . . . oracles in a frenzied state induced by vapours rising from a chasm in the rock, and that she spoke gibberish which priests interpreted as the enigmatic prophecies and turned them into poetic dactylic hexameters . . .” The main difference in this analogy is that current AI systems distill content in easy-to-understand and bite-sized ways.

Due to social media, and the dependence on mobile, people like to be entertained and spoon-fed answers with as little effort as possible. They are now starting to use AI chatbots because, unlike traditional search, they conveniently aggregate answers and copy and paste them into reports and other types of documentation. It is a combination of fun interacting with a chatbot and focused content in shorter periods of time, which delivers time-savings and efficiency.

Chat GPT 4.0 has just released its new version, “Sky,” which further anthropomorphizes AI by speaking content in a perky human voice. A popular cultural example of this is the movie “Her,” about a lonely writer who develops an unlikely relationship with an operating system designed to meet his every need. Scarlett Johansson is the voice of Her, and in a schadenfreude moment for Open AI, they approached Scarlett – twice – to be the voice of Sky to ” . . . bridge the gap between tech companies and creatives and help consumers to feel comfortable with the seismic shift concerning humans and A.I. He said he felt that my voice would be comforting to people.” Unfortunately, Open AI deployed a voice very similar to Scarlett’s and had to retract it when challenged by Scarlett that she did not give permission to use a likeness of her voice.

This move to emulate human behaviors will become more difficult to identify as we move from narrow artificial intelligence to artificial general intelligence. AI is anthropomorphized further with the equivalent of deep fake avatars delivering content as digital sherpas. We are seeing in Japan where digital avatars in mobile platforms such as Replika are embraced to combat loneliness, John recognizes that anthropomorphization is a delusional trick to convince humans the anthropomorphic interface is equal to a human and granted immediate agency.

John believes there needs to be “HIL” or humans in the loop to shape AI platform technologies and identify possible errors or problems. Unfortunately, the more significant problem looming with AI is it is a black-box technology that humans do not fully understand. The input of any AI system is the corpus of information that trains it and the many iterations it goes through to refine its answers. The outputs are the answers to specific prompts that benefit humans, but they also train and refine the system to create more refined answers for greater efficacy.

The problem is that the process of how an AI system makes decisions is hidden from view between the inputs and output. No recorder or log file diagrams show how the system rationalized an AI output. Saurabh Bagchi from Scientific American highlighted that ” . . . researchers don’t fully understand how machine-learning algorithms, particularly deep-learning algorithms, operate . . .” If we do not understand the decision-making process, how do we know it to be accurate, or if not, how do we fix aspects of decision-making to gain greater efficacy? Anthropic recently announced that it could forensically understand AI decision-making by identifying features through “mechanistic interpretability,” but it is at a nascent stage.

Three Design Epochs and An Uncertain Future

John has defined three design epochs: classical design, design thinking, and computational design to show how each one informs the other, but also that computational design will desiccate the other two over time:

• Classical Design, which included discreet specializations such as graphic and industrial design, focused on specific skills and craft to create “canonical design” analog objects of their discipline. To John, this type of design is holding design back.
• Design Thinking shifted design from focusing on objects to frameworks that defined the right problem to solve using human-centered methods. The double diamond model, informed by several disciplines co-creating together, improved value to markets.
• Computational Design exploits Moore’s Law of doubling computational power and capability in half the time. Designers and engineers work together using computer algorithms, simulations, and data analysis to create computational media that speeds the design process to continually morph based on ever-changing data and technological delivery.

The practice of design is deeply rooted in classical design in that it took the human craft model and ported it to digital technology in the form of user experience and user interface. For years, designers labored to create digital assets for interfaces, and there was a lot of manual customization of common interface elements. Digital design is now commonly called the “Figmaization” of design, referring to the prototyping platform creating user flows. While from a brand perspective, having thousands of versions of buttons, mastheads, dropdowns, and other UI elements can differentiate one platform from another, it is also very time-consuming and expensive.

The problem for John is that “design is poorly designed.” Design is quickly being open-sourced, and most design outputs are becoming automated. Canonical design outputs can now be automated by AI to create hundreds of variations based on existing ingested pattern variations. Anyone can create what they need without design knowledge, just some rudimentary design judgment. This turns design into efficiency of options, production, and cost savings.

Is Design ag(ai)nst AI possible?

As a trained designer, I find increased automation liberating and troubling for design practitioners. Liberating in that many classical design skills can be automated, easily tweaked, and quickly prototyped, allowing designers time for higher-order thinking and concepts. It is troubling because the craft of design takes years of learning a craft of making, only to see it automated as just good enough output, turning design into a meme and a shallow stereotype. The diffused design communities struggle to find relevancy because of these trends and a real fear that designers with classical knowledge and skills will no longer be relevant, needed, or affordable.

Never in human history have we disconnected the act of creation from a human creator (that is why humans created gods to do the unexplainable). Attribution to a person is a very rooted concept in creation, and when this is de-linked, it creates very thorny problems such as what are authorship and originality? Any AI system trains on established content and then generalizes this content into seemingly “organic” responses. It then can iterate to the point that it improves itself based on human interactions. Many professionals who use chat-based systems find them compelling and reasonably accurate.

Because any profession is the summation of its historical knowledge, techniques, and outputs, adapting to disruptive and transformational change in short periods of time does not bode well for any profession to adjust quickly. Professional knowledge is essentially a canonical knowledge system that can be automated. We have seen this with general accounting, and it will be integrated into law, medicine, and business, where much of the day-to-day skills and knowledge will become automated.

Classical design, design thinking, and computational design may merge into “bespoke design.” This may take various forms, connecting individualized designer knowledge, skills, and experience augmented by machine learning and sandboxed design patterns connected to limited production prototype platforms for unique design outcomes.

Whatever the future holds for AI and design, it will be a very disruptive and uneasy relationship. It will upend classical design in terms of knowledge of creating historical canonical artifacts, and the merging of physical and digital artifacts will create new typologies. Design will always be about turning intention into intentionality by identifying greater value. Designers then marshal social meaning and social and economic resources to create products and services that move markets or increase people’s agency for the common good. This means humans must feel they have agency and freedom to create new patterns by breaking old ones. AI could assist us in that effort or become the status quo that keeps us from achieving our desired potential.