DDD Episode 003

The Future of Manufacturing ft. Mike Kuniavsky

Mike: How do you take advantage of all of the essential information you have about people and of all of the capability in the factories to create a new way of inventing things? 

Xuan: Welcome to Design Despite Disciplines – a podcast series from the Master of Design program at UC Berkeley, where we explore the important fissures and emerging territories of interdisciplinary design practice.

This is “Episode: The Future of Manufacturing” featuring Mike Kuniavsky. We are your hosts. 

Xinyi: This is Xinyi.

Xuan: And this is Xuan. 

Xinyi: Imagine… the year is 2054, and you’re thinking about shopping for some new clothes. Your computer intuitively responds by suggesting shirts that it knows you will love. You buy them, because your computer was right. Later that week, you have a long day at work and decide to go to a bar. You sit at the counter and without exchanging a single word, the bartender serves you the exact beer you were thinking about ordering.

If you have seen Steven Spielberg’s science fiction movie, the Minority Report, then that hypothetical reality may not sound so surprising to you. In that movie, Spielberg imagined a future where advertisers spoke directly to individuals with intimately personalized messages. The messages were automatically generated by scanning a person’s retinal patterns, and following their movements.

Xuan: The Minority Report was released in 2002, and it is no longer science fiction anymore. Companies like Google and Facebook generate billions of dollars from advertisements every year. Whole Foods uses cameras to capture data on how humans navigate their stores. Our conversations, behaviors, and decisions are being tracked on and offline, as third-parties race to provide people with increasingly personalized experiences that have been informed with data. When provided with our data, AI and machine-learning are able to convert personalization into monetary value.

Today, Netflix and Amazon probably know what you will watch and buy before you do. China’s largest e-commerce platform Taobao owned by Alibaba can be pretty creepy too. I remember one day when I was talking to some friends about buying a new speaker. And later that day, Taobao’s banner was full of ads for speakers. That example was one of many, and over the past few years, more people are reporting similar experiences. Although the Minority Report was set in the year of 2054, it might be a lot closer to the reality it predicted than we realized.

Xinyi: We are living in a world where data has been used to influence almost every aspect of human life. It is everywhere. While digital services continue to invest in creating highly personalized experiences that leverage tremendously massive amounts of behavior data, similar investments are also being made in the manufacturing world. Data and digitalization has changed the way we work with products and the way we produce products. In 2015, German engineer and economist Klaus Schwab first introduced the idea of Industry 4.0, an umbrella term for smart and autonomous manufacturing, empowered by data and machine learning. Industry 4.0 envisions a future where computers and machines are digitally connected and are able to communicate with each other to create and share decisions based on data. In the scenario of Industry 4.0, manufacturers will become more efficient and productive and less wasteful: they can customize what they produce on short notice, and they can only produce what the customers really want, down to each niche group and even to each individual.

As we can see now, there are some products that have already stepped into data-driven and customized service. In my previous company, we used to work with a client in cosmetics and personal care who wanted to design a facial renewal device, which can collect a user’s skin data, such as skin type, skin tone, and hydration level, and customize its function of lifting, massaging, and toning for each user accordingly. In the near future, it is possible for us to envision niche manufacturing, encompassing smaller lot size and more product variations, serving more diverse users, or in other words, more niche groups. In the far future, there will be personalized manufacturing where products are designed for individuals and produced in ultra-small lot size.

Xuan: However, such a transition has been going slow and we are still very early in the game. We are now in an era of mass manufacturing and  niche marketing, which means that manufacturers can still make hundreds of millions of identical things and target different niche groups with the same product using different marketing strategies. Privately customized products have not yet become the mainstream of manufacturing and data has not yet entered the practice of our daily product.

We talked to an expert, Mike Kuniavsky who is working on applications that can facilitate digital transformation in manufacturing. Mike was one of the pioneers of the UX design in the late 90s, and currently runs a research and development lab in a consulting company. He is super fascinated by what is going on in manufacturing and it’s taking a step ahead to make changes.

Mike: We’re going to go from a world of niche marketing to a world of niche manufacturing. I see the way that product invention is happening in a world of massive amounts of data. I see product inventions happening very differently than it used to. Like it used to be that people sit in a room or you have like some vice-president who wakes up and goes, right, we need to make a blue one or whatever. Then there’s also the IDEO style thing, which is like, Oh, everyone let’s make some post-it notes. Let’s put up some post-it notes and then like, “let’s pick one, all right, that’s what we do for the next two years”. But in a data-driven world, that’s not how things can work. You have a constant stream of data about what people are doing and how they’re doing it. And that should be how, what you use to make your, or one of the inputs for making your decisions as a designer.

Xinyi: One of his major roles at Accenture is to work on solutions to build digital threads and create a simple universal access and continuity to data across products, processes, and people through a cloud-based service.  If you are not familiar with the term digital thread, it is the foundation behind a digital transformation. It refers to a data driven architecture that connects data generated across the product life cycle, and allows integrated views of the product’s data. According to Mike, digital thread plays an important role in achieving niche and personalized manufacturing as it will power the future design process with a source of data on design, manufacturing, and usage.

Mike: So, you know, one of the things that Accenture does is we actually build all these cloud-based services and we’ve really been thinking very hard about digital threads, about how do you connect the different pieces of everything with not a single representation, because you’re going to have many, many different kinds of representations, but a single way to connect all the representations. So, what does that look like? How do we create that? We kind of don’t know, but we’re kind of thinking about different pieces of it.

Xuan: As we mentioned before, digital thread is the foundation behind a digital transformation, which enables every product to be connected to a set of comprehensive, readily accessible data about it. With the emergence and rapid development of new technology, manufacturers nowadays are able to acquire enormous data in different ways. For example, There is customer interest data generated by CRM systems. There is product usage data from the IOT system. There are also manufacturing processes, such as computer numerical control machining that allows manufacturers to control a range of complex machinery with pre-programmed computer software. In the designer’s world, there are also computerized design tools, such as CAD and PLM to allow designers to define products with high precision. But we still seem to be far away from niche manufacturing and not to mention about the personalized manufacturing to have every product and every service connected with digital thread. Take Apple watch as an example.

Mike: You know, if you look at the way that Apple watches are made, each single watch is carved out of a piece of aluminum. They can carve with a CNC machine. They could carve it out on a different shape. They could carve it out or a different way, but they don’t. Well, why is that? You know, Apple also has enormous amounts of CRM data about how people use their Watchers, what people do with them. They have an enormous amount of product usage information. How do I use my watch versus how do you use your watch? The watch knows that the watch has all of it,  but they can in fact reconfigure their product lines very quickly to do it, but they don’t.

Xuan: So the question is why don’t they do that? What are the constraints that keep us away from moving into niche manufacturing? Here are several hypotheses.

Mike: Here’s a hypothesis one, you know, right now analyzing all that data takes a lot of time. And a lot of specialized skills designers don’t have that. They don’t have time for that. The second idea is that there’s no digital read. There’s no kind of connection between the different pieces of the design process. You know, maybe there’s a PLM system, but there isn’t anything that really connects. From the customer and their data into the design process, through into the manufacturing process, through into the use process, through into the disposal process. There’s nothing that connects that right now. And finally, my theory is that the design tools are not designed to, to create these product variations, our tools themselves are not sufficient.

Xinyi: Mike also points out that some existing attempts on achieving niche manufacturing such as Nike ID and miadidas failed in a way that they’re asking for too much design and configuration work from the users, instead of the professional designers and the AI-assisted or CRM data-driven parametric product generation. This notion resonates again with the importance of building digital threads to enable data-driven decisions.

Mike: I think it’s because all of the current systems make end-users do too much work.  People don’t like configurators, people don’t like to do that. Like if people wanted to be designers, they would have a different job, you know? You’re essentially making them do a job they didn’t want. So the way that I see this working is essentially that you want to give people kind of the experience they have with shopping, where they see lots of different options and they are predesigned for them.

Xinyi: According to Mike’s theory, the most important reason that refrains us from achieving personalized manufacturing is the third one: our current design tools are not sufficient. The way that Mike interprets the problem of manufacturers and designers’s capabilities of working with data and AI reminds me of an article called UX Design Innovation: Challenges for Working with Machine Learning as a Design Material from Dove and Halskov. In the article, the authors suggest that one of the reasons that machine learning has not experienced extensive design innovation compared to other technologies is the lack of prototyping tools for designers to work with machine learning. They also call attention to the need to develop effective tools and techniques that can help designers envision the potential of working with machine learning and allow them to incorporate it into their works. Similarly, Mike also highlighted the importance of computational design as a solution to increase designers’ capabilities of working with AI and data.

Mike: Computational design embracing the notion of computation at the most fundamental level. It’s not a new idea. It’s existed for a long time. These days it’s called generative design, it has been called parametric design. So computational design. I’ve kind of identified these four components. So the first one is software defined everything. What you get is that everything becomes not a single product design. Everything becomes a Metapod. So it means that everything is encoded as software, not a value in software. And this is where I firmly believed that like a model of a product should generate not just the physical appearance, but also the firmware that runs on that thing. It’s agile, which means that it is rapidly iteratively adapted using data, and that everything is connected with digital threads. If it’s a piece of electronic hardware, we should generate the mechanical models. If it is a mechanical device, it should generate all of these things as part of the same representation. Everything should be ideally connected to its own digital twin up on the cloud so that when something happens, there’s either in the model or in the actual physical product that the model represents. There’s a change up on the cloud, you know, representation of that object, so that you can collect all of this information so that you can make it better.

You know, I’ve often thought that my laptop should pick out what my next laptop should be. Cause my net and my laptop knows a lot more about how I use laptops than I do. And if it could pick out my next laptop for me, I’m not sure why it doesn’t.

Xuan: As we move from computerized design to computational design, the role of designers will inevitably change. There are several concerns around the race of AI will displace or disrespect designers or eliminate the human values in the design process. Just like what we have been witnessing in the field of graphic design or architecture, a machine learning algorithm can generate hundreds of posters within minutes and the program can automatically generate a plan for the architect. So will AI replace designers? However, in Mike’s theory of computational design, instead of being replaced by AI, designers need to collaborate with it. Because humans and machines can share different skills and specialties. Well, we will talk about that later. So how can designers and AI  collaboratively and effectively work as a team? In response, Mike developed a design framework that facilitates collaboration between designers and AI.

Mike: For making your decisions as someone who is going to be inventing new things. And the, and the other thing is, is that so many factors and so many things are getting so automated that you can actually make those decisions much more quickly than you used to be able to. So how do you do both of those things? How do you take advantage of all of the essential information you have about people and of all of the capability in the factories to create a new way of inventing things? So that is the kind of core idea that we’re trying to address with this framework.

Xuan: Mike first introduced the three basic principles of computational design in his framework. Firstly, there are designers and AIs working as a team. Secondly, designers and AIS take on different roles as they work together on shared representation. Lastly, everyone does what they are best at.

Mike: And also a really, really core component of this is thinking about how do you make everyone do what they’re best at? You know, what do AIs do really well? Well, they can take on computer tasks that only computers can do. They can analyze enormous amounts of information. They can evaluate thousands of variants. They can prioritize, they can run finite element analysis. They can do all this stuff in the background. They can distribute ideas, they can do all of this stuff that a human being can’t do. And so that’s their strong point. So you let them do that. Humans. On the other hand, I have to do the things that only people can do. Only people can define goals because people have to represent and they have to be responsible for the end result. Only people can evaluate simulations. For unquantifiable criteria, not everything is going to be quantifiable, not everything like we’re never going to get to a point where we can parameterize everything. Always. There’s always going to be ambiguity. There’s always going to be that. So human beings have to be the ones that resolve that and human beings also have to make trade-offs because human beings are the ones that essentially have to make those choices.

Xinyi: According to Mike, instead of doing user research, finding insight and building solutions by hand, designers today are more like decision makers to create constraints and relationships, define a general style and set functional goals, which is very different from the previous working process of designers. And more importantly, the capacity of designers improves a lot.

Mike: So what we want to do is essentially we want to use computational design, to use data and AI to increase the capacity of designers so that it’s no longer linear. It is now exponential. So one designer can now do a lot more work or create a lot more different kinds of designs than they used to and in order to be able to support these niche markets.

Xinyi: In order to make the collaboration process more implementable to others, Mike built the framework based on a classical design model called “Double Diamond”. Double Diamond is originally a design process model popularized by the British Design Council in 2005 and adapted from the divergence-convergence model proposed in 1996. It suggests that the design process should have four phases: discover, define, develop, and deliver.

Xuan: When AI comes into the design process, what will be the new version of double diamonds? Specifically in the discovery phase, human and AI need to capture business and users’ needs together. While AIs are good at performing quantitative research, designers can define the goal for the parameter space. Next in the Define phase, they need to specify problems, functions requirements and initial parameters based on the research conclusion in the last phase. So they work on a shared representation of intelligence requirements system and style grammar system in which designers work as a system analysts and AIS as constraint managers. In the next exploration phase, designers and AIs use generative design tools to generate possible solutions where designers compose generative design systems, and select the fitness functions while AIs generate multiple possible solutions and refine solutions based on the feedback. Finally, in the last evaluation phase, designers evaluate the subjective design functions and provide suggestions to select solutions and move forward, while AIs will provide objective cooperations and capture data feedback from users.

Xinyi: It is worth noting that this Double Diamond framework is agile, which means that designers can always loop back and iterate the process over and over to make adjustments and improvement at any of the stages to create a better result. The core idea is that the goal of the design may be, and nearly always will be underspecified. The designer’s aim is to provide a set of constraints that allow for certain, presumably desirable, modes of evolution.

Mike: For example, like the acoustics of the room that it’s in, we also know what kinds of things people ask it to do? Like, are you playing music or playing news? Are you playing your game? Are you doing something? So we know something about that. So what we can do is you can start with that and we can start by analyzing that information. Then we can go and we can say, okay, now, we want you, we want to have a parametric smart speaker that fits into your home. Like essentially Google Alexa smart speakers are designed to be kind of as generic as possible. They’re designed to be these like fuzzy cylinders because no one objects to fuzzy cylinders. but in reality, it’s because they’re afraid that someone’s not going to buy it because it’s not going to fit into their, their world. What if you could walk around, scan a space, understand the environment and you could say, Oh, your house, isn’t this style. You have this high modernist style where you have a lot of Dieter Rams stuff, or you have this like, like as a, my house, I live in a 110 year old house in San Francisco. And so like you want your smart speaker to fit into that space. So you know, you can have it, figure that out. Then you can essentially say, okay, well, what do you prefer? Like let’s generate. So there’s a thing that you’ll notice in the third section, which is called the customer genome, it’s a project that my team did a couple of years ago that essentially defined people’s rather than profiling people, we try to profile people’s preferences by what they did. And so you create this genome rather than a specific instance of a person. You create this representation of a person of a person’s likes. And so you can kind of combine all of these things. So then can I ask the person some questions? You can define some constraints.  This is the part where the person might, the end customer might be doing this or designer might be doing this because they might be selecting these constraints. Then that then generates a whole bunch of different components. The AI kind of goes through this entire analysis of all of the different economic performance, functional factors and, It lets you simulate how it would sound or in your space because you know, you could walk around with your phone and take a lot of acoustic readings. And so you could get a simulation of the sound that as it’s going to happen in the space where it’s eventually going to be put and then you give it to a designer curator and they select the final things that they can then present to you or you select as the end consumer the final, right?

Xinyi: The example is quite inspiring in a sense of creating a unique speaker that is specifically designed to match a person styles and values. It can save customers the effort of choosing from different existing products or manually customizing one that best answers their needs. While personalized manufacturing will give people the pleasure of uniqueness and conveniences. There are also many criticisms around the topic. After learning about the backgrounds, principles, and practices of personalized manufacturing, we also shared our criticisms with Mike.

Xuan: The first critique is about sustainability in niche or personalized manufacturing. When humans step into the mass manufacturing era. The number of products was growing exponentially, hundreds and thousands of commodities enticing people to buy things they didn’t really need. This is what we know as consumerism and the cause of this problem to hurt our environment by producing garbage and waste.  Many of the doubts towards niche or personalized manufacturing is the high number of variations and customizations might feed consumerism instead of encouraging the reduction of the waste.

Mike: You know, we’re in a late capitalist economy where everything feeds consumerism and that’s not an excuse, but I think that one way to ameliorate it is by essentially using the digital twin of various products in order to be able to essentially identify how you fix them, how you unmake them and also you only make the products that people actually want. You know, think about   what one of the possibilities in computational design is that only the products that people actually want get manufactured rather than having millions of products. They don’t want, you know, fast fashion is a great example, I guess, a huge waste. Right. You know, we all know how fast fashion produces, you know, hundreds of tons of waste  every year. How can you do that? Well, one of those to undo it is to not make things people don’t want.

Xinyi: Our second criticism touches on the ethics of data. We suspect that personalization will reinforce the bubbles between people. See what is happening on social media. The practice of personalization has been frequently discussed in the internet world. Just like Mittelsatdt points out in the article, The Ethics of Algorithms: Mapping the Debate, the personalization algorithm is “treading a fine line between supporting and controlling decisions by filtering the information presented to the users, based on the understanding of there based on the understanding of their preferences and behaviors”. Users’ autonomy in decision-making is harmed when the information relevant to themselves is in fact determined by the interests of the producers above the interest of their own.

The same thing can also happen in manufacturing and design. As Mike previously suggested, he sees niche manufacturing and even personalized manufacturing work in a similar way to the experience of shopping: giving people a range of pre-designed and pre-created choices generated from their personal data. While these products may successfully match the interests and the expectations of the users, they might also confine the users in the same bubble by always offering relatively similar and safe choices and exclude them from exploring other possibilities besides their existing preferences.

It is interesting that global chain fast fashion companies, such as Zara, have released different products in different cities. There is also an Italian clothing brand called Benetton who sells specific colors of sweater in different cities by looking at the data and knowing what colors are selling well in what parts of the world. People who become aware of the situation and are interested in getting the clothes other than the ones offered in their current cities, actually have to travel from one place to another in order to get the products. So what if AI and machine learning limit the possibilities a consumer can choose or tend to have? What if the users are pushed to make the “institutionally preferred action rather than their own preferences”?

Mike: We, in a sense, live in a kind of a monoculture. But like that to me, I think, is a product of the industrial revolution style manufacturing, where you have like, One design and just sell it everywhere. And every design is a global design. You’ll probably still see that when you go to places that are really different than kind of the, very, affluent places in the world, like you go to,  places in India or you go to Nigeria or you go to Peru. I mean, at the same time, it does allow for more individual vitality in people, in a sense theoretically, where someone can say, Oh, I’m only going to wear green. And you know, before it might’ve been really hard to find only green clothes and it would’ve been very, very hard. But now you might be able to have a lot of choices for clothes that include a lot of green.

Xuan: We are only just beginning to understand the potential of data and AI. And we have barely begun to experience the consequences. So combination of AI and data has resulted in several new domains of science, engineering, and designs that are growing rapidly. As our community embraces and applies these relatively new technologies, it is important to know that there are a number of questions that remain unanswered. What principles might be used to apportion authorship? Who is ultimately responsible when an AI powered thing is released into the world. And who do we blame? When do things act autonomously and make harmful decisions?  When we will get serious about universally addressing how data mining violates People’s privacy countlessly times every day. Will we ever be the owner of our own data or might we profit from the value it has? The Minority Report argues that a critical shift for the future of humanity will be our merging with data becoming one with them. Today, we can see that along the past exists something worse, being manipulated by and represented by data.

It is true that modern mass manufacturing provides convenience and choices for a significant number of people, but there are still massive problems in sustainability, privacy, and traceability. Whether we like it or not, it seems like corporations are more interested in racing towards the future and embracing new technologies without reservation. Today the more data you have, the more powerful you are.

As designers, we have a responsibility to represent the people who are affected by the things that we contributed to making. As debate around the balance of profitability and ethics evolves , we can all act on our responsibility today by starting the conversation and exposing people to a more accurate view of the AI and the data driven system that guides many of their experiences.  By sharing knowledge, maybe 2054, won’t be so bad.

Xinyi: You have just listened to “Episode: The Future of Manufacturing” from the Design Despite Disciplines podcast series. Over the course of the Spring 2021 semester, eight teams of MDes students researched, interviewed, presented, and produced episodes featuring invited speakers from the colloquium. We’d like to thank Mike Kuniavsky for sharing your time and insights with our class. To learn more about Debates in Design and the Berkeley Master of Design program, visit design.berkeley.edu.