<aside> 🫐 Shan Wang is currently a Product designer driving meaningful and responsible AI design at Microsoft Research, and has previous experience at Deloitte. At Microsoft, she is designing new AI platforms that accelerate research and development in commercial science through large-scale ML models and new search frameworks, making scientific discoveries more efficient and scalable.

👀 Linkedin

👀 Website

</aside>

Screen Shot 2023-01-31 at 11.24 1.png

Disclaimer: Opinions reflected below are Shan’s own opinions, and are not representative of Microsoft.

🫐 Who are you and what was your personal journey into your current role?

My name is Shan and I graduated from CMU in 2020 with degrees in Architecture and Human Computer Interaction. My journey into product/UX started back in my sophomore year, where I took a coding class and took an interest in projects and research in the HCI department. I started doing more independent studies and research projects with Professor Jason Hong, with a focus on mobile and privacy, and biases in machine learning, alongside some design projects. After graduation, I worked at Microsoft Bing (search engine) before I joined Deloitte consulting in 2021 under their Applied AI team vertical for a year and a half where I did a lot of different projects for clients in aerospace, tech, and finance.

One of my main projects at that time was voice experience design. Being able to ship that experience for auto-finance clients was really interesting. Thereafter, I transitioned to my work at Microsoft research – designing an AI platform in an incubation team called Project Science Engine.

🫐 What is your role as a Product designer in Project Science Engine?

Project Science Engine (Project S) is a scalable enterprise AI platform that allows research scientists to accelerate their research and development process and leverage a lot of the large scale, foundational AI models to understand data across different siloes and modalities such as documents, tables, and even molecular structures. The platform serves different research fields in industries such as biopharma, chemicals, materials and alternative energies. Researchers could leverage the AI capabilities of the platform for a variety of use cases including finding similar proteins/genes for drug discovery.

Our incubation team has over 100 people, including designers, product managers, researchers, front and back end engineers, and applied scientists. Because of the start-up-like environment, designers wear multiple hats and work on multiple projects all at once. Some of the project areas I am focusing on right now are the platform-wide semantic search, ML model management, and, very recently, the new concepts to integrate GPT into the product.

🫐 How do you maintain cohesion when designing such large complex, scalable platforms such as AI tools?

First, I focus on the holistic experience across the user journey and different use cases when using the platform. It’s helpful to identify all the different touch points and entry points such as filter options, searching and import/export workflows. I’ll push for similar search experiences across different touch points (i.e. similar typeahead, suggestion dropdown) to make sure users’ expectations are met and that they don’t encounter unfamiliar patterns.

<aside> 🫐 “We also have to design with modularity and scalability in mind from a component level.”

</aside>

To accommodate future needs and scalability with minimal disruption, we also keep modular design in mind to allow for the addition of elements. In terms of standardization, we also make sure design elements are consistent and presented in a coherent manner. We have designers working directly on design systems whom we work closely with, keeping a constant feedback and communication loop to ensure that we are all aware of each other’s work and input.

🫐 What are some of the most important considerations when designing models and AI tools?

One of the most fundamental questions you should always ask when designing AI products is “why are you using AI and ML to solve this specific problem?” What’s the advantage with this technology over traditional programming? You want to be very intentional with how you use and develop the model, and designers should be involved early on to evaluate the use cases and potential business value. Understanding the context of the initiative as well as the bigger picture of these model decisions is also very important in thinking about how to better leverage AI to solve problems.

<aside> 🫐 “Transparency and interpretability is very fundamental to designing for AI, and we want to help users understand how models from a high level work.”

</aside>

For example, when looking at how we can provide the most relevant and actionable insights users can trust, we can start exploring ways to build in transparency and explainability of the model to users to help them understand how the model works and ways of interpreting the result. Since machine learning is not always 100% accurate, we may also want to give users the ability to correct mistakes the system has made, and assist the users in how the system is working so they can make a more informed decision. Lowering that barrier of entry to make it easy to use would allow non-technical experts to effectively leverage the capabilities and potential of AI.

There are also a lot of considerations regarding ethics, such as the representation of your data and making sure it’s not biased as possible. In this case, we may also need to look into how you identify biases early on, and, if it’s a model, how can we effectively communicate that to users. Ultimately, it comes down to how we provide information to help users make informed decisions more effectively.

🫐 Where do you personally see AI heading towards with design? How would you like to contribute to that future you see?

I think we’re still in the early stages of grasping the full picture of AI, and we’re continuously trying to establish familiar, good patterns in the landscape of Human-AI interaction through professional design practices, trial and error, and usability studies using AI products. As designers, we are able to synthesize a lot of learnings from these projects and contribute to a ‘handbook’ of Human AI interaction practices to uphold.

For example, looking at how we engage with Bing and Google as well as multi-modal interactions such as AR/VR surfaces a lot of new interactions and behavior patterns we need to carefully observe and study.

🫐 What advice would you give people who are looking to take on design roles that overlap with AI?

I would take a step back and think about what you are looking for in terms of design careers. Explore the different types of working environment such as consumers and enterprise models instead of fixating on the technology itself. Be mission driven and don’t get bogged down by the details.

I would also try to understand the technology better at a high level such as how the ML pipeline works and how applied scientists develop and manage models. Both of these would help facilitate conversations in cross-functional teams. Through this, you would be able to better understand the capabilities and social impact design and technology can have.

<aside> 🫐 “You should also consider how the human in the loop fits into the bigger picture – where should human intervention happen in the process?”

</aside>

A quote I resonate with comes from one of the Sunday Letters newsletters by Sam Schillace (CVP, Deputy CTO Microsoft)“AI isn’t a feature of your product, but your product is a feature of AI.” Essentially, he’s referring to how real value comes from understanding the technology and building AI as a platform, rather than adding AI as an afterthought to a product.