Designing backwards: fighting biases and avoiding unintended consequences in AI-driven experiences

A member of our team setting up our speculative design workshop, FutureShift, in a large classroom.

How might we empower people to discuss, critique, and reflect on ways in which AI programs find their way into unexpected areas of our lives? FutureShift is a prototyping workshop for exploring alternative future scenarios and values in emerging technologies.

For my Master’s capstone project in Human Centered Design & Engineering, I created a speculative design workshop to explore the implications of machine learning algorithms in the near future. Participants explored Black Mirror-esque scenarios and created boundary-pushing prototypes that might exist in those scenarios.

Skip to our Process Book where we documented our workshop and learnings.


Project Team:

  • Paul Roberts: Design Strategist & Project Manager
  • Ariel Duncan: UX Researcher

Role: Visual Design Lead

  • Responsible for creative direction, and writing & editing for the workshop and process book.
  • Led literature review and partnered with Research Lead on expert interviews.
  • Collaborated on all other aspects of this project.

Timeline: 10 weeks

Is frictionless, seamless design good for society?

From fake news on Facebook to predictive policing software, we realized we were blind to how machine learning algorithms were dictating many aspects of our lives. As algorithms increasingly make decisions for us, they gain agency without us noticing it.

We wanted to shift the agency back towards people, and let them decide how much power AI programs should have over their day-to-day decisions.

To understand how we want to shape and be shaped by AI programs, we need to create space to challenge presumed virtues and dominant values inscribed in them.

Speculative design gives us an opportunity to challenge our own biases (our “givens”) and critically reflect on what we want and don’t want to see in our future as a society.

Project Aims

  1. Enable participants to explore the social and ethical implications of technological predictions through making a “speculative object” that expressed their (one of many) desired relationships with AI programs or machine learning algorithms.
  2. Display these objects at a “yard sale of the future” where the fictitious products would be sold alongside items one might find at a real yard sale. The yard sale provided a real context to start conversations with visitors around the values embedded in fictitious products.

Final Design

We created a workshop called FutureShift

We conducted our workshop with 4 participants who all came from the University of Washington. Diversity of participants was critical, so the university environment allowed us to recruit people from different majors, from Junior to Ph.D level.

Workshop Materials:

A member of our team facilitating our speculative design workshop, FutureShift.
Participants explored future implications of technological predictions through structured activities. They picked a future “signal” and explored it by considering its social, technological, economic, environmental, and political dimensions. Then they created a narrative using those considerations and their desired relationship with that signal. Finally, they created a “object from the future” using materials we pre-selected for the making activity.
Parts of the FutureShift Workshop which included a warm-up activity, facilitation prompt cards, and prototyping near future objects.
We took an iterative approach to developing the workshop. We created two warm-up activities to make an abstract topic more concrete and relatable. We also developed cards with questions relating to the STEEP (social, technological, economic, environmental, political) dimensions to help people imagine more diverse scenarios. To aid people in expressing their product’s narrative, we participated in the making process.
An example of a speculative object that a participant made during the workshop.
An example of a “speculative” object a participant created. She wanted to help teenagers disguise their faces when facial recognition technology was everywhere.
Examples of objects that participants prototyped during the speculative design workshop.
Other examples of speculative objects participants created to express their point of view about a future prediction.
A member of our team selling items at the FutureShift Yard Sale.
A yard sale provided a real context in which people from our community can evaluate participants’ objects from the future.


1) Workshop

While we didn’t expect to cultivate agency in 3 hours, we saw hints of agency emerge during our discussions and post-workshop evaluation.

A participant holding his prototype of an idea based on "social reputation coach" to undo the effects of race and class biases.
A participant holding her prototype of an idea based on tracking nutritional needs of children.

2) Yard Sale

We had a total of 19 visitors and engaged in conversations about workshop participants’ speculative objects with 10 visitors. We filmed some of the visitors’ responses.

A neighbor who shared her perspectives on personalization algorithms that might inhabit digital products of the near future.

Since conducting this workshop, we discovered other companies have used a similar process.

Design Process

Understanding the issue at scale

We conducted 6 interviews with experts in the field and read over 50 pieces of academic literature and industry reports.

Research goals:

  • Unpack the interplay between algorithms and their impact on culture & society, and uncover methods for revealing how algorithms work (i.e. reverse-engineering algorithm’s outputs).
  • Explore strategies from the field to inform how we could inspire awareness, discussion, and reflection around algorithms and people’s desired relationship with them.
A screenshot of a Skype call with Pamela Pavliscak to discuss her work with emotional design, artificial intelligence, and practical ethnography.

Findings from discovery research

Slide with a quote from Audrey Desjardins, an Assistant Professor of Interaction Design at the University of Washington.

1) People were conflicted about how machine learning algorithms “see” them because they didn’t trust the intentions behind the companies (i.e. Facebook, Google).

But at the same time, they were frustrated when these systems didn’t know enough about them to personalize the experience.

2) People make up stories about how algorithmic systems work–and then try to train the system to better align with their preferences.

People see the data and content that are served up but not how they’re being pulled together or weighted.

3) We risk creating algorithms that exclude and discriminate if we don’t actively challenge our narrow view of the world.

  • It’s hard to challenge a decision made by algorithms when we only see the output – much less hold the system accountable.
  • It’s easy to accept a result as a given when algorithms are perceived to be objective, even when they’re not. They encode human biases in the form of rules, categories, and criteria.

Bottom line:

The ability to change how algorithms work rest with developers, designers, and researchers. Without challenging our own personal assumptions that go into designing algorithms, we recreate current inequalities in algorithmic systems.

Reframing the Challenge

As designers, developers, and researchers, we have an opportunity to shift the agency back to “non-experts” – in a process to highlight people, perspectives, and values that might have been forgotten during product development.

New challenge:

Empowering non-experts to voice their concerns and tensions about how machine learning algorithms are currently designed.


How might we empower people to discuss, critique, and reflect on ways in which AI programs might find their way into unexpected areas of our lives?

Two immediate audiences:

  1. Non-experts: We want to hear from people who aren’t usually included in developing machine learning algorithms or thinking about the long-term future.
  2. Design, development, and research teams: These people have direct impact on creating algorithmic systems. We want to empower them to challenge their own perspectives about “what’s desirable.”

Why it matters:

By working backwards from preferred visions, we can enable tangible actions today while including alternative values and considering the ethical implications of rules, categories, and criteria we embed into algorithms.

Arriving at Our Concept

Push into the “ridiculous” to provoke what we think is a “desirable future”

We took inspiration from Dunne & Raby’s ideas on speculative and critical design and Montgomery & Woebken’s hands-on, participatory approach to challenging conventional values.

To see how we might apply these ideas in practice, we borrowed from Dr. Dan Lockton’s “designing for agency” and A.B. Kocaballi’s agency-sensitive design qualities.

Our “solution” will seek to:

  1. Empower people to challenge the status quo of how the future is currently being imagined and to add their own personal critiques and desires into consideration.
  2. Make algorithms and their impact more visible and tangible.
  3. Include a more diverse audience when we discuss future implications of algorithms.

How can we help everyday people touch, see, smell, and imagine “the future” today?

We decided to modify a participatory workshop framework from the Extrapolation Factory. Participants made speculative, future-facing objects and then placed them in everyday contexts to engage the public in spontaneous discussions about possible futures.

The set up for the speculative design workshop that the Extrapolation Factory created.
Workshop setup from the Extrapolation Factory.
The Extrapolation Factory sold participant-made objects in a real convenience store in New York City after the speculative design workshop.
Speculative objects from the workshop were displayed in a bodega in NYC where unassuming customers can see them next to real merchandise.

Iterating the Workshop Concept

1) From present solutions to future implications

We modified the Future Cone and the Futures Wheel to stretch participants’ imaginations beyond present-day scenarios.

Brainstorming visual frameworks for the speculative design workshop.
Version 1 of the visual framework for the speculative design workshop that we set up in a classroom wall.

But we eventually took out the STEEP lenses (Social, Technological, Environmental, Economic, Political) because participants were getting stuck at categorizing the future signals instead of articulating their desired scenarios.

We simplified the Futures Wheel to keep up the momentum of coming up with implications on society from participants’ chosen signals.

Visual framework of the Future Cone and Storymaking portions of the speculative design workshop we set up in a large classroom wall.

We curated a list of future signals from research firms like Gartner and Intel to provoke discussions and probe for people’s desired relationships with AI programs.

Prototyping the speculative design workshop where participants pick out a few near future scenarios that pique their interest.
Examples of near future scenarios.

To embed their own personal critiques and desires, participants created narratives to articulate their point of view about the implications of their chosen signals.

Even through we removed the STEEP lenses, I brainstormed provocative questions about each lens and created a deck of “STEEP” cards to help participants articulate their thoughts.

FutureShift card deck to help facilitate conversations about the near future scenarios.

2) From barely noticeable to visceral

Participants made tangible objects to playfully materialize their hopes and concerns about AI programs. By using materials that people can see, feel, and smell, we were able to talk about other possible versions of future in a more concrete way.

A prototype session where a participant tests out the object-making portion of the speculative design workshop.

One challenge was to overcome professional designers’ and engineers’ point-of-view already encoded into finished products.

We broke apart finished products to make them more modular and “playable.” We also carefully selected items that loosely communicated how they should be used.

A participant investigates materials for object-making during the speculative design workshop.

An appropriate context encourages people to consider near-future products side-by-side with the everyday and the “outdated” products. This sort of comparison brought out visceral and emotional connections to potential scenarios embodied in future products, and we used them to generate discussions with the public.

Selling ordinary items alongside participant-made objects at the yard-sale-of-the-future event in a residential neighborhood.
Brainstorming public engagement ideas on a whiteboard.

We tested our assumptions around how “familiar” a context would be to ensure that it grounded both the public and workshop participants.

Researcher conducting concept testing to prioritize public engagement ideas.

We landed on yard sale of the future concept based on these criteria: accessibility, familiarity, natural evaluative-ness, and feasibility.

3) From deductive thinking to inductive thinking

Future predictions are often dictated by research firms and academic think tanks. To make it more relatable to everyday life, we tested two warm-up activities: Meet My Algorithms and Algorithmic Self Drawing Exercise.

Several participants placing popular consumer digital products and services on a 2x2 matrix during a warm up exercise for the speculative design workshop.
Close up shot of a warm-up exercise during the speculative design workshop.

We sparked spontaneous 1-on-1 conversations with participants to help them articulate their hopes and concerns about a future prediction.

Facilitator guiding a workshop participant during the scenario selection step.

We also worked 1-on-1 with participants during the prototyping activity to help them “think with” the materials.

In the post-workshop survey, participants said that our facilitation made the workshop more enjoyable and helped them materially express the narrative they wrote.

Facilitator leading a workshop participant in the object making exercise to prototype personalized experiences based on a future scenario they chose.

Walkthrough of the FutureShift Workshop

Our workshop, which we named FutureShift, consisted of several phases.

High-level overview of the speculative design workshop.

Let’s Use an Actual Example

Example of how we moved a participant through the design workshop, took their speculative object or prototype, and displayed it in the yard sale. The purpose of the yard sale is to engage the public in spontaneous conversations about the implications of future digital products that increasingly rely on sophisticated learning algorithms to provide personalized experiences based on users' data.

Experiencing the Workshop

Here are the steps through the FutureShift workshop.

1) Meet My Algorithms

As they entered, participants were invited to chart their relationship with common algorithms (e.g. Facebook, Amazon, Google, Netflix) on a wall. They mapped these algorithms on a 2×2 diagram using two scales: “trustworthy” to “creepy” and “gets me” to “doesn’t know me at all.”

2) Algorithmic Self Drawing Exercise

To reveal the invisible work of algorithmic systems, participants sketched and shared versions of themselves from the perspective of three popular algorithms: Facebook, Netflix, and Amazon.

3) Choosing a signal

Participants were invited to review 35 predictions about the impact of algorithms on culture and select one or two that they’d like to explore further. These “signals” challenge stereotypical ideas about what counts as “the future” by exposing participants to a variety of scenarios about how algorithms might influence unexpected areas of their life.

4) Sorting futures

Participants chose whether their signal about the future belongs in “probable”, “plausible”, or “possible” future. Using a pink thread, they mapped a route through the appropriate section of the Future Cone and into the first arrow of the next phase of the diagram. The Future Cone helped participants understand “futures” in the plural.

5) Storymaking

Participants fleshed out the “world” of their signal, creating multiple narratives that explored the implications on other aspects of life in that future scenario. A deck of provocative questions helped encourage participants to consider different perspectives. We created a two-level chevron as a visual cue to push them to explore secondary and tertiary consequences of their signal’s scenario.

6) Making speculative products

Participants were invited to choose materials that were of interest to them and physically render any aspects of the narratives they created. We placed the materials close to the Future Cone area to help them make connections between the materials’ qualities and the scenarios they were considering.

7) Putting the products up for sale

Participants were invited to imagine their object existing in the context of a yard sale of the future. They created narratives about how their object ended up for sale and the price they would ask for it. Doing this helped them think about how their object might be used, become obsolete, or be discarded.

8) Yard Sale of the Future

We hosted an actual yard sale where participants’ objects were displayed amongst “normal” items one would find at a typical yard sale. Participants’ objects were used to inspire spontaneous discussions about the possible futures they represented as visitors browsed items for sale. We were able to engage passersby in brief but emotionally-charged conversations about the role algorithms might play in the near future.


As designers, we focus around the narrow goals of the user and the business as it exists today and underestimate the long-term impact of things we design.

Without changing our work processes, we won’t be able to change our outcome. When we haven’t been asking “what if?” and “is this a good idea?” it’s harder to parse out which possibilities are actually preferable and which are terrifying, as Black Mirror invites us to consider.

More Projects