2019 – 20202019 – 20202019 – 2020
Origo
A data platform for governing data platforms.
Harvard University Design Engineering
Master's Thesis
About
After the Cambridge Analytica scandal, the tech industry goes through an ethical reckoning. This thesis for the design engineering program at Harvard synthesizes and distills the cultural discourse around tech ethics into a speculative platform that pragmatically resolves the problem. Although speculative, the platform is not a video but an interactive digital prototype that simulates an object and context in a fictional near-future. The undertaking of this project was also a methodological synthesis of wicked problems, design thinking, and data engineering. Methods developed from this class was applied to the USCIS and generalized into a studio curriculum at the University of Michigan.
Problem Spotting
I worked in Silicon Valley at an enterprise data management company from 2016 - 2018 that was an AWS competitor.
The Cambridge-Analytica Data scandal broke out April 2018. My Statement of Purpose was written November / December 2017. At the time of writing, I saw theoretical problem of tech ethics as an interesting intellectual exercise.
The theoretical framing was possible due to my undergraduate training in architecture and design at UC Berkeley. The program has a history of interdisciplinary engagement with social justice and wicked problems.
My continued time in practice led me to focusing the thesis on data ethics.
The Cambridge-Analytica Data scandal broke out April 2018. My Statement of Purpose was written November / December 2017. At the time of writing, I saw theoretical problem of tech ethics as an interesting intellectual exercise.
The theoretical framing was possible due to my undergraduate training in architecture and design at UC Berkeley. The program has a history of interdisciplinary engagement with social justice and wicked problems.
My continued time in practice led me to focusing the thesis on data ethics.
Statement of Purpose
The core problem that I wrote in the SOP that is the subject of this thesis are the "inscription errors of unconscious human bias.
Photo: Xinhua / Barcroft Images
01Problem Spotting
01
01Problem Spotting
Historical Analysis
The problem of "tech ethics" is very broad. Since I wanted to design a platform intervention, I approached the problem atomically by reducing system components to their smallest parts.Instead of looking at the system outcomes / outputs through an enumeration of risks and benefits, I looked at the system inputs to guide the research and development of my problem statement.
Starting at the macro-level, I needed to understand the range of tech's ethical problems from multiple perspectives. I decided to collect case studies from media articles due to their public-facing nature and contribution to mass understanding of the discourse. I then structured the articles to identify the stakeholders that needed to be accounted for in the platform design.
Starting at the macro-level, I needed to understand the range of tech's ethical problems from multiple perspectives. I decided to collect case studies from media articles due to their public-facing nature and contribution to mass understanding of the discourse. I then structured the articles to identify the stakeholders that needed to be accounted for in the platform design.
Timeline of Silicon Valley Disruption
Living and working in Silicon Valley from 2009 - 2018, I had front row tickets to the tech industry's innovation and disruption.
When I left, Silicon Valley politics became inter/national politics.
The timeline shows a sample of major events during this time, including my own company's acquisition by ARM Holding as part of their device-to-data IoT strategy.
When I left, Silicon Valley politics became inter/national politics.
The timeline shows a sample of major events during this time, including my own company's acquisition by ARM Holding as part of their device-to-data IoT strategy.
Learning 1
There are four major stakeholders involved in tech ethics: 1) venture capitalists, 2) employees / workers, 3) consumers, and 4) regulators.
"Cities" is not a stakeholder, but technology impacts urban space.
"Cities" is not a stakeholder, but technology impacts urban space.
02Historical Analysis
02
02Historical Analysis
Mainstream Media Landscape
The primary characters exerting influence on the problem of tech ethics is Big Tech and the media.
Given the history of yellow journalism, the media is not an impartial reporter. It fails to have the domain knowledge for understanding the root causes of problems and is incentivized to sensationalize. What the public receives from the media are observations of effects of mechanical failures, not the mechanical failures themselves. It is critical to be able to distinguish between the two in order to mitigate internal bias as a designer. Scapegoating is unethical.
At the same time, the media has immense power to create change through influence by creating a shared cultural consciousness. Like how it is important to assess the failures of tech, it is equally important to understand the failures of the changes that have been introduced due to media reporting. A system map synthesized the literature review.
Given the history of yellow journalism, the media is not an impartial reporter. It fails to have the domain knowledge for understanding the root causes of problems and is incentivized to sensationalize. What the public receives from the media are observations of effects of mechanical failures, not the mechanical failures themselves. It is critical to be able to distinguish between the two in order to mitigate internal bias as a designer. Scapegoating is unethical.
At the same time, the media has immense power to create change through influence by creating a shared cultural consciousness. Like how it is important to assess the failures of tech, it is equally important to understand the failures of the changes that have been introduced due to media reporting. A system map synthesized the literature review.
Systems Map of Media Effects
Media has the capacity to provide information at the collective level, but can also cause hysteria. Both has gone up over time as tech draws national attention.
The information reported are usually outcomes and effects, not causes, which means public literacy of technology does not increase. Attitude towards tech also becomes increasingly negative, which creates a potential risk of scapegoating.
As governments respond to problems, the regulations are not designed with technology in mind. A case study is the consent form now on every website which do not benefit the public and are only for companies to protect themselves.
Since there is an obvious problem, people want to step in and help. Others, including academics, want to step in and capitalize on a hot moment (Floridi). This creates an influx of information and can lead to ethics-washing (Floridi). Technologists may be caught in a vicious cycle of scapegoating which is, ironically, unethical.
Misfit regulations can also lead to infrastructural bloat which can impact future use cases, particularly in domains of healthcare and climate change, in unknown ways.
The information reported are usually outcomes and effects, not causes, which means public literacy of technology does not increase. Attitude towards tech also becomes increasingly negative, which creates a potential risk of scapegoating.
As governments respond to problems, the regulations are not designed with technology in mind. A case study is the consent form now on every website which do not benefit the public and are only for companies to protect themselves.
Since there is an obvious problem, people want to step in and help. Others, including academics, want to step in and capitalize on a hot moment (Floridi). This creates an influx of information and can lead to ethics-washing (Floridi). Technologists may be caught in a vicious cycle of scapegoating which is, ironically, unethical.
Misfit regulations can also lead to infrastructural bloat which can impact future use cases, particularly in domains of healthcare and climate change, in unknown ways.
Learning 2
Media articles create vicious cycles of scape-goating, misinformation / misinterpretation, and infrastructure bloat,
03Mainstream Media System Mapping
03
03Mainstream Media System Mapping
Semi-Structured Interviews
Once I identified the range of stakeholders, I needed to zoom in and understand one in depth. I chose to research the worker / employee experience since they were the ones who are directly involved and building the products. A solution that targeted this context would impact everything downstream.
I needed to understand their mental model of the situation so I conducted semi-structured interviews and reviewed literature to understand how employees framed the causes of ethical problems.
But what is an ethical problem?
Based on the historical analysis, there is a broad range to what is understood to be an "ethical problem." I needed to understand how "ethics" was identified in this specific context. As an abstract concept, I knew that a singular definition is diffuse and would not be useful. I was not trying to be a philosopher. Pragmatically, the definition and types of ethics does not matter. For the purposes of designing an intervention, how the human identifies and finds an ethical problem is the design requirement that needs to be accounted for.
I needed to understand their mental model of the situation so I conducted semi-structured interviews and reviewed literature to understand how employees framed the causes of ethical problems.
But what is an ethical problem?
Based on the historical analysis, there is a broad range to what is understood to be an "ethical problem." I needed to understand how "ethics" was identified in this specific context. As an abstract concept, I knew that a singular definition is diffuse and would not be useful. I was not trying to be a philosopher. Pragmatically, the definition and types of ethics does not matter. For the purposes of designing an intervention, how the human identifies and finds an ethical problem is the design requirement that needs to be accounted for.
Employee Mental Modeling
Learning 3
Identifying what is ethical requires an internal moral compass which will depend on personal experience.
Learning 4
Standards and Policies codify ethics and is where people usually start their validation process.
04Employee Semi-Structured Interviews
04
04Employee Semi-Structured Interviews
Task Analysis
In addition to understanding the mental model, I needed to also understand the actions employees took and what happened. This would help identify gaps.
Existing Employee Reporting Experience
Learning 5
Managers are point persons for ethical concerns.
Learning 6
Technologists believe they are responsible, but they lack adequate tooling for assessing impact.
05Employee Task Analysis
05
05Employee Task Analysis
System Mapping
Finally, using the employee report of problems as a framing lens, I telescoped out, to create a system map. I traced and mapped the socio-technical environment of production based on my interview notes with stakeholders.
System consequences, inefficiencies, and failures were identified, structured, and contextualized.
System consequences, inefficiencies, and failures were identified, structured, and contextualized.
Organization System Map
I selected a Series B Start-Up in the middle of scaling as a model for my systems map. agile management, design thinking, and quality management frameworks were synthesized into a diagram.
Following Conway's Law, I assessed for gaps in information and communication, then identified where and how bias / insensitivities might enter the system to cause ethical problems. I also identified bugs / deviations / and inconsistencies to be a source of ethical risks due to their impact on the user experience.
Following Conway's Law, I assessed for gaps in information and communication, then identified where and how bias / insensitivities might enter the system to cause ethical problems. I also identified bugs / deviations / and inconsistencies to be a source of ethical risks due to their impact on the user experience.
Learning 7
There is a lack of procedures for dealing with changing regulations in frontier domain products (such as bio-technology). This creates complexity for quality management teams. Quality is not fully defined and/or has a shifting definition.
Learning 8
Information and communication silos create time pressures that make it difficult to assess for ethical problems.
06Organizational System Mapping
06
06Organizational System Mapping
Research Synthesis: Problem
As I went through the research, I continued to refine my problem statement.
The final form is as shown.
The final form is as shown.
Problem Statement
It is how* we design product, not what product we design, that leads to ethical problems.
* how = people + process
07Problem Statement
07
07Problem Statement
Research Synthesis: Object
In keeping with a systems approach, the "process," the "people," and the relationship between the two have been researched. However, "ethical problem" is still extremely broad and dependent on individual evaluation.
Ethics Synthesis Diagram
Ethics is concept understood and manifested in different ways. Some are implicit, others are explicit. Understanding what is ethical is also not distributed evenly as individuals always have their own perspectives and understanding of the world. The culture of an organization also varies as leaders influence individuals by modeling appropriate behavior.
Learning 9
Assessing the potential ethical consequences of a product feature is a team sport.
Learning 10
Ethics exceeds compliance and includes culture and community.
08Ethics Conceptual Analysis
08
08Ethics Conceptual Analysis
Research Synthesis: Organization
With the components of the problem identified and structured at all levels of detail, I synthesized everything into a hypothetical task analysis with pain points that would serve as the backbone of the platform.
I then roughly designed the objects in accordance with the pain points.
I then roughly designed the objects in accordance with the pain points.
Task Analysis & Object Design
09Workflow Design
09
09Workflow Design
Branding Strategy
The brand of a para-fictional platform that synthesized the research and operationalized the hypothetical task into a workflow was developed.
A Platform for Ontological Design
The platform is designed to be applicable to any domain problem.
In the following prototype, it is looking at the domain of digital education (MOOCs), but it can also be applied to affordable housing, criminal justice, etc.
In the following prototype, it is looking at the domain of digital education (MOOCs), but it can also be applied to affordable housing, criminal justice, etc.
Vision Statement
Origo believes the future of tech ethics lies in creating a continuous feedback loop between software development and regulatory legislation while enabling a plurality of stakeholders to participate in the construction of data models without needing to be "technical." By democratizing the construction of data models and grounding ethics in everyday context and operations, teams can spot, assess, and mitigate ethical risks in a sandboxed environment while collaboratively sharing their learnings with lawyers and regulators. Technology can be designed to be safe, and regulations can update closer and more elegantly to the speed of innovation.
Mission Statement & Reasons to Believe
Origo connects the product team with the information they need to anticipate and mitigate ethical problems at the point of design.
1
Lawyer-in-the-Loop
Base requirements are collected directly from lawyers so the product team can bootstrap their data model and design ethically without costly refactors.
2
Legible Ethical Value
Protocols that become design requirements can be created and reused by lawyers, traced, and audited at scale so that the product team's work is visible and of consumer value.
3
Input Diversity
Context on design problems are collected through references representing a spectrum of temporal change and diversity of stakeholders.
4
Product Knowledge Management
References and requirements are collected and managed in direct relation to the product data model so that context is never divorced from object.
10Branding Strategy
10
10Branding Strategy
Concept Design
Data objects were designed for interaction based on stakeholder analysis.
Protocol Conceptual Model
The concept of an ethical protocol generalized the different ways that ethics manifested in textual documents into a common object that could bridge between stakeholders in different institutions.
Ethical protocols would also serve to normalize the set of words and their referents into a coherent namespace by pairing lexons (internal vocabulary) with keywords (external vocabulary). As a data object that packaged relevant information, protocols could be traced and a digital space for institutional stakeholders was created for data governance.
Ethical protocols would also serve to normalize the set of words and their referents into a coherent namespace by pairing lexons (internal vocabulary) with keywords (external vocabulary). As a data object that packaged relevant information, protocols could be traced and a digital space for institutional stakeholders was created for data governance.
Protocol Coordination
As a common object, protocols act as a bridge between stakeholders in different institutions.
Namespace Systems Map
The systems map traces information from outside to inside the product at the global, local, and structural level to show how feedback loops can be constructed to counter the ones created by the media, while working cooperatively with governance.
11Data Design
11
11Data Design
Role Design
All the stakeholders interacting with the problem domain were structured, and new emergent roles were identified.
Stakeholder Map & Role-Based Design Futures
A new role was created for the primary user / change-maker: a lawyer from the future. The platform sought to create communication channels for this lawyer to interact with the product design team.
12Role Design
12
12Role Design
Ontological Design
Emergent Domain Ontology
The lawyer develops a domain ontology of concept. The ontology unifies namespaces across multiple concepts. In a unified view, it also counters the problem of jurisdictional fragmentation that comes with intersecting local, state, and government laws by serving as a boundary object at the macro-level.
The ontology can also provide a view into optimizing laws created for a common problem by detecting patterns and reverse salients across and within jurisdictions. Over time, data governance can be complex without being complicated.
As feedback with product team are maintained, the ontology can also serve as aa data collection mechanism for experiments that might be generalized into new laws.
The ontology can also provide a view into optimizing laws created for a common problem by detecting patterns and reverse salients across and within jurisdictions. Over time, data governance can be complex without being complicated.
As feedback with product team are maintained, the ontology can also serve as aa data collection mechanism for experiments that might be generalized into new laws.
Collaborative Data Design
Where the ontology exists for lawyers at the macro-scale, the data model exists for product teams at the mezzo-scale and provides a space for designing data representations where questions of identity, meaning, and semantics need to be discussed.
Given that much of the cost of UX designs are due to architectural changes needed because impedance mismatches, the data model also aids engineers in planning and making trade-offs for product development.
By making a technical object legible, non-technical stakeholders can participate in problem-spotting and designing data.
By abstracting data away from interface, designers can make more conscious design decisions.
Given that much of the cost of UX designs are due to architectural changes needed because impedance mismatches, the data model also aids engineers in planning and making trade-offs for product development.
By making a technical object legible, non-technical stakeholders can participate in problem-spotting and designing data.
By abstracting data away from interface, designers can make more conscious design decisions.
Semanticisation Workflow
Information is extracted and semanticised within the platform to create a real-time emergent ontology based on a collection of data models.
13Ontological Design
13
13Ontological Design
Onboarding Design
I prototyped both the backend and frontend for the platform throughout the project to create the underlying data model / API for the platform and all workflows / interactions.
Provisioning Flow for a Two-Sided Platform
Landing Page w/ Onboarding Resources
Onboarding is critical to the success of employees / users and organizational leadership by providing steps for transition into the product.
The landing page of Origo bootstraps the user requirements-gathering process by asking them to define their jurisdiction and anchoring the product in a context and culture.
It then creates a boundary object by visualizing the data model, and directs users to a glossary to give them a common vocabulary for discussion.
The landing page of Origo bootstraps the user requirements-gathering process by asking them to define their jurisdiction and anchoring the product in a context and culture.
It then creates a boundary object by visualizing the data model, and directs users to a glossary to give them a common vocabulary for discussion.
Documentation and Training
The glossary lays out my synthesis of over five years of practice-based research on designing data models, all brought into the creation of this parafictional platform for ontological design.
For clarity and ease of learning, the glossary eschews the production of new marketing / self-marketing terms to describe what has already been described in other fields to produce a unified vocabulary.
This aids in the creation of a shared multi-disciplinary framework, reduces silos, improves collaboration, and helps in the search for more information.
For clarity and ease of learning, the glossary eschews the production of new marketing / self-marketing terms to describe what has already been described in other fields to produce a unified vocabulary.
This aids in the creation of a shared multi-disciplinary framework, reduces silos, improves collaboration, and helps in the search for more information.
14Onboarding Design
14
14Onboarding Design
Task Flows
Phase 1: Gather Context
Phase 2A: Share Protocols / Develop Protocols
Phase 2B: Share Protocols / Tag Data Model
Phase 3: Develop Product Ethically
15Task Flows
15
15Task Flows
Visualization Design
The visualization of the data model was developed to educate the team.
Data Model Specifications
The devil is in the details. This is the same for data ethics. The visualization is created to surface properties of data relevant to decision-making.
By surfacing this within the product, it also passively educates less data-savvy stakeholders on the nuances of data.
By surfacing this within the product, it also passively educates less data-savvy stakeholders on the nuances of data.
Surface Modeling Data
Many activities occur around the data model. Different views are created of its surface for different interactive purposes.
Data Properties
Visualizing the data model also creates a simple chart of all its keywords.
Properties of the keywords relevant to ethical decision-making are also visualized / surfaced.
Properties of the keywords relevant to ethical decision-making are also visualized / surfaced.
16Data Model Visualization Design
16
16Data Model Visualization Design
Concept Validation
After the concept was fully prototyped dynamically, I found a person who fit my hypothesis for a user. I asked him to describe his current workflow and showed him the prototype. He validated the prototype's utlity.
Being able to associate individual field names... which will vary quite a bit as you’ve mentioned... under legally operative concepts is actually a huge value add. Especially to the extent that someone from the legal team can define what the legally operative term encompasses, someone from the product team can tag the individual fields... what I see this tool doing is providing a common language and framework...streamlining conversations between people who are dealing with the same concepts but in very different worlds.
- Mason Kortz, Clinical Instructor @Harvard Law School & Affiliate @Berkman Klein
17Concept Validation
17
17Concept Validation
Go-to-Market Strategy
For platform enablement, I developed an ethical maturity model to help users make sense of which ethical problems were easier / harder to approach. The maturity model also aligns with their product roadmap.
18Go-To-Market Strategy
18
18Go-To-Market Strategy
Final Review
The thesis was presented May 2020 during Covid lockdown over Zoom.
Attendees: Dr. Michael Smith (Advisor) Andrew Witt (Advisor) Dr. Tom Dayton (Advisor)
Dr. Martin Bechthold (MDE Program Director) Dr. Mary Tolikas (MDE Program Director)
Bryan Boyer (Future UT Faculty Director)
Attendees: Dr. Michael Smith (Advisor) Andrew Witt (Advisor) Dr. Tom Dayton (Advisor)
Dr. Martin Bechthold (MDE Program Director) Dr. Mary Tolikas (MDE Program Director)
Bryan Boyer (Future UT Faculty Director)
19Final Review
19
19Final Review
Next Step
After graduating, I was hired at the University of Michigan to teach their design research studio. The ontological data design methods developed as a part of this thesis formed the basis of that curriculum.
I also presented my thesis work and its curricular evolution at various venues both inside and outside the University of Michigan.
I also presented my thesis work and its curricular evolution at various venues both inside and outside the University of Michigan.
20Next Steps
20
20Next Steps
Data EthicsData EthicsOntologyOntologyData GovernanceData GovernanceData ModelsData Models