Overview

The goal of our process is to ensure teams have a guide to explore key questions and considerations to ensure we are developing ethical products following a responsible process that aligns with our values.

Benefits of using our process include:

• Team alignment during early planning
• Intentional conversations on key ethical considerations and potential risks
• Shared accountability in tracking and monitoring ongoing value and harms

Our Responsible Product Development Process has four key phases that allow us to be successful:

• Vision

  As we start, we gather business requirements and make sure we have a risk mitigation plan in place based on the vision we’ve identified. A shared vision ensures we always prioritize having humans in the loop where they are most needed.

• Framing

  Before actively engaging, we pause and ask important questions to get clarity. We always want to weigh the need for generative AI against its risks and ensure we have identified valid use cases before getting started.

• Discovery

  Next we dive into research and ideation while involving the affected communities as much as possible. We want to make sure we are assessing the available data sources while identifying embedded biases and how we might mitigate them.

• Delivery

  As we move into delivery, we identify an MVP to test and iterate on as we learn more about user needs. Our team moves into a continual process of model evaluation, training, deployment and monitoring in order to identify issues as they arise.

• Scaling

  Once we have learned from an initial MVP, we can begin to expand on the identified value and reach of the service. We will create a plan that is sustainable and provides steps for teams to have long term stewardship.

Guardrails

Responsible Product Development introduces four guardrails to ensure proper product governance:

• Ethical champion

  We identify someone whose responsibilities are to ensure a minimum viable quality throughout development. Depending on the nature of the product this role may be assumed by one or more people on the team.

• Community partnership

  Given the nature of our work, we deepen our relationships with our community partners in order to make changes as needed. Many of the potential solutions we have worked on in the past have direct effects on traditionally marginalized communities, so it is important that we invite them to the table from the beginning.

• Moments of truth

  As we move the process, we pause and check on our alignment to our original goals. We establish ceremonies between each phase of work to measure key responsibility metrics that we have stablished and course correct as needed based on what we are learning.

• Experimentation

  Advances in this space are continual, so it’s imperative that our teams remain adaptable and flexible in practice and outcomes. We adjust our approach as we learn from ongoing experimentation with a wide variety of tooling, including popular models, third party services and open source initiatives.