Responsible AI Checklist

tl;dr

At Spoke, one of our core values is the belief in using technology ethically. Since we work with and develop AI products, we are building a meaningful process around Responsible AI (cf. Google AI, TDS).

These core ethical practices should be incorporated into our processes. Our approach for that is the creation of a check – much as we have for adding unit tests and other engineering best practices – in our code contribution flow.
‍

Objectives

• Collaboratively build a list of important considerations in building AI responsibly, which can be updated continuously
• Ensure those values are considered during development and not only in retrospect or, worse, “in theory”
  ‍

User Value

👉 "Responsible AI is the practice of designing, developing, and deploying AI with good intention to empower and fairly impact humans and society"

Crucial to this objective of responsibly delivering user value is ensuring that AI is used to help with problems users actually experience and that the AI achieves what they want it to.
If a task is enjoyable for a user, then it probably shouldn’t be automated away with AI. Conversely, if an unenjoyable task can be automated with AI, then we should engage in a serious effort to ensure that this task can be automated effectively – across all our user groups.
‍

‍

Challenges

• There are an increasing number of resources on Data Ethics available, however, many of them focus on particular areas, such as Fairness in AI or Data Security. We want something comprehensive, yet not overwhelming!
• In the case of Data Security & Collection, a reasonable amount of robust and legible legislation exists within the EU, which can guide you in framing the relevant checks. Unfortunately, in many other other areas there remains a dearth of governmental and regulatory input…
  ‍

Resources & Outcome

Some resources that we drew from in developing the checklist:

• People + AI Guidebook (Google)
• Principles of GDPR (EU)
• Example Ethics Checklist (data.org)

The checklist itself was simply included in a PR request template for our data services, living within our Github repository. This template then automatically opens for a new PR, with the requester being responsible for considering the included items and the reviewer for discussing any concerns. See the checklist template included with each PR below:

Conclusion

The focus of this work, to begin with, was for processes within our data team, but as per our core values this can and should be extended to the entire tech and product organisation. Many of the questions in the checklist equally apply in developing back- or front-end applications, and many more are explicitly product-related.

Potential other places within Spoke / other organisations where this checklist could be implemented into existing processes, include Product Discovery, Tech PRs, RFCs, and many others.

In case, you would like to implement a similar checklist in your team, or would like to know more about this topic, please feel free to reach out to Gráinne with any questions!