5 Principles for Responsible AI Design
Editor’s note: A version of this article was previously published on the Salesforce Design blog on Medium.
As architects and technology creators, it is not enough for us to deliver only technological capabilities — we also have an important responsibility to ensure that our solutions are safe and inclusive. This article explores five principles employees at Salesforce rely on to ensure AI is used responsibly, accurately, and ethically as part of our Ethics by Design approach.
Architects have a unique opportunity to drive toward a more ethical use of AI. Ninety-two percent of IT leaders say ethics are a critical part of any AI strategy, while 41% note their organizations already use chatbots and the rate of adoption is expected to grow 95% within just two years. With that type of digital transformation afoot, we have a responsibility and opportunity to ensure that the underlying AI is implemented responsibly. To start thinking about ethics, let’s consider the fact that 89% of Americans now say they would be willing to share personal health information with employers to help keep themselves and their coworkers safe from COVID-19. What happens to that data, however, when the threat of infection diminishes? Can we ensure technology helps, rather than harms, the people it serves?
Unfortunately, in the United States, contact tracing efforts have, for the most part, failed to curb the spread of the virus. Manual contact tracing is both expensive and labor-intensive. The out-of-the-box contact tracing solution in Work.com (Salesforce’s suite of back-to-work and back-to-school tools) had two challenges with which to contend:
- Contact tracing efforts often fail because just 19% of Americans say they generally answer their cell phones when an unknown phone number calls, whereas the vast majority of text messages are read within minutes.
- Receiving news that you may have been exposed to a potentially deadly virus from anything other than a caring human being could be both strange and unsettling.
In response, Principal Solution Engineers Gerard Iervolino and Chris Gilmore conceived of Tracey Bot — a contact tracing bot that increases contact tracers’ effectiveness by augmenting their work. I teamed up with Iervolino and Gilmore to help create a humane contact tracing bot experience using Salesforce’s Ethics by Design methodology.
“Working with Yoav and the team truly leveled-up our design process. We learned it’s important to build technology from the perspective that our customers are different from us in ways we cannot always anticipate. ” — Chris Gilmore, Principal Solution Engineer of Tracey Bot
Five Principles for Responsible Design
The five principles we sought to bring to life through our Tracey Bot design process are responsibility, accountability, transparency, empowerment, and inclusivity.
Responsibility
Part of taking responsibility for the technology Salesforce puts out into the world includes empowering the people who implement our products — such as admins, architects, and developers — with the knowledge necessary to succeed in using it. This is why we created a comprehensive installation guide with Tracey Bot.
As an example, one directive we offered in the install guide was clear — only send a message if an agent is going to be available to handle an inbound call and provide the caller an estimate of the likely wait time if possible. Why? A poor outcome would be receiving a message about possible COVID-19 exposure and having to then wait on hold interminably to speak to someone — or not having an agent available at all. We led with a human-first approach to the design of Tracey Bot to help ensure our customers’ success.
Accountability
When building the interface between employers and employees, businesses and customers, or students and schools, accountability is key in creating interactions that foster trust. To engender trust, employers should be explicit with employees about what data are being collected, its purposes, the retention policies, and how the data will be secured. That type of accountability will build relationships that stand the test of time.
Iervolino offered some insights into the technical side of crafting accountability into the Tracey Bot experience:
“The majority of Tracey Bot leverages out-of-the-box constructs inside of the work.com contact tracing data model. For example, there are a number of Lightning Flows in the package that were created based on the existing out-of-the-box end-user evaluation Lightning Flow that was created for manual contact tracing. We used these existing constructs because they had already been approved by the product and legal review processes, and were created with the CDC guidelines in mind.”
“Minimal custom metadata was added in the package, including just a few fields on the Person Account and Contact Encounter Participant objects. Tracey Bot even uses the existing consent management framework in the contact tracing Health Cloud package to communicate the reason for outreach and capture authorization from the end-user.”
Transparency
We believe that, as simple as it may seem, a bot needs to identify itself as a bot, and any attempt by a bot to pretend to be a human agent deceptively undermines trust. As a result, our Acceptable Use Policy at Salesforce states that any Einstein bot must immediately disclose this information in the name or in the chat itself, hence the name Tracey Bot.
Empowerment
Empowering users with information that’s crafted in a caring and supportive way benefits society as a whole. We wanted to mitigate the tension that comes when someone receives serious health news from a non-human bot. We couldn’t imagine an experience that would include receiving a message like: “Hi, you may have been exposed to a potentially deadly virus. Press 1 to hear more.” So we carefully crafted the Tracey Bot messaging scripts using templates from the Centers for Disease Control and with guidance from our Conversation Design Principal, Greg Bennett.
An example of this is in Tracey Bot’s first message:
“In order to stop COVID-19 from spreading in the community, we follow up with people who have been potentially exposed. We want to work with you to help you get the care that you may need.”
“Contact tracing is about creating a map of people’s relationships, and that’s very intimate. And it’s coming from a sterile form of communication, a bot. Conversation design can bridge that gap by crafting messaging that is authoritative but not scary and caring but not too friendly. This is similar to the messaging nurses, and doctors use in their bedside manner. We want to give Tracey some good bedside manners.” — Greg Bennett, Conversation Design Principal
Inclusivity
Salesforce’s Ethical AI commitments state that inclusivity means “AI should respect the values of all those impacted, not just those of its creators. To achieve this, we test models with diverse data sets, seek to understand their impact, and build inclusive teams.” In the context of the COVID-19 pandemic, we are all doing our best to create solutions for a problem that is unlike any we’ve ever experienced. Approaching issues around access and effectiveness with inclusivity in mind can help architects come up with innovative solutions that actually work.
We initially designed our Work.com contact tracing solutions for the ideal world in which everyone answers their phone when a contact tracer calls. However, the reality is that most people do not answer phone calls from unknown numbers, while 90% of text messages are read within three minutes of receipt. So, in the hopes of making contact tracing inclusive and accessible to everyone, we made certain that Tracey Bot could be used across a variety of channels, including SMS, chat, and WhatsApp.
The Challenges of Contact Tracing as U.S. Battles COVID-19. Pew Research Institute. October 30, 2020
Conclusion
As architects and technology creators, it’s imperative we remember that the experiences we craft now are likely to persist beyond the present moment. How we choose to build technologies today has important implications for the life of tomorrow. If we are building our own future, then ethics must be at the root of all we do. It is not enough to deliver only technological capabilities — we also have an important responsibility to ensure that our solutions are safe and inclusive for all.
Interested in learning more?
Learn more about Tracey Bot and how to implement it as part of manual contact tracing with Work.com.
If you’re curious about how to become a more ethically-minded architect and creator, get started with the Ethics by Design Trailhead module. Or explore how you, too, can make a more ethical product by completing the Responsible Creation of Artificial Intelligence Trailhead module.
You can also learn more about Salesforce’s Office of Ethical and Humane Use or Ethical AI Practice.
Read our recent story on How to Choose an AI Strategy for Salesforce to learn how to adopt a process-driven approach for AI success with Salesforce.
Finally, learn more about Salesforce Design at www.salesforce.com/design, follow @SalesforceUX on Twitter, or check out the Salesforce Lightning Design System.
