How an AI Ethics Committee Works (Jasper’s in Particular)

AI ethics is a complex and multidimensional field that requires a deep understanding of both technology and human values. This understanding is increasingly important as AI’s capabilities advance and more people use the technology professionally. Based on that understanding, companies that make AI-driven products and those that use said products should enact AI policies that are ethical, unbiased, and proactively mitigate concerns. 

Jasper is a company that both produces and uses AI systems. It was vital that the business develop a framework around how employees should ethically build and employ AI across every aspect of its internal- and external-facing work. However, leaving this responsiblity to a single person would be challenging and potentially irresponsible if it weren’t their full-time role. So volunteers were asked to create an internal AI ethics committee to address some of the biggest AI concerns of the moment. 

In episode 7 of The Prompt Podcast, host and Jasper’s Head of Enterprise Marketing Samyutha Reddy spoke with two members of the company’s internal AI ethics committee — VP of Marketing Meghan Keaney Anderson and Director of Information Security John Bullough. The leaders discussed the ins and outs of what Jasper’s AI ethics committee does, what their meetings are like, and how the group stays on top of the biggest developments in the space. 

“We wanted a place to take the questions that come up in the AI space because we don't have all the answers,” said Bullough. “We wanted a diverse group of people from inside the company to answer questions like ‘How do we talk about AI outside the company? What about inside? How do we use AI responsibly ourselves? What models do we use? And how do we ensure that those models are diverse and don’t have extreme biases?’”

Check out the full conversation via YouTube below. Keep reading to get a quick synopsis of the biggest insights from the discussion.

Launching the AI Ethics Committee and Its Discussions

AI ethics committees/councils/panels/etc. are great ways to get more collective input on how to navigate the complexities of AI using the brainpower of a varied roster of professionals. Jasper’s ethics committee consists of internal volunteers from the people team, marketing, security, customer success, and general counsel (legal).

Anderson said everyone signed up based on their curiosity about AI and a desire to understand and influence the company's strategy. Discussions were a little loose at first but eventually, everyone’s role got more defined and conversations grew more structured. What was most important in this stage was that the group aligned on a purpose and meetings were operationalized around it. 

Disagreements are bound to happen with any collection of unique viewpoints, especially around complex topics of great consequence. Bullough said that there are multiple correct approaches to any challenge. Agreeing on a best practice with little to no conflict requires first taking everyone’s opinion into account. From there, you can find the approach that has the biggest impact, serves the most people, and creates a system to be proud of.

Anderson emphasized the need for discussions to be as objective and impartial as possible. She shared an example of a case the group reviewed involving hate speech where details about the customer's political stance were removed to ensure a fair assessment. She mentioned the challenge of making judgment calls and the need for collaboration to determine the best decision-making approach.

“The whole point of ethics is it's complicated,” said Anderson. “If it were easy, you wouldn't need to have these hard debates and discourse around it. But you can welcome the disagreements and find a structured way to work through them to get to a reliable decision-making process.”

Staying Well-Informed

Host Samyutha Reddy asked the two leaders how they stay informed and prepared for discussions, especially given the endless amount of AI research and discussion going on at the moment. Anderson mentioned that sharing articles from trusted sources and unpacking them within the ethics committee has been effective. However, she believes that it’s crucial to increase AI literacy and knowledge across the entire company in the same way.

Bullough and his security team discovered a game Gandalf from Lakera where players use prompt injection to get an AI bot to reveal a password. It’s a gamified way to explore how prompt poisoning, a key concern in generative AI, works. Jasper’s ethics council looked at the example to question how well the platform is protected from such threats.

“As this space evolves, this is how that evolution takes place,” Bullough said.  “Somebody raises a concern and we get proof of concept systems that will demonstrate that concern. Then we put in place protections and controls to not only protect a system like Jasper, but also help users make the right choices and protect themselves.”

Speaking of customers, their feedback is very important to the committee. Customers bring up great questions around responsible deployment, potential liability for inaccurate content, plagiarism and other concerns as they adopt AI across their teams. This gives Jasper’s ethics committee plenty of worthwhile challenges to discuss, understand, and address. Doing so allows the council to not only improve Jasper’s functionality but offer guidance to users as they invest in their own AI strategies (like releasing a template for AI transparency statements and defining standards for AI use). 

What’s Next for AI Ethics Committees?

To round out the interview, Reddy asked about the future of AI counsel and its potential evolution as the market, industry, and technology mature. Anderson responded by acknowledging the importance of ethics, not just in AI but in companies that could have a significant societal impact overall. She mentioned the need for ethicists on staff at companies or on their board of directors to help businesses think through ethical standards and practices. 

Meghan shared that she recently connected with Olivia Gambelin who runs a consultancy called Ethical Intelligence where she assisted businesses in making ethical decisions and provided frameworks for addressing ethical questions. Professionals like Gambelin might be in greater demand as AI continues to proliferate the business world and society. 

“Create an operating model for dealing with these questions, whether it's internally or you're bringing in ethicists from the outside who can help,” said Anderson. “I absolutely think that's going to be an ever-increasing role and probably should have been a role many years ago, even before AI.”

There’s even more to the conversation that I couldn’t include here, like Bullough and Anderson’s thoughts on how to demystify AI so it’s less of a “black box” and how they see mitigating biases in AI outputs. Be sure to listen to/watch the full episode to learn even more!