How KQED is enriching its ‘Forum’ archive with generative AI

More

Mike Janssen, using DALL-E 3

This piece first appeared on Medium and is republished here with the author’s permission.

Here at KQED, our newsroom has been thinking deeply about how to integrate and balance generative AI into our workflows and toolkits. We’ve worked on a number of early-stage projects, including:

At KQED, part of my job is to drive these efforts and think intentionally about how KQED can work with AI. A working group of senior editors, KQED leadership, and others meet regularly to build on these initiatives and explore holistic thinking about AI with the newsroom here at KQED.

One especially exciting project we’ve been working on is Ask Forum, a project connected to Forum, KQED’s flagship call-in radio program,

Forum on Discord

For more than three decades, Forum has distinguished itself as a space for conversations about all topics and different viewpoints, reminding us how special conversation can be and why we want to connect to one another. Less than a year ago, KQED launched a new online community for Forum on Discord.

screen shot of KQED Forum on Discord
KQED Forum on Discord

Discord, while traditionally a space for online gamers, has recently been embraced by a variety of organizations as a platform to host online communities. The shared emphasis on community in conversation made the platform a natural fit for experimentation, particularly as it relates to AI, with Forum’s content.

Discord helps create online spaces and support for communities by providing digital tools to organize and communicate with users. The platform enables real-time discussions, multimedia sharing, and live broadcasting. Even when Discord lacks certain native features, its capabilities can be extended by integrating third-party applications.

Enhancing the Discord ‘Forum’ experience with Ask Forum

We created our own AI-powered assistant called Ask Forum to help users easily find and engage with archived episodes of Forum on Discord. Ask Forum is regularly updated, and in keeping with the lively pace of our community’s exchanges, we make frequent changes to the Forum Discord server to better suit this dynamic group.

AI experiments and learnings from Ask Forum

I spent this past year working more closely with the Forum team, increasingly inspired to hear the power of conversations. Working with such an incredible team, I wondered how AI could work alongside this group to help share these stories, and when does AI do more harm than good?

A series of AI experiments begins to answer some of these questions.

KQED’s goals are to inform, inspire, and involve our community. When developing with AI, it’s essential for us to hold new projects to these same ideals.

Also, KQED is committed to informing users about how the AI technology was involved in the process whenever using AI to create or display content. Plus, if the AI system modifies a user’s search terms to find a match, it will openly communicate this to the user. By being transparent about when and how AI is involved, KQED aims to help users understand the role of AI and build trust in both the system and the community.

An example of AI transparency

Ask Forum harnesses the power of a large language model (LLM) to offer users a more flexible and expandable search experience to find archived Forum episodes.

In addition to traditional search methods, Ask Forum supports natural language, allowing users to ask for content using their own words. For example: “What was today’s show?”, “When was Rachel Maddow on?” Are there any “shows about gender?”, and any “shows about camping?”.

Search result examples (Ask Forum)

While natural language search is not entirely new, AI-powered bots also allow us to explore other types of functionality to best serve our audience. For select interviews with available transcripts, Ask Forum enables users to ask questions about a past interview. By using the transcript as its primary source, Ask Forum provides responses that are specific to the interview content, ensuring that the responses provided are accurate and directly relevant to the content discussed in that particular interview.

By focusing on the transcript, Ask Forum minimizes the occurrence of “hallucinations,” which are instances where the AI might introduce information from other sources or mention guests and topics that were not actually part of the interview in question.

Example of using Ask Forum to dig deeper into a Forum episode transcript

Ask Forum is in constant development, evolving and adapting not only to the new content added daily but also based on valuable feedback from users who share ideas on its future direction.

Transparency and AI moving forward

As we continue carefully refining and expanding the Ask Forum experience, we look forward to learning from our online community. This is new territory, which is why transparency, responsible experimentation, and openness to input will be key to getting it right.

Exploring AI and generative AI (genAI) presents a space filled with potential pitfalls. However, we feel compelled to explore this domain to better understand how we might serve our audiences more effectively, both now and in the future. As we create new modes of content discovery, we are excited to have the help and support of all our members who believe in the value of public media across various platforms.

By responsibly navigating the challenges and opportunities associated with AI and genAI, we aim to uncover innovative ways to enhance our services, engage our members, and continue fulfilling our mission in an ever-evolving digital landscape.

Lowell Robinson is senior producer of voice and AI at KQED in San Francisco.

Leave a Reply

Your email address will not be published. Required fields are marked *