MANILA, Philippines – Rai, Rappler’s AI-enabled chatbot that was initially introduced as a chatbot for focus group discussions (FGD), is gaining new skills. 

You can now ask Rai questions about people, places, events, and issues. Rai will limit its answers to trusted data sources and curated facts.

In 2023, Rappler introduced Rai as an  FGD moderator, allowing it to gather inputs from users, synthesize discussions, and ask questions. From these, Rai crafted suggestions for rules and policy ideas, including the governance of AI, and car-free policies.

Rai has grown since then. It’s now an AI-powered chatbot that made its debut at the Social Sood Summit 2024 on October 19.

But what makes Rai a cut above other chatbots?

Rai gathers information from the Rappler website, getting the latest articles every 15 minutes. This is unlike other chatbots whose data sources include random websites whose content are not necessarily vetted.

This makes Rai the most up-to-date and reliable chatbot when it comes to news that matters to Filipinos and other citizens interested in the Philippines and the region. This also gives Rai a unique Filipino and Asian perspective when it comes to world events.

Guardrails in place

What motivated Rappler to develop Rai? This is a fair question considering misgivings over generative AI technologies and their tendency to hallucinate.

“AI hallucinations” refer to instances when these chatbots make up false information while responding to questions. While AI companies have been hyping the capabilities of their bots at general intelligence, bots would notoriously blurt out responses from time to time that could be out of this world or totally made up.

Gemma Mendoza, head for Digital Services at Rappler explained Rai’s purpose: “For more than a decade, Rappler has produced stories and research around various issues that our team has covered and verified. We now want to make these more accessible to readers, allowing them to keep up with current issues and get the information they need — when they need it.”

Must Read

[ANALYSIS] The future of news

AI-powered chatbots have this capacity. But to mitigate risks and minimize hallucinations, Rappler developed Rai with guardrails in place.

“With Rai, we’re not just launching another AI tool, we’re defining how newsrooms can harness data and AI to build public trust,” said Rappler Data and Innovations head Don Kevin Hapal. “By anchoring Rai in the integrity of Rappler’s journalistic rigor, we’re showing that AI can be responsibly integrated to support truth in the face of disinformation.”

Designed to be an extension of the way Rappler’s multi-awarded team digs for the truth, assesses facts, and debunks falsehoods, Rai’s data source is the stories and the datasets that have been gathered, processed, and vetted by Rappler. 

“In all of our investigative and in-depth stories, as well as daily news stories, we have placed importance on primary sources, be that data, documents or interviews,” explained Chay Hofileña, Investigative Editor and Head of Training at Rappler.

“Ethical standards guide our decision-making whenever we are confronted with dilemmas that concern stories and sources,” Hofileña added. “We have learned from mistakes along the way and maintain a corrections policy, all aligned with Rappler’s value for accuracy and truth-telling in our reporting.”

By only using information that the Rappler newsroom has verified over the years, Rappler is essentially making sure that Rai is consistent with the organization’s editorial and ethical standards.

The bot is designed to provide source articles and links for responses it generates. This makes tracing and correcting the source of errors in responses easier.

Beyond this, Rappler’s tech team equipped the new conversational Rai with a powerful architecture that allows readers make the most of generative AI in mining Rappler’s wealth of information — while minimizing the risk of hallucinations that bots are prone to.

Among other things, Rai makes use of a technique called Retrieval Augmented Generation, a method for enhancing the accuracy and reliability of an AI model by grounding it using an external source of knowledge or data.

This was made possible by various fundamental technical development work that Rappler’s tech team has rolled out over the years. Before developing Rai, Rappler re-engineered its website in 2020 to adopt a more scalable, topics-based way to organize content.

Ontology, knowledge graph

Ahead of the 2022 elections, Rappler also developed its own ontology and knowledge graph.

Ontologies are ways to structurally describe a subject matter in a way that machines will eventually understand. What is so powerful about Ontologies is that they make it possible to establish relationships between specific concepts and data points. This includes relationships between people, organizations, places, and other themes and topics. The BBC is one media organization that organizes its content using ontologies.

Rappler’s ontology and knowledge graph house link Rappler’s stories with data about people, places, events, and other key concepts in topics and themes that the newsroom covers. Even before OpenAI’s public release of its chatbot, Rappler used generative AI to create profile pages of almost 50,000 of local candidates in the 2022 elections.

Rappler used the same to roll out  its Topics Directory, a way for Rappler readers to find information on specific people, places, organizations, and other subject matters that figured in the news.

Ahead of Rai’s release, Rappler cautiously rolled out AI features on its platform. In June 2023, AI-generated summaries were made available to Rappler+ members. Alongside the launch of Rappler Communities, Rappler also introduced Newsbot and Gamesbot to gamify audience engagement in news.

More than just ChatGPT

The team has put in place a number of other guardrails to ensure — in the best possible way — that Rai behaves a certain way. This is apart from constraints in its design that limits its data sources to trusted and curated facts.  

This combined human-and-machine approach to using AI is meant to make sure that Rappler is able to maximize the use of AI while minimizing risks.

Most noteworthy of all, while Rai uses the language processing powers of existing large language models such as OpenAI’s GPT4, Google’s Gemini, etc., it is designed to be LLM-agnostic. This means Rai can make use or combine the use of the best models available in the market.

Here are current use cases for it: 

• Fighting disinformation by always providing sources for its answer
• Sifting through difficult sources and complex datasets, helping journalists (and the public) find the root of the story
• Civic engagement and promoting civic discussion
• Ending obfuscation in bureaucratic processes, where people try to “hide” anomalies in paperwork 
• Understanding your world
• Teaching students difficult topics
• Talking in different languages and helping people find a common ground: ask Rai a question in any supported language to see how it works! Cebuano, Japanese, it can do it all
• Show you how people think and creating a place where it feels safe to ask questions in the open
• Understanding and contextualizing the news in a flash

Here’s an ask: let’s power and grow Rai together.

Must Read

Rappler launches Rappler Communities: Here’s what you need to know

And what if Rai makes a mistake? As in everything Rappler does, Rai is also covered by Rappler’s corrections policy. Users may report errors to feedback@rappler.com. A team will assess to find the cause of the mistake. If necessary, corrections will be made on the source data.

Rai is also governed by the AI guidelines of Rappler. As the first newsroom in the Philippines to adopt and publish rules in newsroom use of AI, Rappler’s guidelines puts a premium on the supremacy of human critical thinking and judgment, and accountability.

Rai will be available exclusively to users of Rappler’s Communities app. Members of Rappler+, Rappler’s premium membership program, will be given early access to new features beginning Monday, November 4.

Be the first to try Rai by becoming a Rappler+ member today. Visit rappler.com/plus for details. — Rappler.com