‌*The following is the speech delivered by Yuan Lin, Product Manager of Toxics-Free Corps AI Assistant, at the China Foundation Development Forum - AI Empowering the New Public Welfare Ecosystem Parallel Forum on November 23, 2024.

In 2023, we decided to combine the two sections of product investigation for the e-commerce industry and consumer advocacy for the public. At the beginning, partners could not find a clear and specific action point. Should we do live streaming, community operations, or offline workshops and activities? By April 2024, Dr. Mao, the head of the organization, said that after thinking about it for so long, we had to start doing something. At that time, we once again confirmed the original reason why we wanted to do this, which was to hope that friends around us could avoid pitfalls and stay away from the health effects of these toxic and hazardous chemicals. So at that time we used the most basic method to start. The partners of the organization took turns in the product mutual assistance WeChat group to actively collect, respond to, and organize questions raised by group members about product safety.

In June, we found the Tencent SSV Carbon LIVE team. They were actually looking for us, too. It was truly a two-way effort. After several months of hard work, the Toxics-Free Corps AI assistant quickly expanded from an initial 60+ questions to over 700, with over 100,000 words of information. The hit rate of the knowledge base increased from 58% to 75%. Not only has its text responses become increasingly professional, but it has also developed features such as photo recognition of ingredient lists, quality inspection reports, and voice recognition, and has also implemented artificial intelligence to support personalized questions. Going forward, we will continue to promote the collaborative development of the AI knowledge base and product safety corpus.

The past and present of AI assistant

2021-Present

Continue to do scientific popularization and actively answer questions about product safety from friends around us

2023-2024

We shifted from product research to product research + consumer action advocacy. But partners couldn't find a clear and specific action point: live streaming? Community building? Education? Workshops?

June 2024

We contacted the Carbon LIVE AI technical team. It was a two-way effort, and we worked together. The first batch of more than 60 questions and answers formed the initial knowledge base.

April 2024

Back to the original intention: to help friends avoid pitfalls and stay away from the effects of toxic and hazardous chemicals. Start with the most basic method: take turns responding to and editing various questions and answers in Wechat community, and record them in a separate document.

August 2024

The internal test product is released, and the internal test team is established to form a rhythmic working mode. The knowledge base is constantly increasing (700+) and new functions are constantly being developed.

October 2024

Expanding to more communities

Increased knowledge base hit rate from 58% to 75%

New users

Satisfaction rate for question and answer annotation

November 2024

The Toxics-Free Corps AI Assistant has been officially released to the public, enabling features like photo recognition of ingredient lists and quality inspection reports, and voice recognition. We will continue to promote the collaborative development of the AI knowledge base.

During the research and development process, I increasingly felt the enormous potential of AI technology for public welfare. AI technology possesses a strong public nature, impacting every industry, much like water, electricity, and coal. This can easily lead commercial companies, focused on return on investment and competitive barriers, to be wary of AI applications. This, in turn, allows public welfare organizations to embrace innovation with ease. Furthermore, the barrier to entry for applying AI in specific verticals is not as high as many imagine. Initially, it doesn't require significant capital or cost investment, making it relatively accessible to many small, local public welfare organizations. Most importantly, public welfare organizations can leverage AI technology to strengthen knowledge dissemination and public engagement in their specific verticals. This, I believe, is the foundation of AI-based public welfare. For example, during the development of the Toxics-Free Corps AI assistant, we continuously compared the text generated by the assistant with text generated by existing language models for the same questions. Our initial conclusion is that, in the online world, knowledge about product chemicals in non-commercial contexts is underappreciated. Therefore, we remain confident in the assistant because it holds unique public welfare value.

What are the issues that matter most to your organization? Is it environmental protection, rare diseases, or public health? What knowledge and information do you hope more people will understand? We think this is a good place to start thinking about AI development.

Here are some of my experiences:

1. Clarify the institutional agenda. Don't pursue AI simply for the sake of pursuing it, or simply to follow a trend. Starting with the core social issues your organization is focused on will help you identify the type of knowledge base you want to build.

2. Look outward. Try to first consider what knowledge and information you want to output externally and what users don't yet fully understand. When many public welfare partners discussed developing AI knowledge bases with me, they said they hoped to use a robot to improve their own learning and internal knowledge sharing efficiency. What I want to say is that if your initial goal is to output externally, your driving force, the external feedback you receive, and even the support of resources will likely continue to flow, forming a positive cycle. At the same time, you will also achieve improvements in internal knowledge efficiency and learning. Considering the positioning and value of AI assistant products from the outside in is a very important point.

3. Rapid iteration. Just like learning to drive, you don't need to know how a car is built to learn it. The first step is getting in. Thanks to the support of our AI technology partners, public welfare organizations don't need to invest heavily in algorithm or computing power R&D. Whether or not to get in depends more on your understanding, willingness, and commitment than on resource or technical limitations.

4. The most important factor is people. Public welfare and AI involve cross-team and even cross-institutional collaboration, so people are crucial. Our collaboration with the technical team is like finding people who are looking for each other.

During our collaboration, the AI technical team did one thing exceptionally well, and I'm very grateful for it: they spoke human language. They consciously minimized technical jargon and barriers to entry during communication, resulting in a very positive, mutually reinforcing working model that has developed over the past few months. As soon as the technology evolved, our knowledge base expanded quickly, and front-end promotion and internal testing began. Once the front-end had some progress, the technical team immediately began integrating the next round of development requirements and technical iterations. We also have a division of labor and coordination within our team. The two knowledge base operations and maintenance partners are here today, and they've done crucial work on every detail of the knowledge base. In addition, there are our institutional communications team member, and of course, our strategic director, Dr. Mao Da. Including me, there are five project members in total. Besides the AI assistant project, each of us also holds other responsibilities within the organization, making us a multifaceted team. This has honed our ability to react quickly yet methodically.

The social uncertainties brought about by AI technology are gradually becoming apparent. As partners in the public welfare sector, we should actively explore and respond to these challenges to become a meaningful and influential player. Whether it's building a human-centered AI architecture, exploring AI ethical practices, creating AI corpora and algorithmic applications for good, or leveraging technology to create greater well-being and sustainable development for humanity and the planet, we public welfare should not be absent from this AI revolution. The AI assistant product from Toxics-Free Corps is just a small first step. We encourage everyone to work together to develop AI products and applications for public welfare. Please pay close attention and provide us with feedback.