[Behind the Bot] Chronicles of a Protection Officer
Hello everyone and welcome to the first installation of our “Behind the Bot” series!
In this series, we’ll be sharing insights directly from the notes that are taken on a daily basis by the Signpost team that is both de-risking the AI and also checking it for quality.
As we all know, the stakes are really high - Humanitarian needs are increasing, resources for aid aren’t keeping pace, and we have the promise of new technology that will either help us - or maybe complicate things even further.
This journal is about the derisking of Signpost’s first AI “agent” - a chatbot, more specifically, a chatbot that will surface answers drawing from the knowledge base of articles written by Signpost teams. In the ensemble of the 30 Signpost projects around the world, there is somewhere around 30,000 articles written on a basis of the self-expressed information needs of the users of Signpost. This creates a huge knowledge base to train AI from. While this sounds quite straightforward, many considerations make this job quite sensitive. Clients /users of Signpost are often in extremely vulnerable situations, they are facing huge problems and big decisions, and there is nearly zero margin of error when it comes to the performance of this chatbot agent, were it to be interacting directly with a person asking for help.
The Protection Officers are the ones that are doing this critical work - evaluating the quality of responses against a number of different angles, closely collaborating with the Red Team (Red Team tests AI by pretending to be attackers to find flaws and make it safer), and communicating with the engineering team to make changes to the AI agent set up to improve the bot to the point where it 1) does not do harm and 2) helps every time, as best as it possibly can.
The Role of Protection Expertise in Enhancing Humanitarian AI
Protection officers (PO) play a vital role in training AI systems to effectively support individuals in crisis:
Informing Immediate Relief: PO insights guide the development of AI chatbots to offer real-time, accurate, and compassionate support to those in distress.
Scaling Services: PO understanding of logistical challenges ensures AI systems can efficiently deliver crucial information and support to everyone in need.
Bridging the Linguistic Divide: PO experience navigating language barriers enables the creation of multilingual AI chatbots that communicate vital information in diverse communities' native languages.
Augmenting Protection Officers' Capabilities: PO expertise is essential in training AI chatbots to handle routine inquiries, allowing POs to focus on complex cases and provide more in-depth, personalized support.
By integrating the knowledge and experience of protection officers into AI development, humanitarian aid can be more efficient, empathetic, and accessible to all those who need it, while ensuring the human touch remains central to the protection response.
Welcome to Signpost AI
We’re not just theorizing about AI's potential in the humanitarian sector, we’re rigorously testing it and building it. Step inside our reality, where we’re chronicling the progress, challenges, and breakthroughs in our blog series, “Behind the Bot | Protection Officer’s Diary”. Labeled [BtB | PO Diary] in our weekly updates, the series will:
Pull back the curtain on our weekly testing iterations, sharing the successes, the bugs, and the evolving algorithms that shape our AI.
Analyze real-world conversations between our team, country team moderators, and our chatbots, demonstrating the power of natural language processing and machine learning.
Examine the ethical considerations of AI deployment in humanitarian contexts, assessing potential risks and trade-offs to ensure responsible and ethical use.
Share our experiences and lessons learned to contribute to the broader conversation about the appropriate and effective use of AI in humanitarian response.
The Vision: AI That Empowers and Protects
Our ambition is to create AI chatbots that replicate the most effective and empathetic human responses possible. Chatbots that:
Provide accurate, reliable information that empowers individuals to make informed decisions.
Offer personalized support that acknowledges the unique challenges faced by each user.
Navigate complex legal and logistical issues with ease, providing clear guidance and resources.
Continuously learn and adapt, evolving to meet the ever-changing needs of the populations we serve.
We recognize the importance of replicating empathy and compassion in our AI interactions, and we are actively exploring how this will be perceived by users who are aware they are interacting with a chatbot. Protection expertise is helping us reimagine AI in humanitarian response, and you have a seat at the table.
Join us on this journey of innovation, share your thoughts and perspectives, and help us build a future where technology and compassion work hand-in-hand to protect and empower the most vulnerable.
Stay tuned for our next post, where we'll reveal the initial findings of our AI testing initiatives.
