INTERNAL: Announcing the Signpost AI Prototyping Lab

Signpost AI Prototyping Lab!

To start off 2024, Signpost is announcing internally our concept for a Signpost AI Prototyping Lab. This news should not yet be shared outside our network just yet but marks the beginning of an exciting new era of Signpost! In addition to this news article, you may find the attached concept note for the lab here. 

What is the Lab? 

In order to ensure that information is never a barrier for people in crises to exercise their human rights, understand their options, and find critical support, Signpost will create an AI prototyping lab to de-risk and evaluate the effectiveness of generative AI so that we can scale community-driven information programming to a truly global scale. Through the Signpost Project’s global footprint and consortium partners, we will prototype promising solutions and scale successful models globally. With this initiative, we will get closer to Signpost’s goal of providing information to 50% of the world's displaced population by 2027.

Signpost proposes the creation of a Humanitarian AI Prototyping Lab. Based on principles developed by IRC’s Airbel Innovation and the technology backbone scaled across the Signpost Umbrella. 

This lab will be dedicated to building “public good” prototypes of humanitarian tools that repurpose and reprogram open source and private sector technologies for humanitarian use cases and, when necessary, build new foundations of a more ethical digital landscape. The Signpost AI lab will be grounded in practice, prioritizing new products based on needs coming directly from people about the quality of their services and from responders about the hiccups in their responses. We will build prototypes across our consortium (IRC, Mercy Corps, Internews, and local partners) depending on their demand. 

What will the lab do?

• Create Partnerships: We will form an open, global consortium partnership with aid organizations, governments, tech companies, research institutions, and news media, casting a big tent to jointly research and develop AI solutions. 
• Create Prototypes: We will develop, deploy, and scale (within Signpost) cutting-edge AI tools to address humanitarian challenges and enhance aid.
• Evaluate Impact: We will run practical impact research to evaluate the quality of AI-generated information against human information and relative returns on investment.
• Disseminate Research: We will regularly disseminate practical toolkits and research to provide humanitarian workers with AI development principles and guidelines for building on their own.
• Advocacy: We will begin direct communication with and regular public advocacy with technology companies about how to make AI technology support humanitarian clients better and be more just. We are starting this process with Signpost's engagement in Meta's Open Loop program as experts to help formulate better risk management frameworks for AI.

Why is Signpost doing this? 

It is essential to our mission to provide actionable humanitarian information to as many people as we can, advances in AI present huge opportunities, but also enormous risks. 

1. Harm Reduction: 

Whether by our own interests, that of funders, or pressures from our sector we will all be asked to utilize AI in the coming years. It is clear that AI is not going away and that its momentum will soon affect our work. Inappropriate use of AI has harmed people, and will continue to do so. Many organizations are evaluating the game-changing and existential nature of AI. “Harnessing the power of AI for good will require more than focus on existential and potential damage. It will demand a positive vision of what AI can do and effective measures to turn that vision into reality.” (Kent Walker, President of Global Affairs at Google)

Through this initiative, we need to do massive-scale harm reduction - not only build access to the right information, but we also need to obliterate half-baked, harmful information models. Chatbots of this sort are already being sold to humanitarian organizations by profit-incentivized organizations with disregard for their impact. It must be noted that to ensure do-no-harm in an aid sector context, extensive de-risking is required from both a technical and programmatic perspective and deep knowledge in protection and humanitarian work is a must. While some vendors will have strong technical skills, the human element may not be understood - i.e. how will humans work with this technology in a way that is safe and ultimately more beneficial.

2. Meeting the Information Needs of our World: 

With over 100 million people displaced worldwide, a stark increase in armed conflict, growing incidence of climate-related disasters, and economic hardship, more and more people are forced to make life-changing decisions under duress with limited to no information about their options. Signpost, the largest information project in the aid sector, only serves an estimated 3% of the global population of displaced people. Aid organizations are overwhelmed by the flood of information requests present in every crisis. With projections that over 1.2 billion people will be displaced by 2050 due to climate change, our 3-year target is to reach 50% of the world’s displaced people so we can get ahead of the staggering need. Therefore, we need solutions that stretch our limited resources further. AI is an enabler that can bring better information faster and more scalable than ever before with a business model that can fundamentally change how the sector delivers to clients and mainstream information as aid.

3. We are the best positioned

Reputation: Signpost is a UN SDG Action Award-winning consortium that is well-networked in the aid sector and research institutions and is positioned to scale technology through these networks.

Vessel for scale: Through the Signpost Project’s global footprint (35 countries and over 750,000 clients per month), we will pilot promising solutions and scale successful models globally. Signpost is the only program at scale, reaching 78 million people with 12 million all-time users in over 70 active social channels, with 400+ staff experts providing information to vulnerable people in crisis. Through each instance of Signpost, we can control tests and include a diversity of contexts to pilot and scale prototypes.

Local partnerships: Signpost projects have mapped and created relationships with thousands of local NGOs in every country we work. Each of these projects is a possible “customer” for successful AI products built by the lab and a huge part of the localization of any public good or open access. 

Better data: Signpost is the custodian of the largest and richest dataset in the world of interactions between humanitarian responders and clients. There are 300,000+ conversations in 30+ languages and 22,110 source language articles (over 500,000 with translations) to train our AI models. Unlike most aid data, this is information directly from clients in their words without asking 20 questions. Signpost has created the largest and richest dataset of client-initiated interactions and responses in the aid sector. 

Data Infrastructure: Signpost already has cloud architecture that houses all of our data after a two-year build process. Our existing systems are flexible and can be modified for building an AI testing environment and creating synthetic data. 

Data Protection: Signpost has proven protocols for data protection and inclusivity, as well as a core database infrastructure developed to be secure and pliable.

How does it impact existing Signpost Programs?

Reputation: AI Lab work will position Signpost further as a sector-wide innovator. Early engagement with AI will future-proof Signpost programs and increase its long-term sustainability for funding, technology approaches, and reputation.

Expanded funder base: AI Lab with different networks will introduce Signpost to new mediums of funding and new networks of potential partners for existing Signpost work. Work to expand our funder network is already beginning and is a priority of Andre, Briana, Liam, and the IPP (Institutional Partnerships and Philanthropy) team at IRC. 

Subsidy and Joint Fundraising for country programs: Country programs across the globe, especially in areas with more neglected affected populations by funders and media, would have the ability to pitch prototype AI technologies to donors when aligned with program objectives, covering program overhead with fewer stipulations and effort than more traditional funds. We hope that our leadership on AI will also make our project more attractive for institutional and government funds to help stabilize country-level and regional fundraising for Signpost programs. We want to use the attractiveness of AI to funders as a means to more evenly and objectively match funds to the places that need it most.

Subsidy for global team: Given that many of the skillsets of the AI Prototyping Lab will be shared with global Signpost staff, funds from the open market of humanitarian AI fundraising will help us support existing staffing with an additional means to raise funds.

Shared IT Services: AI will be a catalyst to bring on IRC organizational resources for collaboration. Signpost will retain levels of support by creating mutual incentive models with existing IRC IT resources. The IRC HQ began an AI Community of Practice in AI which is beginning to open collaboration across the IRC related to AI.

Tech Support Implications: Our processes will not change, however, who does the work on which issue may change slightly. With finite resources the technology team will be diverting some existing development resources to build AI tools. This means our team will have some dedicated resources toward fixing web related issues, but will prioritize the creation of new tools as see fit. Will this increase the timeline for your bug to get fixed? In most cases no, especially if it doesn’t require significant developer time. Only for low impact issues that require significant development will there be any change in our speed and uptake of new bugs.

What outcomes do we want? 

1. Help people faster and cheaper in more places, languages, and contexts

2. Decreased harm from the proliferation of AI in irresponsible ways in the AI Sector

What next? 

January 2024 - Hiring: We are hiring a Project Manager, AI in the United States. The role went live at the end of December 2023!

February 2024 - Google AI Accelerator: The Signpost team is excited to announce that Signpost was accepted into Google’s Generative Artificial Intelligence Accelerator Program to begin the creation of a Signpost AI Prototyping Lab that will de-risk and evaluate the effectiveness of generative AI and build products that deepen the impact of Signpost information programs. This Accelerator will offer dedicated support and access to best-in-class experts on artificial intelligence, data protection, AI ethics, and development from Google for 6 months. We will begin with work on improving quality analysis and with human-in-the-loop two way chat communication. 

Conclusion

For more information about how Signpost is viewing AI please reference Signpost's FAQ on Generative AI. 

The new lab concept is an exciting mission. It can only be successful if it is a partnership between the technology and programming of Signpost. With this in mind, any ideas, worries, or project ideas should be shared to support@signpost-help.zendesk.com and Liam.Nicoll@rescue.org.