No Voice left behind: A Call for Research on AI’s Role in Amplifying the Insights of the Systemically Unheard

By Jessica Mayberry and Lieve Fransen

Summary

The article argues that systematically ignoring and silencing the voices of the poor and marginalized worldwide does not serve society or democracy well and must be countered. While new technologies such as AI could provide an opportunity for change, we contend that these technologies need to account more effectively for their realities. Therefore, we propose a research and policy- oriented action agenda to confront this issue in the near future.

Being systemically unheard remains a problem for the poor and marginalised, despite the explosion of social media. In a democracy, Voice means speaking out on issues that matter. Whether the immediate issue is not using one’s ‘voice’ or lacking the power to use it effectively, the outcome remains the same in both cases: those individuals remain unheard. 

In this article, we distinguish the terms “systemically silenced” or “systemically unheard” from “voicelessness.” As political theorists and sociologists have shown, the real problem is not voicelessness itself but rather the systematic ignoring and silencing of certain groups. (Read here). While these communities have plenty to say, their concerns and experiences are often dismissed or overshadowed—an outcome determined by those in power choosing not to hear.

Through this article, we seek to shift that paradigm. With the rise of AI and Large Language Models (LLMs), we now have an unprecedented opportunity to address these inequalities—if we choose to take it. After all, meaningful change begins not simply by giving others a voice but by refining our collective capacity to listen.

It is commonly believed that the world’s ‘unheard’ populations reside only in Low- and Middle-Income Countries (LMICs), but that is not true.  People not being heard are everywhere in the world and that is not only an access or technology problem. People in LMICs, in the last five years or so (depending on the country), have gained access to technology. And though there has been an increase in posting by citizens, of photos, audio messages or video clips, there has not necessarily been a corresponding increase in digital behaviors that could be considered the exercising of ‘active citizenship.’ 

Platforms like YouTube appear to give everyone a voice, but in reality, the number of creators from  truly marginalized backgrounds has been fairly low. For two years, Video Volunteers – a platform that promotes community voice and enables co-creation with community media content — has been seeking to understand the extent to which, for instance, the homeless were creating content on homelessness, or people with disabilities  on the experience of disability, and found it sorely lacking. The results from this small study seem to suggest that the widely available social platforms are clearly not helpful to change this reality. (Read here)

Data sets produced by marginalised and poor citizens worldwide are more effectively collected through visuals and self-produced videos rather than text. Internet-based video is therefore increasingly recognized as THE tool for listening to the voices of those with low literacy, who are not inclined to write essays. 

Video Volunteers’ and some other NGOs’ work demonstrate that ‘active citizenship’ doesn’t happen automatically – it needs empowerment, nurturing and education, and the high investment into nonprofits that develop active citizenship is evidence of a widespread belief that this is the case. People who are introduced to content creation and social media through training sessions and on-the-ground efforts by NGOs do in fact use these tools in ways that strengthen democracy. 

Video Volunteers’ research hypothesis is that government officials and policymakers would, potentially, be amongst the biggest winners of better untapping the lived knowledge of the marginalized and poor in the following ways:  First, their insights can provide officials with information on the local context surrounding a problem.  Second, community voice provides unique information about how a problem affects people. Third, the insights shared by communities may shed light on systemic problems and issues of discrimination against certain groups.  Fourth, government-facing expressions of community voice can illustrate citizens’ willingness and ability to take action to solve a problem. In short, we believe that community voice is the collective expression of concerns from marginalized communities aimed at influencing decision-making and it can help officials prioritize and solve problems in a manner that is more in line with community needs.  

Tanu Kumar, Professor at Claremont Graduate University, working with Professor Gabrielle Kruks-Wisner at UVA and Video Volunteers, has defined ‘voice’ in the context of governance (Read the full paper here). Their survey of more than 1000 local public servants in one state of India found that officials who access citizens’ voices are more empathic and attentive and have a healthy sense of upward accountability. Read the policy brief here

VV has been observing for years how  communities use their voices and how officials respond, and we’ve found that local officials, when approached in a non-aggressive manner, often appreciate the work of local citizens.

Furthermore, citizen generated data matters for development, and is increasingly sought after and respected by many including the UN. It is now considered a key component of measuring progress towards the Sustainable Development Goals or SDGs. To know more, see Collaborative on Citizen Data and Global Partnership for Sustainable Development Data. More citizen-generated data, and particularly the qualitative data, will enable policy makers to make better informed decisions. They may even hasten, at least to an extent, the accomplishment of the Sustainable Development Goals.

Newer technologies, such as AI, seem promising as they could tap into large data sets and make it easier for the voices of the marginalized and the poor to be heard. A human can hardly listen to the individual stories of 100 people, let alone ‘listen to the stories of a billion people.’ But AI can, and that is why AI has such potential in a democracy. ‘A billion unheard people’ could be ‘heard’ not just on the day  they cast their ballots, but every day they speak up about something that matters for the country. Only AI can ‘listen’ at scale. The question is, can we make it ‘listen’ to the voices of an entire nation, and do so  fairly, safely and with respect for privacy? If the answer is yes, this would clearly benefit poverty eradication and democracies.

Of course, this would only be happening if the data sets are available. There remain significant ‘data gaps’ in achieving the SDGs, and AI algorithms are being trained on data sets  biased towards the English language, and thus towards the life experiences of English speakers. According to a quick google search, only 5% of the world speaks English as a native language. 17% of the world speaks English and 59% of the Internet is in English. It is therefore a challenge that the data sets, which might help  ‘give voice’ to people, hardly exist. In other research-based initiatives, we are tackling the problem of data creation by marginalized communities, an issue involving different sets of stakeholders such as national statistical offices and civil society organizations. We believe progress will be made towards creating those data sets, and so our focus for this research is what needs to happen once the data sets are available, specifically by the technologists who work on AI. 

Therefore, It would be useful and even critical to assess if the AI tools used by entities such as YouTube, Google and Zoom  – and by institutions like the UN –are developed in a way that is equitable to the voices of the marginalized and poor. The data produced by those citizens should be  recognized as coming from people with valuable knowledge and insights based on their lived experience. It will also elevate lived experience as a form of knowledge and move away from defining knowledge solely as theory learned by books. 

The promise of artificial intelligence is indeed great and could mean more automated decision-making at scale, but that also means AI could automate risk at scale.  There is now a timely pressure to figure out how to use this new technology to innovate, increase efficiencies, and generate money or save money, while mainly amplifying the voices of the systemically unheard – a feat never achieved.   

The main concern is: how do we do this and how do we do this safely? Already, many have suffered real damage when algorithms led to discriminatory, privacy-invading, and biased outcomes. When ChatGPT was launched by OpenAI on November 30, 2022, generative AI was thrust into the spotlight.  Some companies had already been working with large language models (LLMs), which specialize in processing and generating text.  As far as we know, there has been slower progress in processing images as well as videos, particularly, raw or documentary-style videos where people are speaking a non-European language. ChatGPT, and subsequently Microsoft’s Bing, and Google’s Bard, made generative AI available to everyone within their organizations. Standard AI Ethics include a focus on bias, privacy violations, and ‘black box’ problems.

We believe that responsible development and use of AI will be important for well-functioning democracy, including accessing the voice of the historically silenced  in policy development, decision-making and in creating products and services for all. 

It is time to support the development of a comprehensive and cohesive  framework for AI that facilitates safe innovation, mitigates adverse impacts and – crucially – amplifies the voices of the systemically unheard. Key elements most probably to be included are accountability, transparency and robust risk management processes that look beyond traditional business risks and address privacy, security, non-discrimination, equality and human oversight and control. The development and use of AI systems impact society at large and affect human rights such as privacy, security, personal freedom, and non-discrimination. It can increase the risk of large-scale misinformation, deception, or manipulation.

If we fail, AI could further increase the marginalisation of unheard voices and bias and the society as a whole will lose out because of it. But if we succeed, AI could amplify those voices, making society more responsive to the challenges faced by the poor and marginalized and make democracy more inclusive. And now is the time to get it right. 

We, therefore, believe that building a research collaborative that is focused on this topic of AI and the voice of marginalized communities, and the unheard  and organizing discussions and webinars to bring that community together is critical. We feel that we need urgently  to invest resources for research towards the following areas:

  • Articulating in detail how AI and its tools and processes are related to the poor and marginalized community voices in the world through literature review and questionnaires with stakeholders and main actors in the field. Though this is a global phenomenon, a specific focus on the voice of poor and marginalized communities in the global south is the priority. Some of the challenges already identified should be examined from the lens of marginality and systemic bias, such as hallucination (AI generating plausible but incorrect responses), lack of genuine deliberation (AI simply predicting language rather than reasoning), and the “shared responsibility” problem (where a few companies control most AI development). A feasibility study would define benchmarks for what “sufficiently safe for deployment” means, especially in contexts involving marginalized groups. 
  • A possible methodology for this would be a mapping exercise on tools for democratic participation and poor citizens’ empowerment built with artificial intelligence, with an assessment of bottlenecks and biases. 
  • Laying out a research framework for answering this question: what needs to happen for AI to be able to tap into the wealth of lived knowledge that marginalised communities in the world hold? To ensure the algorithms are not just free of bias but are proactively focused on extracting insight from the often subtle, or complex, or inarticulate utterances of the poor, and not be biased towards the ‘expert’. 

While Large Language Models (LLMs) are being developed, assessing and diving into the world of LLMs and exploring their potential benefits to the poor and understanding the realistic possibilities and the boundaries of what these models can achieve needs to be studied. Learning about the multimodal future of AI, where models will work with diverse prompts, from text to images, sounds, and more and envision a future where AI can holistically analyze data produced by the poor, from records to visual scans and even auditory interactions. To do this, the researchers could use a real-world example – the data archive of the organization Video Volunteers, an organization with the mission of empowering marginalized and poor citizens’ voices. (See Annexure for more information.)

  • Proposing methodologies and processes to help ensure that AI tools could break down the barriers to make systemically silenced voices heard by policy makers, governments, and the UN institutions. It also must deal with what needs to be done to respect the ownership of the data by the producers of the data and also protect people’s privacy. What needs to be done to enable the building of the data sets, such as incentivizing people to create the content? What policies need to be framed to make this possible?  This should be developed by bringing together different actors for discussions and webinars to propose actions and initiatives .

About the Authors

Lieve Fransen is a medical doctor with a PHD in social policy and Public Health. She has extensive peer reviewed research publications and in the last 40 years she is a well-recognized global policy maker.

Jessica Mayberry is the Founding Director of Video Volunteers. She has spent 20 years working to devise models that elevate the voices of marginalized communities in the public sphere.

Annexure 1 – Background on the VV Dataset

In recent years, NGOs have been initiating impressive projects in which the communities they work with create content. They do this to aid the design or the monitoring of development projects and, more generally, in order to make big expensive development programs more participatory. 

Video Volunteers (VV) is one such organization.  It has accumulated an archive of over 18,000 community videos, all produced by people from marginalized backgrounds. As each video was produced, VV meticulously captured a wealth of associated qualitative data, contained in a separate database. This is perhaps the largest such archive anywhere. It is a record of 20 years of a nation in transformation, in which we see and hear rural citizens documenting their development challenges, negotiating with government officials, positing solutions, and organizing their fellow citizens. 

Furthermore, all of it is produced by citizens themselves, capturing their unmediated articulate voice. In a world where most knowledge products related to poverty alleviation and development are produced by people with no direct first-hand experience of living in poverty, this lived experience knowledge is incredibly valuable. 

The VV archive, however unique and impressive, is a good proxy for the immeasurably vaster kind of data that we argue needs to think about more clearly: the explosion of qualitative data that will be produced in ever greater quantities by citizens from marginalized backgrounds, as more and more people use their phones to create content.

The archive can be seen here