AI for Human Rights

Developing and applying inclusive and responsible AI to safeguard human rights and address humanitarian challenges.

Introduction

Humanitarian needs are growing with the number of displaced, exploited, marginalised and vulnerable people being higher than it ever was before, and resulting in human rights abuses.

  • Approximately 79.5 million people were displaced in the world in 2019 – nearly 1% of humanity. 
  • Over a billion people are without legal identity.
  • It has been estimated that over 40 million people are in modern day slavery – a human rights abuse – at any given time. 
  • Nearly half of human trafficking victims are women, and 35% of women globally have experienced physical or sexual violence at some point in their lifetime. 
  • Women and girls, rural dwellers, ethnic minorities, people with disabilities, migrants and refugees, and the LGBTQ community are systematically excluded and denied human rights, and deprived of opportunities and benefits of societies and economies.
  • Individual, local, community and regional socio-economic or political instability, coupled with natural disasters and public health emergencies increases those already in precarious situations to further harm and marginalisation, and exposes a new vulnerable group to harm and marginalisation.

Data and data science/AI methods have the potential to contribute to approaches and solutions, and to make critical step changes in the advancement needed to tackle societies greatest humanitarian challenges, thereby addressing the SDGs rooted in a human rights approach and serving in the administration of systems and the Rule of Law. To realise the utility and power of data and AI, responsible and inclusive practices must be developed and practiced.

Our vision

We envision a world that is more safe, fair, just and inclusive for all.

We believe that new forms of data and data flows, new algorithms and multi-sector and multi-disciplinary collaborations are the key to this future.

Theory of change

Through co-design and development of inclusive and responsible data and AI methods, tools, frameworks and practices, data and data flows can be unlocked and harnessed whilst preserving privacy, to inform new insights and opportunities for services and interventions for prevention, preparedness and protection against harm; accelerate prosperity and inclusivity for all; improve the resilience of institutions, systems and communities, and serve in broader human services, systems and the Rule of Law.

Our mission

We will empower government (public) and non-profit sectors by co-designing, developing and deploying new inclusive responsible data and AI methods, tools, and frameworks working in partnership across sectors to build new pathways for innovation and change with an aim to safeguard human rights and address humanitarian challenges.

Challenges

 

Related projects

Below are the projects related to this theme, categorised by their individual goal.

Goal: Safeguard people and communities from harm

Identifying national drivers of a hidden phenomenon
How can data-driven approaches shed light on the complexities of national prevalence of modern slavery?

A number of the UN sustainable development goals (SDGs) attempt to measure and quantify subtle and intricate social issues, such as gender equality and exploitation. The adoption of SDGs has encouraged the quantification of complex social issues which are challenging to measure directly, particularly if the problem is largely hidden, as with modern-day slavery. Due to the unique hidden, noisy and disparate nature of modern-slavery data, this work generates hybrid models around a principal of combining theory from the field, informative prior statistical assumptions along with data-driven machine learning techniques. Data is scraped from multiple open sources including the World Bank, Varieties of Democracy Project, The Woman Stats Project, and the Global Slavery Index. This project aims to apply machine learning methods to better predict and explain complex factors that make a particular country or geographic region and its people vulnerable to modern slavery practices. It aims to do this in the context of small n and large p, and studies the causal non-linear structures, predictive thresholds, and dependencies between the variables predicting slavery.

COVID-19 and its impact on the most exploited
How has COVID-19 impacted Modern Day Slavery (MDS) with respect to the scale and nature of exploitation?

COVID-19 has caused disruption to most facets of the economy, and it is expected that it is no different for forced and exploitative labour. Already a hidden population, both NGOs and government departments are trying to understand where these modern day slaves have gone, and what work/industries they may have been transitioned to. This work applies a mixed method approach to examining data capture at the population/aggregate level – open publicly available scraped data, and the small scale individual – interviews. The interviews inform and help to augment both population data capture sources and inferences.

Measuring and capturing hidden populations
How can we measure and capture hidden populations?

From the typologies of modern slavery (including domestic servitude. labour exploitation at sea. sexual exploitation online) to domestic violence to terrorism, vulnerable and exploited people are often hidden in plain sight. This work explores (i) the utility of existing methods to capture hidden populations including multiple systems estimation and survey methods, and (ii) development of hybrid or hierarchical methods that makes (a) the use of unstructured data and proxies, and (b) the role of privacy enhancing technologies (PETs) to open data access to inform and improve data gaps.

Modelling pathways to exploitation
How can we harness individual narratives to inform interventions and policies?

Individuals provide narratives of the accounts that led to exploitation or conviction, an adverse outcome. These narratives require natural language understanding to identify the key events that led to the adverse outcome and potential opportunities to intervene for other vulnerable people. This work

aims to couple the use of natural language processing with chain event graph (probabilistic graphical model) technologies to develop a new tool to create a new data source and approach to inform interventions and policy opportunities.

Mapping and estimating vulnerability to exploitation
How can we unstructured data augmented by individual household data inform interventions and policies to safeguard against vulnerabilities?

Many people across the world are living in vulnerable circumstances because of high poverty, isolation, or weak infrastructure. Collecting data on these people to better understand their needs, and to help protect them from exploitation, is a challenge. This project combines survey responses with digital footprint data such as mobile phone and mobile money records to map and predict such vulnerability, allowing for (i) the identification of communities which are most at risk of being exploited in cities and hyper-local areas; (ii) a more connected landscape of vulnerabilities and where and when a particular intervention might be required, helping to ensure aid is getting to those most in need, (iii) the study of how these geographic risk factors interact with individual level data, and (iv) the investigation of the role of digital data sources as proxies for different types of vulnerability, consequently alleviating the resource expensive task of directly collecting data through surveys.

Goal: Build resilient institutions, systems & communities

Leveraging data to enable investor action in relation to modern slavery risks
What is the potential of data-driven approaches to provide investors with actionable information about modern slavery risks?

There is increasing recognition of the important role that investors can play in addressing modern slavery risks that exist within the private sector. Yet, investors are faced with the combined challenges of (i) scarce actionable information on investee companies’ exposure to modern slavery risks and (ii) information being presented in ways that are difficult to integrate with established and scalable risk management processes. This project addresses these challenges by delivering a landscape analysis that maps the role of data in enabling investors to integrate modern slavery risks into both selection decisions and active ownership strategies.

Participatory AI for tackling sexual exploitation online
How can we measure the scale, identify the nature, and disrupt risk of sexual exploitation online?

Of the 24.9 million forced labour victims estimated in 2016, 4.8 million were estimated victims of forced sexual exploitation. Women and girls account for 99% of victims in the commercial sex industry. Rising smartphone ownership and cheap, fast internet use has taken slavery into the modern digital age where there are potentially more risks for exploitation, more victims, more perpetrators, and customers, and where traditional law enforcement and intelligence techniques and capabilities are not as strong as they are offline. With the internet promoting the availability of sex and sex work in the UK, there is up to 100 “adult services” websites carrying adverts. Whilst advertisements on ASWs may be legitimate sex workers operating legally, there is a risk that some ads may appear legitimate in nature but may be managed by another individual or organised crime groups who are exploiting victims of modern slavery through organised prostitution. This work aims to detect and distinguish between a signal of risk of coerced versus non-coerced activity through features in the language and images of advertisements to measure the scale and nature of exploitation. Lack of ground truth ads to the complexity of the challenge, and hence, we use proxies in the machine learning approach. This project aims to develop methods, tools and best practice principles in the co-design and development of data science and AI techniques to help policy makers and frontline non-profit and law enforcement agencies to tackle sexual exploitation online.

Data science/AI for tackling forced labour at sea
How can we capture and identify risk of hidden and complex human rights abuses and human security challenges at sea?

Capture fishing is a labour intensive global multi-billion US$ business in often remote, isolated and physically dangerous environments. Over the past decade NGOs, media and law enforcement agencies have uncovered the widespread existence of labour exploitation, forced labour, human trafficking and slavery in this industry. The identification and pursuit of these abuses is complicated by several factors including the remoteness of locations and ease at which offenders can hide or disguise their activities. However an increased understanding of the economic and environmental causes and drivers of vulnerability combined with increasing observational ability through remote sensing technology offers opportunities to use explainable Bayesian statistical techniques, informed with domain knowledge, qualitative and contextual information from cases and survivor narratives, and quantitative remote sensing data from a variety of sources on vessel movements and communications, to build predictive models on vessels and individual actors. This project aims to develop statistical models and AI technology to support policy makers and front-line agencies to identify and combat these human rights abuses and human security challenges.

Statistics and the law: probabilistic modelling of forensic evidence
How probabilistic reasoning assist in serving in the administration of justice?

Forensic science plays an increasingly important role in contributing to national security resilience and the successful administration of the justice system. Major ongoing technological innovation has propelled forensic science development and sourced new types of forensic evidence. Such technological and scientific advances must be matched by well-developed tools for the analysis, evaluation and interpretation of evidence in order to successfully exploit innovations in the administration of justice. This project draws upon real cases and engages with the forensic science, legal and policing community to develop a probabilistic framework for the evaluation of complex forensic evidence, that deals with the multiple statistical issues and complex data structures that can occur. There is particular interest in understanding activity level propositions – that is relating the evidence to the activity in question and not just source level attribution, using graphical models.

Enabling trust, security and privacy for policy innovation in tackling modern slavery

Modern slavery is a policy area which often gives rise to a needed cooperation amongst government agencies, non-profit and for-profit institutions to share data, tools and resources to effectively deliver services, inform and enact policies and practices, enable change to safeguard people and institutions, and responds to emerging issues. Despite gallant efforts to create platforms to facilitate data sharing and the increased appetite for data trusts, barriers to data sharing can be described by twinning concerns for (i) data underuse and (ii) data overuse or misuse, giving rise to personal rights, informed consent, and myriad of data privacy, security and trust issues. This project involves a multi-disciplinary team (consisting of members of members of The Alan Turing Institute, the Bonavero Institute, and the Open Data Institute) that will conduct structured and unstructured interviews and round table discussions with members of the UK anti-slavery stakeholder, data and algorithmic user, and data steward/provider community to (i) understand the barriers and gates to trust, security and privacy that would enable data and algorithmic policy innovation for modern slavery and (ii) the necessary first steps to (a) realise the value in data and algorithmic approaches when legal or cultural threshold bar progress, (b) identify personal rights and jurisdictional challenges, particularly in light of UK’s changing status in relation to the EU, and (c) develop and highlight new insights that can inform PET and law research that highlights data and algorithmic governance, and open opportunity for research, policy and international collaboration.

Goal: Accelerate opportunity to build inclusive societies & economies

AI, equality and human rights
How can AI better serve and benefit all members and all sectors of the world?

With a greater need for humanitarian response and the opportunity that data and AI presents in contributing to solutions that can improve responses, this work explores and identifies methods, opportunities and practices (through focused projects) to enable AI development for the benefit of all at an individual, sector, and global level. In light of accelerated AI responses to tackle COVID-19, this work considers data representativeness, opportunities for participation and consent, and access and availability for the social, health and public safety benefits of AI technologies.

Inclusive and responsible data and AI for justice and humanitarian innovation
How can inclusive principles drive responsible AI innovation?

With the excitement of data science/AI capability, this work develops a framework with a set of inclusive data and AI principles and practices from data representativeness to co-design to fairness and access as a way to drive responsible AI innovation (incl. explainability, robustness, etc.) from design to development to sustained deployment. It also considers the Rule of Law, principles of proportionality and necessity, and data readiness, consent and retention as a basis and bounds to consider in the design, development and deployment of AI to respond to both the status quo of services to safeguard vulnerable people and emergent crises local (39 Vietnamese migrants) or international (global migration) levels and public health emergencies (e.g. COVID-19). This project is developing an inclusive AI practice through the participator AI for tackling sexual exploitation online.

People 

Organisers: Anjali Mazumder, Florian Ostmann, Mark Briers

Co-Is, Turing Fellows, Researchers and Collaborators: Jim Smith, Weisi Guo, Rosa Lavelle-Hill, Oliver Bunnin, Ruth King, Jat Singh, Eirini Malliaraki & Ser-Huang Poon

Partnerships

Policy and Evidence Centre for Modern Slavery and Human Rights

Code 8.7

Dr Weisi Guo

Honorary Professor at University of Warwick & Professor of Human Machine Intelligence at Cranfield University

Organisers