Committed to connecting the world

AI Breakthrough Tracks

Track 1: AI + Satellite Imagery

Team Lead: Stuart Russell (Berkeley)
Members: Lars Bromley, Einar Bjørgo (UNOSAT); Miguel Luengo-Oroz, Marguerite Nyhan (UN Global Pulse); Andrew Zolli (Planet), James Crawford (Orbital Insight)

Track Overview
This track will help address how satellite imagery together with artificial intelligence and machine learning can help meet the Sustainable Development Goals (SDGs). The focus of the track will be identifying challenges to large-scale automated analyses of satellite imagery libraries that inform models and knowledge systems on sustainable development. At the same time the track will serve to create partnerships among the artificial intelligence community, satellite imagery providers, research labs, analysts, sustainable development implementers in Member States and the United Nations system, and others. As implementing the SDGs presents enormous policy challenges this track will be ambitious in its scope, discussing artificial intelligence methods for rapid and accurate analysis of satellite imagery that can feed into decision-making processes at national levels. Informing national decision-making for sustainable development requires effective knowledge systems, linking data and information to policy across almost all sectors. Satellite imagery is particularly potent for such purposes as it can potentially measure multiple indicators repeatedly over time and across large areas. Such information allows analysis of underlying issues affecting multiple SDGs and can inform knowledge systems addressing key policy issues. Operational and digital knowledge systems of this type may be relatively far off but the elements involving satellite imagery can already be actively explored by the global research and policy communities. To that end, this track will seek to create a framework for ‘challenges’ whereby specific satellite imagery analytical tasks are posed to the machine learning community to solve. To help focus discussions three crosscutting themes have been chosen:

Poverty: Goal 1 of the SDGs is to “End poverty in all its forms everywhere”, and the track will address ways to concisely map relative poverty for input into relevant models and activities. While most developed countries have rich data collection schemes for tracking wealth much of the world does not. Creating datasets depicting wealth and poverty in many developing countries is difficult and oftentimes haphazard at best. In many of the poorest areas, data is rare, and gathering such data is often time-consuming, expensive and even dangerous. Some countries have not taken a census in decades, birth rates are not recorded, and income is not well understood. Some interesting developments have occurred mapping poverty from satellite imagery and other inputs using AI. However, much more needs to be done to rapidly and accurately map wealth and poverty across all types of environments and biomes around the world, and likewise much needs to be done to better quantify relative concepts like poverty and wealth from such analytical outputs.
Deforestation and reforestation: Goal 15 of the SDGs is to “Protect, restore and promote sustainable use of terrestrial ecosystems, sustainably manage forests, combat desertification, and halt and reverse land degradation and halt biodiversity loss.” Deforestation is a critical problem because forests can provide about one third of the solution to climate change and are critical for species diversity. While clear cutting is the most obvious form of deforestation, forest degradation is an equal or more important threat since as forests slowly lose their density and understory, they become more vulnerable to fire and other threats and much less vibrant in terms of species diversity. Today, we cannot effectively measure the extent of the threat posed by degradation or accurately assess the effectiveness of concessions or payments made to countries or enterprises charged with protecting forests. Similarly for restoration efforts, until the forest reaches a fairly mature state, current methods of remotely measuring tree cover are not very accurate which makes it difficult to measure the effectiveness of restoration efforts. Satellite imagery, along with aerial imagery, and AI can be used for detecting and measuring thinning of upper-story trees, detecting and measuring loss of understory trees and other vegetation, measuring changes in overall water levels, and measuring the growth of new trees in areas being restored.
Agriculture: Goal 2 of the SDGs is to “End hunger, achieve food security and improved nutrition and promote sustainable agriculture.” In many developing countries, local agriculture is the main source of food for rural communities as well as a vital source of income. More than 2 billion people depend on smallholder farms for their livelihoods, and disruptions to local agricultural production can have major adverse impacts on health and livelihoods. The vagaries of the weather is one risk factor, and a growing number of unusual weather events make agriculture even riskier. AI can be used to assess and understand agricultural production to avert famine, maximize market returns for smallholders, and develop tailored insurance policies to mitigate the impacts of crop failures.

Track 2: AI + Health: Artificial Intelligence – a game changer for Universal Health Coverage?

Team Lead: Marcel Salathé (EPFL); Ramesh Krishnamurthy, Senior Advisor, Department of Information, Evidence and Research, World Health Organization (WHO); Sameer Pujari, “Be Healthy, Be Mobile” Project Manager, World Health Organization (WHO)

Track Overview
The Sustainable Development Goals (SDGs) are the 17 global priority goals to be achieved by 2030, agreed upon by the UN member states in 2015. The 17 goals include a more detailed 169 targets and indicators. Achieving the SDGs within the timeframe is an enormous challenge across the scientific and economic community.

Universal Health Coverage (UHC) is one of the SDG targets that aims at ensuring that all people can access quality health services, to safeguard all people from public health risks, and to protect all people from impoverishment due to illness, whether from out-of-pocket payments for health care or loss of income when a household member falls sick.

Artificial Intelligence applications can be a game changer to achieve Universal Health Coverage goals by empowering frontline health workers to enable early stages of diagnostic like Malaria or Cervical Cancer detection or to identify population at risk of developing non-communicable diseases like detecting diabetes/cardiac-risk from the iris, etc. Currently much of this work is done manually, limiting the frequency and scale of coverage. AI can be used also during health and natural disaster emergencies that can significantly increase the efficiency of disaster response and save more lives.

Aim of the track

Identify quick-wins areas and types of AI applications that hold high-potential impact on health outcomes and that are feasible and relatively simple to deploy
Identify bottlenecks and challenges that prevent society from taking full advantage of AI quick-wins in areas such as diagnostic, treatment, prevention or emergency response and how to possibly address those.
Mobilize community to support a number of proof-of-concepts projects that could demonstrate impact in the short and medium term.
To make commitment and seek feedback from health community of a global XPrize Challenge to ‘Transform Community Health through AI’ (tbc).

Potential domains for AI quick-wins for Public Health (workstreams)

AI for primary care and service delivery
Artificial Intelligence can help make certain types of diagnostics almost ubiquitous, sometimes in combination with off-the-shelf cameras or smartphones. Similar tools can help guide primary-care workers through treatments that they might otherwise not have been able offer.

Examples:
- Diagnosis and Management of Chronic Cardiovascular, Diabetes, and Respiratory diseases
- Prediction and identification of populations at risk of non-communicable diseases
- Services for high risk pregnancy and early childhood development disorders e.g., detection of hearing impairments or autism.
- Cervical Cancer screening (Cervicography) and skin cancer detection
- Malaria and Tuberculosis diagnosis
- Snake-bite detection
- Pre-diagnostics, self-assessment and remote screening (ear, nose and throat, mental health, dermatology etc.)
- AI-powered health consultations / telemedicine
Outbreaks, Emergency Response and Risk Reduction
Increasingly, the earliest signs of an epidemic are digital, be it through search queries on Google, or through social media, or through patterns of movement, as can be approximated by things such as movements of mobile phones. Artificial Intelligence can help monitor the vast amounts of data that need to be examined in order to pick up the earliest possible signs.

Examples:
- Analyzing satellite and drones imageries to identify measures for Disaster Risk Reduction or to assess affected areas, level of damage and efforts required
- Mining social media data for early detection of outbreaks and assessing population needs during emergencies
- Better manage dispatcher centers during emergencies by filtering out redundant or less urgent calls, interacting with callers naturally, instantly transcribe and translate languages and analyze the tone of voice for urgency.
- Automated responses to public health related queries for nearest care center for example particularly during epidemics and humanitarian assistance.
Health promotion, prevention, and education
As more information about the health status of individuals becomes digital in a systematic manner, it becomes possible to monitor for conditions at a greater frequency and lower cost. Physiologically, heart rate monitors, etc. can be combined with running times. Additionally, through the use of Natural Language Processing (NLP), chatbots can help people with identifying conditions, as well as with identifying mental conditions.

Examples:
- Monitor movement and activities to make recommendations to maintain mental and physical health.
- In-home health monitoring and health information access to detect changes in mood or behavior and alert caregivers.
- Personalized health management to mitigate the complexities associated with multiple comorbid conditions and/or treatment interactions.
AI health policy
Personal medical data is of a highly sensitive nature, as such, the use of artificial intelligence in health applications, which is currently still in its infancy, gives rise to a number of ethical questions, such as those relating to data privacy.

Examples:
- Ethics
- Equity
- Privacy and data protection
- New WHO Resolution on Digital Health to recognise the potential of digital technology including ehealth, mhealth and emerging tech such as AI - opportunities, challenges, gaps.

Track 4: Trust in AI

Team Lead: Huw Price, Francesca Rossi, Zoubin Ghahramani, Claire Craig
Members: Stephen Cave, Kanta Dihal, Adrian Weller, Seán Ó hÉigeartaigh, Jess Whittlestone, Charlotte Stix, Susan Gowans, Jessica Montgomery
Theme Managers: Ezinne Nwankwo, Yang Liu, Jess Montgomery

Meet the team: trustfactory.ai

Track Overview
Artificial intelligence (AI) has the potential to dramatically accelerate the pace at which the United Nations’ Sustainable Development Goals (SDGs) can be achieved. Maximising AI’s potential for good will strongly depend on building trust in AI, in several dimensions. This track will focus on three dimensions of trust. Developers of AI solutions must earn the trust of communities to which such solutions are offered. AI developers and others working for beneficial AI must trust each other, across cultural, national and corporate boundaries. And AI systems themselves must be demonstrably trustworthy.

This track will explore these three dimensions of trust under three Themes:

Building trust for beneficial AI – stakeholder communities
Building trust for beneficial AI – developer communities
Building trust for beneficial AI – trustworthy systems

We will present three projects under each Theme. Participants will be invited to refine the projects, to propose potential research, impact, and funding partners, and to join projects as collaborators.

Theme A: Building trust for beneficial AI – stakeholder communities

Dr. Becky Inkster (Department of Psychiatry, Cambridge)
Building better care connections: establishing trust networks in AI mental healthcare
Dr. Dina Machuve (Nelson Mandela African Institute of Science and Technology and Technical Committee Member for Data Science Africa)
Assessing and Building Trust in AI for East African Farmers: A Poultry App for Good
Irakli Beridze (United Nations Interregional Crime and Justice Research Institute, UNICRI)
Building Trust in AI – Mitigating the Effects of AI-induced Automation on Social Stability in Developing Countries & Transition Economies

Theme B: Building trust for beneficial AI – developer communities

Prof. LIU Zhe (Peking University, Beijing)
Cross-cultural comparisons for trust in AI
Dr. Kanta Dihal (Leverhulme Centre for the Future of Intelligence, Cambridge)
Global AI Narratives
Prof. David Danks (Carnegie Mellon University, CMU)
Cross-national comparisons of AI development and regulation strategies – the case of autonomous vehicles

Theme C: Building trust for beneficial AI – trustworthy systems

Dr. Jess Whittlestone (Leverhulme Centre for the Future of Intelligence, Cambridge)
Trust in AI for governmental decision-makers
Dr. Rumman Chowdury (Accenture)
Trustworthy data: creating and curating a repository for diverse datasets.
Dr. Krishna Gummadi (Max Planck Institute, Saarbrücken) and Dr Adrian Weller (Leverhulme Centre for the Future of Intelligence, Cambridge and Alan Turing Institute, London)
Cross-cultural perspectives on the meaning of 'fairness' in algorithmic decision making