Artificial Intelligence for Social Good: Our Approach at Wadhwani AI
Total Page:16
File Type:pdf, Size:1020Kb
Artificial Intelligence for Social Good: Our Approach at Wadhwani AI P. Anandan Sep 13, 2018 #AIforALL Wadhwani Institute for @WadhwaniAI Artificial Intelligence (Wadhwani AI) is an independent, nonprofit research institute and global hub for developing AI solutions for social good. Our mission: AI for ALL AI for ALL Wadhwani Institute for Artificial Intelligence WadhwaniAI.org OUR FOUNDERS DR. ROMESH WADHWANI MR. SUNIL WADHWANI Tech Entrepreneurs Dr. Romesh Wadhwani and Mr. Sunil Wadhwani are committing $3M/year for 10 years (total $30M) as well as their own time and the benefit of their entrepreneurial experience AI for ALL Wadhwani Institute for Artificial Intelligence WadhwaniAI.org #AIforALL Our main goal is to develop AI @WadhwaniAI solutions to benefit the under- served billions in developing countries, in domains including: • Health • Agriculture • Financial inclusion • Language • Infrastructure • Education AI for ALL Wadhwani Institute for Artificial Intelligence WadhwaniAI.org Wadhwani AI launched on Feb 18 by the Hon’ble PM of India Shri Narendra Modi OUR MISSION Create innovative solutions to Societal Challenges based on Artificial Intelligence and Innovate related technologies Deploy the solutions in underserved communities through Government programs Impact and the efforts of non-Governmental organizations Catalyze AI for social good innovation by being a hub where researchers and agencies Catalyze from across the world collaborate in solving societal challenges Become the world’s leading applied research institute focused on AI for social and Pioneer economic good UNIQUE APPROACH PARTNERS AND Our AI researchers work CHAMPIONS alongside diverse team PATH, Gates Foundation, WISH members: data scientists, Foundation, and Government engineers, designers, product Usability of India are among our early Tests managers, domain experts, partners and champions. partnership managers, and Product Researchers from Stanford, entrepreneurs. Requireme Tech POC nts USC, University of Washington, Collaborative and NYU are among our research partners. Human-Centric Nonprofits / Dev Orgs Alpha & Use Cases Lean & Agile Governments Field Tests Corporates / Startups Research & Academia Specific Large Find Problem Pilots Small Scale Partners Areas Problems Scale Our Initial Focus Areas FRONTLINE COTTON TUBERCULOSIS HEALTH FARMING Frontline Health September 2018 Frontline Health: Background ● Primary Health resources: 27k PHCs and 150k Sub-centers, with 32k doctors, 250k ANMs and 1 million ASHAs form the frontlines ● Govt wants to upgrade existing facilities to 150k tech-enabled Health & Wellness Centres (HWC) ● Opportunities to: ○ overcome skill gaps ○ expand care beyond mother-and-child, cope with workload ○ digitize measurements and tests ○ connect the data dots ● Partners: WISH, Gates Frontline Health: Journey to date Early exploration (Mar-July 2018): Identified a list of initial use cases based on research and interviews. Interviewed creators of (or surveyed) apps used by primary health workers in India (e.g., CAS, Mobile Kunji, etc.) and globally (Ada, Babyl, Blackwell) Field research (June 2018): Team members visited rural PHCs and Sub-centers managed by WISH Initial use case: Advisory app for frontline workers WISH/Gates/ Wadhwani AI workshop (August 10th 2018): Led a day-long workshop to generate ideas for AI solutions in primary health, with inputs from the field, program and systems levels New use cases: Identified and prioritised three new use cases after workshop • Detection & management of high-risk pregnancy (Ongoing) • Anthropometric imaging for growth tracking (Ongoing) • Triaging app for frontline workers Upcoming workshop at Gates Bihar Partners Meet (Sep 20-21 2018): Conducting a second AI- focused workshop to generate deeper insights and identify specific solutions High-risk pregnancy (HRP): Solution Areas Possible solutions mapped to hypotheses No ANCs or check-ups, Some ANCs happen HRP is detected but HRP is detected but hence HRP is never but HRP is not properly not managed in the not managed during detected detected in time antenatal period childbirth E.g., Geotagged HRP E.g., Tools for better E.g., Geotagged E.g., “Capacity + hotspots using early outreach, HRP reminder / alert logistics planning” for socioeconomic Diagnostic kit for system for regular PHCs and First indicators + satellite ASHA + AI-powered follow-up with high- Referral Units (FRU) imagery for better ultrasound + App to risk pregnant woman for childbirth last-mile outreach calculate a risk score / risk vector Leading hypothesis Hypothesis #2 Conducted Gates+Wadhwani AI+WISH Workshop on AI for Public Health (Aug 10th) • Day-long workshop (10 AM to 6 PM) at the • Excellent participation and engagement Gates Foundation office on August 10 • Identified over a dozen broad use case areas and • 27 participants across seniority levels: from prioritised 5 field staff at WISH to country leads at Gates • Buy-in from Gates and Wish Foundation High-risk pregnancy (HRP): Overview What makes HRP a compelling problem area ● High-impact problem area with cascading effects on mother and child Critical Public health and well-being Health Issue ● Strong political will: aligned with UN-SDG and National Health Policy goals ● Data collection processes in place, rich datasets available Rich data ● Individual pregnant women already tracked by ANMs, records digitised and available through the national Mother & Child Tracking System ● Strong partners in the form of Gates (rich experience and insights at the Pathways to scale policy, strategy, and implementation levels) and WISH Foundation (deep expertise in implementation) Anthropometric Imaging: Overview • Anthropometric imaging identified as a high-priority need by Gates India and WISH Foundation, especially to identify low birth weight babies • Had a preliminary conversation with Kenneth Brown (Nutrition team, BMGF Seattle) on 7 Sep to discuss the anthropometric imaging tool developed by Gates grantee Body Surface Translations (BST) • Currently evaluating different technical approaches with a focus on accuracy, cost, and computational power Tuberculosis Eradication September 2018 Tuberculosis: Background • 2016: 2.8 million new cases, 0.43 million deaths in India • 2016: 10.4 million new cases, 1.7 million deaths globally • 2016: DR-TB cases grew 13% from 130,000 to 147,000 in India • Indian government target : eliminate TB by 2025 • WHO target : eliminate TB by 2050 • Partners : PATH (and WISH) Tuberculosis TB is recognized as a crisis, predominantly affecting low-income populations. Drug-Resistant TB (and Multi-Drug-Resistant TB) are growing at alarming rates. • TB affected 2.8 million and killed 0.43 million in 2016 in India. • Data & Piloting: PATH is the leading international non-profit • DR-TB cases grew 13% from 130,000 to 147,000 from 2015 to working on TB. Its flagship TB program is in Mumbai (in 2016 partnership with Mumbai Corporation), with support from • Government has set the ambitious target of eliminating TB by government’s Central TB Division, WHO, and Gates. The 2025* program has patient level data and also screening and diagnostics data. • Use-cases • Diagnosis through sputum slide analysis • Scaling: PATH advises government at central level. Elements of • Decision support for Multi-Drug Resistant TB the program are already being borrowed for national level • Treatment adherence & patient risk assessment scaling as part of RNTCP (Revised National TB Control Program). • Automated X-ray analysis for screening • Outbreak prediction • Partners • PATH • Will approach Mumbai Corporation, and relevant state and central bodies as needed. *National health policy Guiding framework: The tuberculosis treatment Treatment failure, relapse cascade TB patients who died, DROP-OFFS TB patients who do TB patients who sought care but TB patients failed treatment, were lost AT EACH not seek out care were not diagnosed or were who were to follow-up or were “not STAGE diagnosed but not reported. diagnosed, but evaluated never initiated on treatment or 2,500,000 1 never formally registered. TB patients who 2,000,000 2 relapsed within 18 months of 1,500,000 3 completing 4 Persons 5 treatment. 1,000,000 500,000 - 1 2 3 4 5 6 Patients New active TB Accurately seeking care Initiating first Completing Incidence cases diagnosed with from a health line treatment treatment Outcome (incidence) TB provider Cascade: Path that a TB patient takes from development of active TB to incidence outcome following treatment. We structure challenges in TB by stage of cascade 19 NOT TO SCALE – TO BE FINALIZED Losses at each stage of care cascade in public sector in India (Private sector numbers largely unknown) Previously treated (15% - 20%) ~40% Indians with ∑ latent TB (increasing 2,500,000 pool), some convert 1 No care-seeking (~17%) into active disease 2,000,000 Not identified as TB Deaths 2 patient or not reported Loss to follow- ~423K up, pre- No robust Private sector: Other (e.g. cured 1,500,000 3 treatment ~1.2M Loss to follow-up, private sector) data on (52%) 4 on treatment 2.8M Private sector: ? Treatment private 1,000,000 (partially known 0.4M) Private sector: ? 5 failure Private sector: ? sector post Persons Public sector: Public sector:1 care- 500,000 Public sector ~1.1M ~1.4M Public sector: Public sector: (48%) ~1.3M Cured: seeking ~1.04M ~960K - stage Patients seeking care New active TB cases Accurately diagnosed Initiating first line Completing 1 from2 a health 3 4 5 Incidence6 Outcome (incidence) with TB treatment treatment