Back during the year 2018, I was looking for ways to solve pressing problems in India and globally by applying modern Machine Learning (ML) techniques. I made some notes in that pursuit. Sharing them so that others can find it useful if not outdated completely. Thanks to my friends Anirudha, and Akshay and smart & lovely Kalzang who helped me with multiple rounds of reading and discussions.

I used the resources put by the organizations like McKinsey, NITI Aayog, and UN to identify challenges and domains that the world today faces. I also used the ideas from the 17 UN Sustainable Development Goals (SDGs) which although are not directly related to ML, help to anchor thoughts and strategies.

Here are the few domains where ML can be most impactful in India and other emerging economies:

• Crisis Response: Search and Rescue post-disaster, disease outbreak, help firefighters during a wildfire, help drones find a missing person in the wilderness. Here , the need for rigorous testing to ensure accuracy is essential. If the solution incorrectly indicates that a road is clear of flooding and directs thousands of people toward it, there could be significant consequences, such as increased risk of harm and reduced evacuation speed.
• Economic Empowerment: Help exploit economic resources and opportunities even to the poorest and left out population. Eg: Farmbeats for enabling data-driven farming with an edge-computing technology.
• Education: The goal here is to maximize student learning and teacher productivity. For example, adaptive learning technology could be used to recommend content to students based on past success and engagement with the material. ML could also be used to detect student distress early before a teacher has noticed.
• Environmental Challenges: Sustaining biodiversity, controlling pollution, combating natural resource depletion, and climate change. For example, robots with AI capabilities can be used to sort recyclable material from waste. The Rainforest Connection, a Bay Area nonprofit, uses AI tools such as Google’s TensorFlow in conservation efforts across the world. Its platform can detect illegal logging in vulnerable forest areas through analysis of audio sensor data. Other applications include using satellite imagery to predict routes and behaviour of illegal fishing vessels. Planet Labs, an Earth Imaging Startup that uses satellite images with geospatial data to monitor changes in any reef ecosystem.
• Equality and Inclusion: Addressing equality, inclusion, and self-determination challenges, such as reducing or eliminating bias based on race, sexual orientation, religion, citizenship, and disabilities. One use case, based on work by Affectiva, which was spun out of the MIT Media Lab, and Autism Glass, a Stanford research project, involves use of AI to automate emotion recognition and provide social cues to help individuals along the autism spectrum interact in social environments. Another example is the creation of an alternative identification verification system for individuals without traditional forms of ID, such as driver’s licenses.
• Health and Disease: Early stage diagnosis. Other use cases include combining various types of alternative data sources such as geospatial data, social media data, telecommunications data, online search data, and vaccination data to help predict virus and disease transmission patterns.
• Hunger: Optimized Food Distribution especially for areas that are struck or about to get struck with shortages and famines.
• Informal Verification and Validation: This means to provide helpful, valuable, and reliable information to all which otherwise can mislead and disseminate polarizing information.
• Infrastructure Management: Energy, water and waste management, transportation, real estate, and urban planning. Eg: Maximizing traffic throughput, identifying malfunctioning components in an operational bridge.
• Public and Social Sector Management: Strengthening institutions, transparency, financial management. Eg: Detecting tax fraud, automated question answering at a public-government interface.
• Security and Justice: Tracking criminals, reducing bias in police forces, deducing safe path in burning buildings through IoT devices. Thorn: uses face detection and person identification, social network analysis, natural language processing, and analytics is being used to identify victims of sexual exploitation on the internet and dark web.

Challenges/Bottlenecks:

• Outdated educational curriculum: lack of practicum integration and below par discourse quality have left most of the world wanting. This calls for an easier to use ML framework which even uninitiated ML folks can use through their phones or any other medium of their convenience. ML has a high technical learning curve barrier to entry otherwise. Competition in for-profit sector is fierce and so there is a shortage of talent.
  Potential solution: Online courses can be helpful. High level AI expertise is concentrated in US, so there is a need to foster the connection of communities in this socially driven ecosystem in the developing world with them.
• The measurement problem: The social value of AI in terms of how much human suffering they alleviate is hard to measure, and one metric usually doesn’t hold across multiple domains. Hence it is hard to gauge and compare them.
• Bureaucratic inertia
• Privacy, transparency, causality issues in ML. If not taken care of, they might end up hurting people whom they were supposed to serve. Solutions should comply with the law like the one EU made lately, legit use of data can be mandated. This area is still a bit grey although recent events like that of Cambridge Analytica has sparked a great deal of discussion around it.
• Last mile implementation challenges
• Problems while scaling up.

Conclusion:

• Any progress in any field is by asking the right questions. Running AI on the grounds comes with a lot of challenges that do not show up while only working in labs. Because these challenges are what stops ML from seeping into our societies, research should be guided by these problems and hence are wonderful motivation to keep doing great research by helping us ask the right questions.
• ML is definitely not the first or obvious solution to most of the problems, however can play a crucial role by being in the solution mix. It can be used where volume of data is so huge that humans can not do it, greater accuracy than human is needed, and pick up characteristics that humans might miss (example, person in a picture).

References:

1. https://www.mckinsey.com/mgi/overview
2. MGI review paper
3. NITI Aayog India AI strategy pre-read
4. Gates Foundation
5. AI for good by Google
6. AI for good initiative by Microsoft
7. 17 UN SDGs – sustainabledevelopment.un.org
8. partnershiponai.org/
9. disasterscharter.org