Published 2 years ago.

AI for Social Good (AI4SG)


As a tech enthusiast like many of us here, my passion for AI has originated from the cool applications that bring considerable commercial value, say Tesla; or the endless effort that keeps pushing the margin further from academia, you name it; or even the AI generated artistic vibe like Deep Dream. Along with my study, I have collaborated with projects like podcast highlighter, music generator, sentiment estimator and so on. I used to relate being cool with simply developing cool applications that attract capital, or researching cool stuff that gains me more followers on Twitter, without any specific purpose. However, as I recently finished the book ‘How I Learned to Understand the World‘ by Hans Rosling, a Swedish physician who devoted his life and research to narrowing the gap of public health levels between countries, and promoting the use of data to explore and visualize development issues. I started to recall how little effort I spent, with all the machine learning education I had, on direct positive social impact. A little Googling showed that I am not alone, much of today’s AI talent is involved in commercial and academic areas only, which is not wrong but certainly something we may want to reflect on. This brings us to today’s topic on how to apply AI technologies to deliver social good.

In accordance with United Nations 17 Sustainable Development Goals (SGDs), The AI for Social Good (AI4SG) aims to establish bridges from AI technologies towards these goals and deliver positive social impact. Many major players in the AI field have also initiated their effort to target applications that directly address social and environmental challenges. For example, Google has funded their team AI for Social Good; Microsoft launched their AI for Earth, Health, Accessibility, Humanitarian Action, and Cultural Heritage program. Conferences like NeurIPS, ICML, and ICLR included AI for Social Good workshops. The University of Chicago, the Alan Turing Institute, and other academic institutes have also launched Data Science for Good academic programmes. Stockholm is taking the lead in addressing the UN’s SGDs along its line of services. In today’s post, I will list five AI4SG applications which may inspire us as AI practitioners on what we can do to address social and environmental challenges.


  • Flood Forecasting by Google AI

Google started its flood forecasting pilot in 2018 as part of its AI4SG efforts. Floods are one of the most devastating natural disasters with an estimated 250 million population affected per year. With both real-time measured water level data and augmented, corrected satellite images from Google Maps, they built a hydraulic model that simulates the water movement on the flood plain. They have also used machine learning models, such as LSTMs, to improve some of the physic-based algorithmic and optimized their model to leverage TPUs parallel computation. With this, they can better forecast when and where floods will occur, how severe they might be, and incorporate that information into Google Public Alerts.

  • Inuktitut Revitalization by Microsoft

Nunavut sits in the northernmost part of Canada and is one of the most exotic, natural places. Although it has been home to an indigenous population for more than 4,000 years, it only hosts 39,000 people today. With Inuktitut as the first language of 70% of the local residents, traditions and knowledge were passed rather inclusively. Together with the Nunavut government, Microsoft has collected data and built a translating system for Inuktitut leveraging AI. Their effort not only serves as a preservation of cultural heritage but also an enrichment of local life with more connection with the outside world.

  • Troll Patrol by Amnesty International & Element AI

In this project, Amnesty International partnered with Element AI to carry out a research analysis of the unprecedented scale of online abuse against women in the UK and USA, with cutting-edge methods in natural language processing. They also evaluated the feasibility of fine-tuning deep learning model in toxic, abusive tweets detection. Their findings, sadly, showed a worrying pattern of online abuse towards women, among which the black group is 84% more likely to experience abusive online behaviors.

  • Deep Learning Indaba by Indaba Abantu

The Deep Learning Indaba is an organization that aims to encourage African participation and contribution to the advances of artificial intelligence and machine learning. Their three principal programmes include the annual Deep Learning Indaba, a week-long event of teaching, practical session, and debate on recent AI principles and practices; the IndabaX, a local Indaba leadership builder; and the Kambule and Maathai awards which recognizes research excellence at an African higher education institution. Their effort helps African groups engaging in AI from not only the view of observers but also shapers and owners of these technology advances.

  • Somalia COVID-19 Hotline by Shaqodoon

Shaqodoon is an NGO aiming to improve citizen involvement and social accountability in Somalia. With a predominantly oral tradition and high levels of illiteracy, Somalian citizens face the challenge of receiving timely instructions on precautions against COVID-19. They have built an Interactive Voice Response system with WHO-approved pre-recorded preventive messages on COVID-19. Their system can also identify suspected cases with speech recognition techniques. Their efficient, easily accessible method not only saved lives but also eased the work for public agencies in monitoring the current situation.


These examples are only a small part of the good initiatives that different parties are taking to address AI4SG. We should also encourage AI experts to actively seek out opportunities that can further empower AI4SG. I always wanted to do something good for society, my girlfriend and I have been helping with an elephant orphanage in Kenya, but out of so many reasons we did not end up travelling there and so little did we know about in what ways we have contributed to. This time I can finally make use of my knowledge, skills and my not-so-big network! I was thinking about maybe grouping some friends, or even start a non-profit podcast where we can brainstorm about what public good we can do with our machine learning education. Does this sound interesting that you might be interested? Do you have even better ideas? Let me know if I can help!

Published by Tianzong Wang 2 years ago.