AI for Social Impact

by TUM.ai department of educationOctober 14th, 2021

Motivation

Ever since I first started learning about artificial intelligence (AI), I have perceived the potential of AI mostly from the perspective of the companies applying it: how AI helps increase the efficiency of their processes, thereby saving costs and generating opportunities for new revenue. When experiencing AI through the lens of commerce only, however, a large part – in my opinion the most essential part - of its potential for humanity remains uncovered: the application of AI for social good.

Desmond

AI for Social Good?

The design, development,
and deployment of AI systems in such a way that they prevent, mitigate, or
resolve problems that negatively impact human life and the wellbeing of the
natural world, and enable socially preferable and environmentally sustainable
developments [1].






An emerging field of scientific research, the concept of AI for social good (AI4SG) looks beyond automation and efficiency and is increasingly gaining momentum in the AI community. It focuses on tackling important social, environmental, and healthcare challenges of today’s world with the help of AI. To deliver real-world social impact with AI, however, we will need to go further than providing methodological advances in terms of newer models and algorithms. This blogpost provides an overview of potential application areas, real-world use cases, and key questions in this emerging area of research.

Application Areas of AI4SG

AI has great potential to achieve positive social impact in a broad variety of domains. Through an analysis of about 160 AI forsocial good use cases the McKinsey Global Institute (MGI) identified and characterized ten diverse domains in which applying AI could have large-scale social impact. The application areas range from crisis response to education empowerment and are described in the following figure. The digits represent the number of use cases in the study falling under the respective domain [2].

Benchmarking AI4SG Use Cases

The potential application areas of AI4SG look promising. But how can we assess the value and success of relevant projects with a cogent framework given the novelty and fast growth of the field? The Oxford Initiative on AIxSDGs  proposes using the 17 United Nation’s Sustainable Development Goals (SDGs) as a benchmark for tracing the scope and spread of AI4SD [3]. To support their proposal, they collected a database of AI4SG projects using this benchmark, which can be accessed here.
The 17 SDGs are illustrated in the figure below.

Examples of AI4SG Use Cases

To give you an idea of what AI4SG projects can look like in practice, here’s three examples of how AI has successfully been applied for social good. Each one is aligned with at least one SDG.

AI for Flood Forecasting
Flooding causes thousands of fatalities, affects the lives of hundreds of millions, and results in huge economic damages annually. Google's Flood Forecasting Initiative aims at providing high-resolution flood forecasts with the help of AI and timely warnings around the globe, while focusing first on developing countries where most of the fatalities occur. In September 2020, Google has expanded its initiative to the whole of India and Bangladesh, where the technology can protect more than 200 million people in India and around 40 million people in Bangladesh [4]. Corresponding SDGs: No Poverty (SDG 1), Good Health and Well-Being (SDG 3)

Insert YouTube Video in Bootstrap Modal
view more on Youtube



 Seeing AI – Converting Visual Data into Audio Feedback
Seeing AI is a free app which uses AI to describe any environment through smartphone cameras and microphones to its user. It has been designed for the visually impaired community by a Microsoft Deep Vision hackathon team with the aim to improve the lives of the 253 million people worldwide living with either partial or total blindness. The software integrates use of the built-in camera to recognize people, specific items, or a general environment around the user. Thanks to image recognition and AI, users can aim their phones at almost anything and receive a vocal description of what it is. Facial-recognition technology enables the app to remember and recognize friends’ faces and even perform some emotion recognition to determine a person’s mood [5].Corresponding SDGs: Good Health and Well-Being (SDG3), Reduced Inequalities (SDG 10).

Insert YouTube Video in Bootstrap Modal
view Seeing AI on Youtube



Cognii - AI for Education & Training 
Cognii provides artificial intelligence-based educational technologies, working with organizations in the K-12, higher education, and corporate training markets to help deliver superior learning outcomes and cost efficiency. Its products include the Cognii Learning Platform, Virtual Learning Assistant, and Assessment Engine. Some of the most proven practices for improving engagement, learning, and retention are more frequent assessments, one-to-one tutoring, writing-to-learn, and open response assessments. These practices are not only costly but also generally depend upon access to a well-trained human. Cognii offers a solution, giving customized feedback, engaging students in active learning, and helping improve knowledge retention [6]. Corresponding SDGs: Quality Education (SDG 4), Decent Work and Economic Growth (SDG 8), Reduced Inequalities (SDG 10).

Insert YouTube Video in Bootstrap Modal
view Cognii on Youtube


How to Design AI for Social Good? - Seven Essential Factors

The AI4SG field is gaining momentum, the number of use cases is on a rise and the impact of successful applications is truly impressive. But what makes AI socially beneficial in principle, and what policies and regulations are needed to reproduce the initial success of AI4SG in practice? How can risks of unintended effects and potential misuses of AI4SG technologies be mitigated effectively? Floridi et al. [7] propose seven ethical factors and corresponding best practices for the design and deployment process of future AI4SG technologies that are critical in minimizing the risks of future AI4SG projects, which are elaborated in the table below.

These factors are a crucial step towards ensuring that AI4SG projects advance beneficial goals in socially preferable and sustainable ways, laying the ground for good practices and policies in this respect. However, there still remain wider ethical and political challenges in the context of decision-making with and about AI that need to be tackled to ensure the positive impact of the “AI4SG project” at large.

Conclusion

As proven by many examples of successful use cases, AI holds enormous potential to deliver large-scale positive social impact in a variety of domains and contribute to the fulfillment of the 17 SDGs. However, despite the increasing attention the AI4SG field receives, I can tell from my own experience that many AI developers are still not using their skills to contribute to unfolding this potential. With this post I want to inspire and encourage everyone skilled in AI to dedicate at least a fraction of their time to applying AI for social good – and think beyond automation and efficiency. Looking for an opportunity to get active? The TUM.ai  Makeathon with its “Social Impact” track is a great chance to start –

get your tickets for the Grande Finale on Oct 17, 2021, HERE VIEW GrandeFinale

Sources

[1] Akula, Ramya, and Ivan Garibay.
"Ethical AI for Social Good." arXiv preprint arXiv:2107.14044 (2021).

[2] Chui, Michael, et al. "Notes from the AI
frontier: Applying AI for social good." McKinsey Global Institute (2018).

[3] Cowls, Josh, et al. "A definition, benchmark
and database of AI for social good initiatives." Nature Machine
Intelligence 3.2
(2021): 111-115.

[4] Matias, Yossi. “A big step for flood forecasts in
India and Bangladesh.” AI. Sep 01, 2020.
blog.google/technology/ai/flood-forecasts-india-bangladesh/

[5] “Seeing AI.” AISDG. www.aiforsdgs.org/all-projects/seeing-ai-0
[6] Cognii. www.cognii.com/
[7] Floridi, Luciano, et al. "How to design AI for social good:
Seven essential factors." Science and Engineering Ethics 26.3
(2020): 1771-1796.