AI for Social Good

CS 691 · IIT Gandhinagar · Semester II, 2024–25

This course explores how artificial intelligence can address critical challenges in healthcare, education, the environment and public safety, ethically and effectively. It draws on current research at venues such as IJCAI, AAAI, KDD and NeurIPS, and covers both the technical and the societal implications of AI-driven solutions.

Instructor
Prof. Nipun Batra
Office
13/401C (directions)
Email
nipun.batra@iitgn.ac.in
Format
Lectures, paper presentations and reviews, guest talks and a paper replication project
Book
AI for Social Impact, Milind Tambe, Fei Fang and Bryan Wilder
Discussion
Course Slack

Lectures

Recordings on YouTube All 10 sessions

Domains

Topics span a wide range of societal domains, including:

  • Healthcare: AI for diagnosis, personalised medicine, public health monitoring and remote care
  • Environment and sustainability: AI in climate modelling, pollution monitoring, energy efficiency and conservation
  • Disaster resilience: AI for disaster prediction, response and resource management
  • Education: personalised learning systems, intelligent tutoring and analysis of educational outcomes
  • Public safety and security: social engineering, detecting misinformation and predictive policing

Methods

The AI techniques covered include:

  • Supervised learning: classification and prediction, such as disease diagnosis, fraud detection and poverty prediction
  • Unsupervised learning: clustering and anomaly detection for environmental monitoring and resource allocation
  • Semi-supervised learning: working with limited labelled data
  • Reinforcement learning: resource optimisation, healthcare interventions and adaptive learning systems
  • Natural language processing: sentiment analysis, misinformation detection and speech recognition
  • Computer vision: object detection, segmentation and classification for environmental monitoring and healthcare imaging
  • Deep learning: neural networks for complex social problems, such as medical imaging and disaster prediction

Questions

How is a paper presentation graded?

Each student gives a 20-minute presentation of a paper, graded by the instructor and by peers out of 10 marks: presentation quality (3), Q&A (1), reasons to accept (2), reasons to reject (2) and explaining the approach (2). See Paper presentation.

What are the two paper reviews?

Review 1 (5 marks) is done in class and looks at the abstract, the introduction and fatal flaws. Review 2 (15 marks) is done at home on a review platform: everything in Review 1 plus a full AI paper review covering risks, ethics and datasets, informed by existing OpenReview reviews.

What is the project?

Groups replicate a NeurIPS-like paper in a social good setting, for 35 marks over three evaluations: a proposal with a Gantt chart, a mid-project progress update and a final presentation. See Paper replication project.

Are guest lectures assessed?

Yes. An in-class quiz on the guest lectures is worth 5 marks. The talks are listed on the Guest talks page.

How is class participation counted?

One mark for each active class participation, up to 5 marks. Attendance is a separate 5 marks.

Ask anything else on the course Slack.

References

Textbook:

  • Milind Tambe, Fei Fang and Bryan Wilder, AI for Social Impact (online)

Research papers from AAAI, IJCAI, KDD, NeurIPS, ICLR and similar top conferences and symposia.

Related courses:

Guest lectures

Guest lecturers announced at the start of the course, in no particular order:

  • Dr. Alok TalekarGoogle India
  • Dr. Harsh ParikhJohns Hopkins University
  • Dr. Aparna TanejaGoogle India
  • Dr. Anupam SobtiPlaksha University