The End Of Knowledge - Vault 6/82 - CVPR - 2023 - Machine Learning with Social Purpose

graph LR classDef social fill:#f9d4d4, font-weight:bold, font-size:14px classDef global fill:#d4f9d4, font-weight:bold, font-size:14px classDef impact fill:#d4d4f9, font-weight:bold, font-size:14px classDef fairness fill:#f9f9d4, font-weight:bold, font-size:14px A[Machine Learning with
Social Purpose] --> B[Social
Purpose] A --> C[Global
Impact] A --> D[Research
Impact] A --> E[Fairness
and
Participation] B --> B1[Discusses AI, ML
with social
purpose. 1] B --> B2[Focus on weather,
climate, global
community. 2] B --> B3[Probabilistic ML
remains a core
area. 3] B --> B4[Generative models
impact earth systems
research. 4] B --> B5[Socio-technical AI
research portfolio. 5] B --> B6[Socio-technical AI
considers social
aspects. 13] C --> C1[Global AI shifts
to general
social purpose. 6] C --> C2[Earth system models
for sustainable
living. 7] C --> C3[ML in global
weather forecasting
applications. 8] C --> C4[Scorecards summarize
model performance
visually. 9] C --> C5[ML supports vital
weather decision-making. 10] C --> C6[Better forecasts harm
marginalized
communities. 11] D --> D1[No system exists
independently of social
context. 12] D --> D2[AI ecosystem view:
research, deployment,
governance. 14] D --> D3[ML research enables
social purpose
claims. 16] D --> D4[Health and environment
intersection needs
ML. 17] D --> D5[Rainfall predictions need
hydrology
models. 26] D --> D6[Open-sourcing ensures
reproducibility, assessment. 27] E --> E1[Participatory design
involves affected
communities. 15] E --> E2[New fairness approaches
for unknowable
characteristics. 18] E --> E3[Diverse researchers explore
socio-technical
problems. 19] E --> E4[Intercultural dialogue,
global community
building. 20] E --> E5[Grassroots organizations support
AI
dialogue. 21] E --> E6[Global AI field growth
due to grassroots
work. 22] class A,B,B1,B2,B3,B4,B5,B6 social class C,C1,C2,C3,C4,C5,C6 global class D,D1,D2,D3,D4,D5,D6 impact class E,E1,E2,E3,E4,E5,E6 fairness

Resume:

1.- Shakir Mohamed, a prominent AI researcher, discusses machine learning with social purpose, connecting it to our planet and each other.

2.- Shakir's work focuses on weather and climate, science and society, global community building, and diversity and equity.

3.- Evidence shows probabilistic machine learning is important and should remain a core investment area for the research community.

4.- Mixing generative models with earth systems research creates pathways to impact, uncovers new questions, builds products, and connects to pressing needs.

5.- Research in technology and society should be undertaken together, deepening a socio-technical AI research portfolio.

6.- A more global AI field and industry can shift machine learning to be more general, magnifying social purpose.

7.- Earth system models, representing planetary processes, are central tools for understanding our changing planet and living sustainably.

8.- Machine learning is making inroads in medium-range global weather forecasting, with applications across commercial, industrial and social needs.

9.- Scorecards visually summarize model performance across variables, metrics, and important data subsets, showing ML can outperform operational weather systems.

10.- Rapid advances are being made in ML for weather, opening up new ways to support vital weather-dependent decision-making.

11.- Better forecasts can sometimes lead to greater harm and vulnerability for poor and marginalized communities.

12.- No technical system exists independently of the social world; the social and technical are enmeshed at every level.

13.- Socio-technical AI adapts the conceptual apertures used in technical work to account for a wider set of social considerations.

14.- An ecosystem view of AI exposes different levels (research, deployment, governance, cooperation) for new research directions and responsible actions.

15.- Including affected communities in AI design through participatory approaches places work on stronger ethical foundations.

16.- Theoretical and methodological ML research, like on generative model evaluation, is vital for enabling claims of social purpose.

17.- The intersection of health and environment is another area of social purpose needing further ML research.

18.- Analyzing model fairness across demographic subgroups is common, but many human characteristics are unknowable, requiring new fairness approaches.

19.- Diverse groups of researchers can leverage their identities and experiences to explore solutions for complex socio-technical ML problems.

20.- Meaningful intercultural dialogue and strengthening varied political communities globally are key to making AI development more equitable.

21.- Grassroots organizations in Africa and worldwide are building communities and movements that support locally-grounded AI dialogue and transformation.

22.- The field of AI is becoming more global due to the committed work of grassroots groups; continued support is vital.

23.- Researchers should motivate their work through a drive for social purpose, expanding the view of their responsibilities.

24.- An ecosystem perspective allows researchers to identify where and how to intervene to reshape the trajectory of AI developments.

25.- Support for grassroots AI initiatives is working to expand global participation in AI; more can still be done.

26.- Once rainfall is predicted, hydrology models are needed to simulate water flow and flooding; an area for future work.

27.- Open-sourcing AI models takes time to ensure reproducibility but enables independent assessment by environmental agencies.

28.- Participatory AI is an ongoing process where affected communities can change the direction of the work, not just deployment.

29.- Public experimentation with AI models raises socio-technical questions around watermarking, open-source, communication, and the basis of value systems.

30.- Machine learning researchers, especially students, should break out of narrow domains and be open to entirely shifting the field.

Knowledge Vault built byDavid Vivancos 2024