The End Of Knowledge - Vault 4/74 - AI For Good - 2022

graph LR classDef greenai fill:#f9d4d4, font-weight:bold, font-size:14px classDef tii fill:#d4f9d4, font-weight:bold, font-size:14px classDef uae fill:#d4d4f9, font-weight:bold, font-size:14px classDef noor fill:#f9f9d4, font-weight:bold, font-size:14px classDef emissions fill:#f9d4f9, font-weight:bold, font-size:14px classDef collaboration fill:#d4f9f9, font-weight:bold, font-size:14px A[The Future of
Green AI] --> B[Dr. Almazrouei presents
on future Green AI. 1] A --> C[TII: Abu Dhabi, UAE
world-class R&D. 2] C --> D[UAE: sustainability,
hosting COP28. 3] A --> E[AI: narrow to
general intelligence evolution. 4] A --> F[Noor: Arabic NLP
model excels. 5] A --> G[AI lacks energy,
carbon footprint reporting. 6] G --> H[Carbon footprint factors:
model size, efficiency. 7] G --> I[Emissions: data storage,
training, model serving. 8] G --> J[Noors footprint: 65% training,
30% development. 9] A --> K[Best practices: efficient
inference, minimal emissions. 10] A --> L[Noors development:
36.5 tons CO2 emissions. 11] A --> M[Small Green AI efforts:
huge impact. 12] A --> N[Noors carbon footprint
report available. 13] A --> O[Progress needs collaboration
on algorithms, hardware. 14] O --> P[Carbon measurement extends
to other domains. 15] A --> Q[AI driven by needs,
sustainability important. 16] C --> R[UAE: knowledge economy,
global talent attraction. 17] C --> S[TII to showcase
at COP28 in November. 18] class A,B,C,D,E,F,G,H,I,J,K,L,M,N,O,P,Q,R,S greenai class C,D,R,S tii class C,D,R,S uae class F noor class G,H,I,J,K,L,M,N emissions class O,P collaboration

Resume:

1.- Dr. Ebtesam Almazrouei from Technology Innovation Institute (TII) in Abu Dhabi presents on the future of Green AI.

2.- TII aims to position Abu Dhabi and UAE as a world-class R&D institute working on applied research across many sectors.

3.- UAE is committed to sustainability initiatives and will host COP28. Advanced technology and AI can address climate change.

4.- Evolution of AI from narrow AI to general intelligence, with large language models being key to achieve human-like reasoning.

5.- Examples show Noor, the largest Arabic NLP model, excelling at text generation, chatbots, summarization by "thinking" and connecting information.

6.- AI community lacks reporting of energy consumption and carbon footprint of large AI models, especially during inference.

7.- Factors impacting AI's carbon footprint: model/data size, hardware efficiency, data center efficiency/location, energy mix.

8.- Sources of emissions span data storage/transfer, experimentation, training, model serving, and other research-related activities.

9.- Noor's footprint: 65% from training, 30% from development. Inferences could emit 30 tons CO2 per day if widely used.

10.- Best practices: optimal model architectures, efficient inference, new algorithms, tailored accelerators, minimized emissions from research activities.

11.- Noor's development alone emitted 36.5 tons CO2, equal to emissions of 2 Americans or 3 Swiss per year.

12.- Small efforts toward Green AI can have huge impact. Advancement in AI should consider environmental impact.

13.- Report detailing Noor's holistic carbon footprint assessment is publicly available.

14.- Progress requires collaboration - new efficient algorithms, hardware, data centers, applications all play a role.

15.- Methodology for measuring AI's carbon footprint could extend to other data-heavy domains like healthcare.

16.- AI development is driven by real needs that it can address, outweighing energy costs, but sustainability is important.

17.- UAE investing heavily in building knowledge-based economy and attracting global talent for research and development.

18.- TII and other UAE organizations will showcase sustainability projects at COP28 in November 2023.

Knowledge Vault built byDavid Vivancos 2024