Knowledge Vault 4 /9 - AI For Good 2017
AI for Common Good and Sustainable Living
Fei-Fei Li et al.
< Resume Image >
Link to IA4Good VideoView Youtube Video

Concept Graph & Resume using Claude 3 Opus | Chat GPT4o | Llama 3:

graph LR classDef vision fill:#ffeb99, font-weight:bold, font-size:12px classDef applications fill:#d3f4ff, font-weight:bold, font-size:12px classDef diversity fill:#ffd3e0, font-weight:bold, font-size:12px classDef progress fill:#d3ffd4, font-weight:bold, font-size:12px classDef challenges fill:#f3d3ff, font-weight:bold, font-size:12px A[AI for Common
Good and Sustainable
Living] A --> B[Vision: animal evolution,
computer vision. 1] B --> C[Computer vision lacks
sight for blind. 2] A --> D[Monitors hand hygiene
in hospitals. 3] A --> E[Car detection predicts
city demographics. 4] A --> F[Translating visual census:
challenges, solutions. 5] A --> G[Co-founded AI4ALL:
diversity in AI. 6] G --> H[Diversity challenges:
education, culture,
beneficence. 7] A --> I[Startups drive exponential
AI progress. 8] I --> J[Digital twins optimize
processes digitally. 9] I --> K[Address privacy,
security, safety issues. 10] A --> L[Machine learning merges
with 5G design. 11] L --> M[Video compression
uses machine learning. 12] L --> N[Interpretable AI:
understanding,
trust decisions. 13] A --> O[AI: optimal decisions
over data collection. 14] O --> P[AI embedded in workflows,
processes. 15] P --> Q[Talent shortage
limits AI industrialization. 16] A --> R[Evolutionary computation
automates AI design. 17] R --> S[Smart agriculture
maximizes crop yields. 18] S --> T[AI: wellness,
early sepsis prediction. 19] A --> U[Regulatory hurdles slow
AI deployment. 20] U --> V[AI proxies protect privacy
in interactions. 21] U --> W[Big tech races
to control AI. 22] A --> X[AI energy efficiency
vs human brains. 23] X --> Y[Standardization needed
for autonomous vehicles. 24] Y --> Z[Democratizing data access
aids AI development. 25] class B,C vision class D,E,F applications class G,H diversity class I,J,K progress class L,M,N challenges class O,P,Q progress class R,S,T applications class U,V,W challenges class X,Y,Z progress

Resume:

1.- Fei-Fei Li discussed the importance of vision in animal evolution and the progress of computer vision technology.

2.- Computer vision has many applications but sight has not yet been given to the visually impaired.

3.- Li's work includes using computer vision to monitor hand hygiene in hospitals to prevent infections.

4.- Another project used car detection in Google Street View images to predict demographics and statistics of American cities.

5.- Challenges remain in translating the visual census approach to developing countries without prevalent cars. Satellite imagery offers possibilities.

6.- Li co-founded the non-profit AI4ALL to increase diversity in AI education and participation, especially for women and minorities.

7.- Challenges include the lack of diversity in academia and industry. Solutions involve education, culture change, and focusing AI on beneficent causes.

8.- Antoine Blondeau said startups are key to going from linear to exponential progress in AI.

9.- Digital twins turn costly real-world processes into digital ones, enabling optimization. AI is one component.

10.- Obstacles like privacy, security and safety need to be addressed for digital transformation.

11.- Machine learning and communications engineering are converging, e.g. in designing 5G networks.

12.- Video coding standards have incorporated machine learning to compress video with high quality.

13.- Interpretable machine learning enables understanding how AI makes decisions, which is important for designing, certifying and trusting AI systems.

14.- Blondeau believes AI should be about making optimal decisions, not just collecting data and making predictions.

15.- AI has to be embedded in workflows and processes to enable self-learning systems that truly augment human intelligence.

16.- The talent shortage is a major bottleneck in industrializing AI. 40 PhD years were needed to develop AlphaGo.

17.- Evolutionary computation and massive distributed computing can help automate AI system design and leapfrog manual engineering.

18.- This approach is being applied to smart agriculture in partnership with MIT to maximize crop yields.

19.- Similar technology could optimize human wellness and enable early sepsis prediction to save lives in ICUs.

20.- Regulatory hurdles slow the real-world deployment of potentially life-saving AI. Changes are needed to enable faster progress.

21.- An opportunity exists for AI proxies under user control to intermediate and protect privacy in interactions with companies.

22.- Big tech companies are racing to control the AI layer to preserve their business models. Alternatives could disrupt this.

23.- AI systems still require much more energy than human brains. Energy efficiency is a frontier.

24.- Industrializing AI may require standardization, especially for applications like autonomous vehicles operating across borders.

25.- Democratizing access to data could help level the playing field globally for developing impactful AI solutions.

Knowledge Vault built byDavid Vivancos 2024