The End Of Knowledge - Vault 4/55 - AI For Good - 2020 - Collective problem solving with AI

graph LR classDef main fill:#f9f9f9, font-weight:bold, font-size:14px classDef aiGood fill:#ffcc99, font-weight:bold, font-size:14px classDef tracing fill:#ccff99, font-weight:bold, font-size:14px classDef discovery fill:#99ccff, font-weight:bold, font-size:14px classDef urgency fill:#ff99cc, font-weight:bold, font-size:14px classDef quantum fill:#ccccff, font-weight:bold, font-size:14px classDef community fill:#ffccff, font-weight:bold, font-size:14px classDef future fill:#cc99ff, font-weight:bold, font-size:14px classDef loop fill:#99ccff, font-weight:bold, font-size:14px A[Collective problem solving
with AI] A --> B[AI for Good
Initiatives] B --> B1[AI for contact tracing,
drug discovery. 1] B --> B2[Researchers must understand
societal impact. 2] B --> B3[Wisdom race:
tech progress vs. misuse. 3] A --> C[Contact Tracing] C --> C1[Covi app uses
self-reported symptoms. 4] C --> C2[Covi outperforms
manual tracing. 5] C --> C3[Privacy challenge:
quantized risk messages. 6] A --> D[Drug Discovery] D --> D1[Active learning for
efficient drug discovery. 7] D --> D2[Graph neural nets
for molecular structure. 8] D --> D3[Reinforcement learning
for drug candidates. 9] D --> D4[Current model incentivizes
secrecy. 10] A --> E[Scientific Urgency] E --> E1[Urgency of science
for global problems. 11] E --> E2[AI accelerates
scientific discovery. 12] E --> E3[AI guides retrosynthesis,
robotic synthesis. 13] E --> E4[AI-driven innovations
in five years. 14] A --> F[Quantum Computing] F --> F1[Quantum computers needed
for simulations. 15] F --> F2[IBM aims for
1000+ qubit machine. 16] F --> F3[Quantum circuits embedded
in programs. 17] F --> F4[Quantum community using
IBMs cloud service. 18] A --> G[Community and Collaboration] G --> G1[Scaling quantum computers
for discovery. 19] G --> G2[Collaborative, multi-disciplinary
discovery efforts. 20] G --> G3[COVID-19 HPC
Consortium example. 21] G --> G4[Consortium: 85+ projects,
600 petaflops. 22] A --> H[Future Readiness and Diversity] H --> H1[Scientific readiness reserve
for challenges. 23] H --> H2[Diversity of innovation
models advantageous. 24] H --> H3[AI, nanotech for
COVID-19 treatments. 25] H --> H4[Indigenous knowledge in
drug discovery. 26] A --> I[Closing the Loop] I --> I1[Implementing discovered
solutions. 27] I --> I2[Technology insufficient
collaboration key. 28] I --> I3[Virtuous cycle of
research, applications. 29] I --> I4[New social contract
for science. 30] class A main class B,B1,B2,B3 aiGood class C,C1,C2,C3 tracing class D,D1,D2,D3,D4 discovery class E,E1,E2,E3,E4 urgency class F,F1,F2,F3,F4 quantum class G,G1,G2,G3,G4 community class H,H1,H2,H3,H4 future class I,I1,I2,I3,I4 loop

Resume:

1.- Joshua Bengio discusses AI for good projects on contact tracing and drug discovery to fight COVID-19.

2.- AI researchers must understand their work's societal impact and collaborate with experts in other fields.

3.- A wisdom race exists between technological progress and collective wisdom. AI can be used for good or misused by the powerful.

4.- The Covi contact tracing app uses self-reported symptoms to provide early warnings and help people avoid infecting others.

5.- An epidemiological model simulates different contact tracing strategies. Covi outperforms manual and binary digital tracing.

6.- Privacy is a major challenge. Risk messages are quantized to prevent identifying individuals. Behavior recommendations use only 2 bits.

7.- Active learning balances exploring uncertain candidates and exploiting promising ones to efficiently discover new drugs.

8.- Graph neural nets can extend deep learning to molecule structure graphs and protein-drug interaction graphs.

9.- Reinforcement learning can generate drug candidates by making a series of molecule changes to optimize binding affinity reward.

10.- Current drug discovery incentivizes secrecy over sharing data and knowledge. An alternative model closer to academia could better align incentives.

11.- Dario Gil discusses the urgency of science during the pandemic and beyond, to solve complex global problems.

12.- AI is ushering in an era of accelerated scientific discovery, enhancing each step of the scientific method.

13.- AI can be applied to the language of chemistry to guide retrosynthesis planning and enable robotic synthesis of molecules.

14.- In 5 years, AI-driven materials innovations could impact carbon capture, fertilizers, batteries, electronic materials, and antiviral drugs.

15.- Quantum computers are needed to efficiently simulate quantum mechanical systems in chemistry and physics, as classical computers scale exponentially.

16.- Superconducting qubits allow exquisite control to create quantum computers. IBM aims to build a 1000+ qubit machine by 2023.

17.- Quantum circuits will be embedded into classical programs and executed on quantum hardware via the cloud.

18.- A fast-growing quantum computing community is exploring applications in science, research, education and software using IBM's quantum cloud service.

19.- Scaling up quantum computers over the next decade could enable a new paradigm of intelligent quantum simulation and discovery.

20.- Accelerating science also requires communities of discovery - collaborative, multi-disciplinary efforts between academia, industry, and government.

21.- The COVID-19 HPC Consortium exemplifies a community of discovery - aggregating supercomputing power and matching it with research proposals.

22.- The consortium has grown to 85+ projects across therapeutics, vaccines, virology and epidemiology, enabled by 600 petaflops of computing power.

23.- We need institutions like a "scientific readiness reserve" to proactively mobilize and coordinate scientific talent to solve critical challenges.

24.- Diversity of innovation models, from individual to institutional to community-oriented, provides a competitive advantage to solve hard problems.

25.- AI and nanotechnology are being explored to develop COVID-19 treatments as potential alternatives to vaccines, though vaccines are still promising.

26.- Indigenous communities' traditional knowledge could inform computational drug discovery when combined with AI simulation and validation.

27.- Scientific communities must close the loop between discovering solutions and implementing them to solve real-world problems.

28.- Technology alone is insufficient - human scientific creativity and collaboration between problem owners and solvers is key.

29.- Solving problems leads to new questions and discoveries, creating a virtuous cycle between fundamental research and applications.

30.- We need a new social contract and support mechanisms for science to rise to the urgency of our greatest challenges.

Knowledge Vault built byDavid Vivancos 2024