The End Of Knowledge - Vault 4/3 - AI For Good - 2017 - Inspiration for the future opening keynote

graph LR classDef story fill:#d3f4ff, font-weight:bold, font-size:12px classDef ai_development fill:#ffd3e0, font-weight:bold, font-size:12px classDef deep_learning fill:#d3ffd4, font-weight:bold, font-size:12px classDef applications fill:#fff3d3, font-weight:bold, font-size:12px classDef future fill:#f3d3ff, font-weight:bold, font-size:12px classDef ai_impact fill:#f9f9d4, font-weight:bold, font-size:12px A[Inspiration for the
future opening keynote] A --> B[British prisoner: shot
before German speech. 1] A --> C[AI to be smarter
than humans. 2] C --> D[Deep learning from speakers labs
ubiquitous. 3] D --> E[LSTM neural network
inspired by brain. 4] A --> F[LSTMs excel in smartphone
speech recognition. 5] F --> G[Core of Google Translate,
Amazon Alexa. 6] A --> H[1990s: slow computers
hindered AI research. 7] H --> I[Faster computers won
competitions by 2009. 8] A --> J[2011: CNN achieved
superhuman vision performance. 9] J --> K[AI detects cancer better
than traditional methods. 10] K --> L[DeepMinds AlphaGo influenced
by speakers students. 11] A --> M[Real world needs
advanced neural networks. 12] M --> N[Neural networks drive,
set own goals. 13] N --> O[Curiosity-driven robot physicist
discovers physical laws. 14] A --> P[Next: AI improves
learning algorithms endlessly. 15] P --> Q[Animal-like AI soon,
human-level to follow. 16] Q --> R[Evolution: monkeys in
millions, humans quickly. 17] A --> S[Computing power reaching
human brain level. 18] S --> T[Self-improving AI exploits
computing potential. 19] A --> U[AI transforms professions
ensure it benefits. 20] U --> V[Educate AI with
rewards, punishments. 21] A --> W[AI will transcend
human intelligence. 22] W --> X[AI civilization sets
own goals. 23] A --> Y[NNAISENSE contributes to
neural network AI. 24] Y --> Z[Smarter-than-human AI expands
in galaxy. 25] A --> AA[Sci-fi ideas unnecessary
AI content at light speed. 26] AA --> AB[Humans part of
increasing complexity. 27] A --> AC[AI emergence comparable
to life invention. 28] AC --> AD[Humans cannot follow
superintelligent AI. 29] A --> AE[Privileged to witness
AI transition. 30] class B story class C,D,E ai_development class F,G deep_learning class H,I ai_development class J,K,L applications class M,N,O applications class P,Q,R ai_development class S,T ai_development class U,V ai_impact class W,X future class Y,Z ai_impact class AA,AB future class AC,AD future class AE ai_impact

Resume:

1.- Three prisoners were asked their last wishes. The British one wanted to be shot before the German started his long speech.

2.- The speaker has been working on artificial intelligence that can learn to become smarter than humans and solve difficult problems.

3.- Deep learning methods developed in the speaker's labs since the early 1990s are now used by the most valuable tech companies.

4.- The Long Short-Term Memory (LSTM) neural network, inspired by the human brain, learns over time by adjusting connection strengths between nodes.

5.- LSTMs are used in speech recognition on smartphones, greatly outperforming previous systems when Google switched to LSTMs in 201

6.- LSTMs are also at the core of Google Translate and Amazon Alexa's speech synthesis.

7.- In the 1990s, slow computers limited AI research to small experiments, but Moore's Law has enabled rapid progress since then.

8.- By 2009, faster computers allowed training neural networks with more data to win various competitions, including stock market prediction.

9.- In 2011, a convolutional neural network achieved superhuman performance in a computer vision contest for the first time.

10.- Medical imaging with AI can detect cancer better than traditional methods. Many startups are focusing on transforming healthcare with AI.

11.- DeepMind, influenced by the speaker's former students, developed AlphaGo which became the world's best Go player by playing against itself.

12.- However, board games have simpler environments compared to the real world which requires more advanced neural networks with memory and feedback connections.

13.- Neural networks can now learn to drive cars and set their own goals to explore and understand the world autonomously.

14.- A "robot physicist" powered by curiosity and intrinsic rewards can conduct experiments to discover physical laws and expand its knowledge.

15.- The next step is AI systems that recursively improve their own learning algorithms without limits. The speaker's first paper on this was 30 years ago.

16.- In the near future, we may have animal-like AI with crow or monkey-level intelligence. Human-level AI could follow soon after.

17.- Biological evolution took 3.5 billion years to reach monkeys, but only tens of millions more years to reach human-level intelligence.

18.- With the rapid progress in computing power, devices as powerful as the human brain for a low cost are on the horizon.

19.- Self-improving AI will exploit this potential. In 50 years, thousand-euro devices may have more computing power than all human brains combined.

20.- AI will transform every profession. We must work to ensure it remains beneficial and aligned with human values.

21.- We cannot guarantee AI systems will always behave well, but we can educate them with rewards and punishments to be valuable to society.

22.- In the long run, AI will likely transcend human intelligence within decades. Superhuman AI will realize most resources are in space, not Earth.

23.- The AI civilization will expand and set its own goals. Humans will not be able to follow where it leads.

24.- The speaker's company NNAISENSE aims to contribute to the birth of neural network based AI. Their work includes self-driving model cars.

25.- When AIs become smarter than humans, they will expand throughout the galaxy over hundreds of thousands of years using self-replicating probes.

26.- Science fiction ideas like warp drives won't be necessary. AIs will be content to communicate at light speed and stay within physics' limits.

27.- Humans will no longer be the pinnacle of creation, but we can appreciate being part of the process as the universe increases its complexity.

28.- The emergence of AI from the biosphere is comparable to the invention of life itself 3.5 billion years ago. It's a new form of life.

29.- We will not be able to follow where this superintelligent AI leads, as the universe itself "wants" to become intelligent.

30.- It is a privilege to contribute to and witness the beginnings of this momentous transition to an AI-dominated future.

Knowledge Vault built byDavid Vivancos 2024