The human pursuit of artificial intelligence | Thore Graepel | TEDxExeter

[@ValerianTexeira]

This is one among the best short lecture that I watched in the TEDx explaining on Artificial Intelligence in simple terms.

[@jessleigh6729]

Such an interesting concept, loved watching this talk again ✨

[@simpsonstephen]

Still loving the animated games of capture the flag!

[@skywalker7589]

Artificial intelligence may be the path to solve complex problems beyond human capability, but there is no telling what will happen to humanity if technology have minds and feelings of their own.

[@terrywilliams9334]

Excellent presentation! I love the comparison between the fictional story of Dr B in The Royal Game and the uncannily similar real success of AlphaZero against Stockfish. The Robert Heinlein quotes were great too.

[@jessicapepperell3061]

This is truly mind blowing. Wow! What an incredible and intelligent man you are! X

[@michaeljelicic4601]

The irony of Atheist. Intelligence begets intelligence. But for some reason non life gave rise to life that evolved into intelligent humans who create intelligence whilst denying a higher intelligence. 🧐

[@karlopeternel7685]

SingularityNET AGI Artificial General Intelligence now on Cardano very possible sooner than ever Quantum computing will accelerate it It will become the first world's supercomputer

[@sdmarlow3926]

.. but you do feed it rules and instructions when design the networks architecture, and use human intelligence to classify the data used in "training." And ML doesn't really learn, as in, shaping how it thinks. Think about the different objects we use to filter, such as deciding when you use a soup strainer, and when to use a spaghetti strainer. Once built, they just do the one thing really well.

[@robosergTV]

they do learn. They start with random actions.

[@sdmarlow3926]

Conditioning and ML are not the same thing. The dog is already operating at a more abstract level than any xNN can provide.

[@patham9]

I agree. Also sticking with Reinforcement Learning will not be the end of the story except for the toy tasks they are solving. And beautiful storytelling like in this talk aside, the biggest issue of DeepMind's approaches to me is the dependence on self-play & need for a perfect simulation, training in millions of games, while I'd argue that another hallmark of human intelligence is clearly also its ability to learn something meaningful and useful from just a few examples, something they do not seem to get. They even celebrate this "massive parallel" training with perfect simulation as something good, it's like selling a broken engine. They showed RL success with Atari games, Go, Capture the Flag, almost certaintly the next story will be just another game. Because their systems do not work in the real world, so it cannot be different. In the real world unforseen situations appear, and an agent often one cannot just gain more training examples about a certain situation, yet has to make meaningful decisions based on its best information available. That's when reasoning becomes necessary.

[@erkn99]

The AI is called Agend that kind of sounds faniliar

[@chris_7711]

This scares me.

[@johnnguyen4053]

Nothing can place the human brain. Even the superpower human intelligence