GOTO Conferences
This presentation was recorded at GOTO Chicago 2019. #GOTOcon #GOTOchgo
http://gotochgo.com
Danny Lange – VP of AI and ML at Unity Technologies and previously led innovative ML teams at Uber, AWS and Microsoft
ABSTRACT
Danny Lange will demonstrate why a game engine is the perfect virtual biodome for AI’s evolution. Attendees will recognize how the scale and speed of simulations is changing the game of AI while learning about new developments in reinforcement learning […]
Download slides and read the full abstract here:
https://gotochgo.com/2019/sessions/719/on-the-road-to-artificial-general-intelligence
https://twitter.com/GOTOchgo
https://www.linkedin.com/company/goto-
https://www.facebook.com/GOTOConferences
#AI #ArtificialGeneralIntelligence #ArtificialIntelligence #ML #MachineLearning
Looking for a unique learning experience?
Attend the next GOTO Conference near you! Get your ticket at http://gotocon.com
SUBSCRIBE TO OUR CHANNEL – new videos posted almost daily.
https://www.youtube.com/user/GotoConferences/?sub_confirmation=1
Source
agi.ai
Absolutely incredible
yeah? yeah? yeah? yeah? yeah? yeah? yeah? Damnnn chill bruh.
Wow, he really 'get' it! Great!
AGI is here. People just don't realize that yet. ANN can learn anything and any incremental tweaks made to it is just a distraction from the fact that we are teaching AI wrong.
AGI needs its own curriculum. Know this and discover your AGI.
Yeah yeah yeah
This is marvellous. I love artificial intelligence. It’s good for human kind, not negative
The whole video is worth watching.
Most interesting timings:
2:30 Siri and Alexa, Amazon and Netflix recommendations, Facebook Feed are not AI, it's just a good software, written by talented people.
5:50 If you want to build an AI for Nature, you could use a game engine with physics simulation like Unity (or Unreal) to train it and scale it.
21:40 When the initial problem has very sparse reward space, you can't solve it with random exploration algorithms only. That's because you have almost no chance of hitting the reward and so you have no data to learn from. Counter-intuitively to solve it you need to do exactly the opposite of what Machine Learning does – instead of minimizing the error you maximize it. That's how you effectively explore the space of choices. And that's exactly how curiosity works in nature. Now think how the Amazon recommendation system could use that principle to explore you.
29:20 Unity built a game with a prize of 100000 dollars to win. Only problem is it's for humans. It's a 100-floor tower with procedural generated levels that get harder and harder. The challenge is to build an ML-agent that learns how to solve it. Search for "Obstacle Tower Challenge".
This presentation seems a bit like an ad
When you start your presentation with the infrastructure of intelligence as the byproduct evolution, then achieving artificial general intelligence is already off its mark. Because your goal is already hampered by your preconceived parameters of reality solely based on evolution.
He starts out by highlighting the inadequacy of ANI's, which I completely agree with, but then attempts to sell reinforcement learning as the foundation of AGI. AGI is so much more than RL; RL is the "safe" step for ANI researchers that want to sound like they know something about AGI architecture. True AGI will not manifest in these little TShirt stage presentations, nor in these deep RL houses.
Can't wait for next century