SciShow
AI can do some pretty amazing things, but if we want it to learn on its own, we’re going to have to teach AI how to evolve.
Hosted by: Olivia Gordon
SciShow has a spinoff podcast! It’s called SciShow Tangents. Check it out at https://www.scishowtangents.org
———-
Support SciShow by becoming a patron on Patreon: https://www.patreon.com/scishow
———-
Huge thanks go to the following Patreon supporters for helping us keep SciShow free for everyone forever:
Adam Brainard, Greg, Alex Hackman. Sam Lutfi, D.A. Noe, الخليفي سلطان, Piya Shedden, KatieMarie Magnone, Scott Satovsky Jr, Charles Southerland, Patrick D. Ashmore, charles george, Kevin Bealer, Chris Peters
———-
Looking for SciShow elsewhere on the internet?
Facebook: http://www.facebook.com/scishow
Twitter: http://www.twitter.com/scishow
Tumblr: http://scishow.tumblr.com
Instagram: http://instagram.com/thescishow
———-
Sources:
https://www.nytimes.com/2018/12/26/science/chess-artificial-intelligence.html
https://www.wired.com/2017/01/mystery-ai-just-crushed-best-human-players-poker/
https://qz.com/636637/the-beginning-of-the-end-googles-ai-has-beaten-a-top-human-player-at-the-complex-game-of-go/
https://www.techrepublic.com/article/ibm-watson-the-inside-story-of-how-the-jeopardy-winning-supercomputer-was-born-and-what-it-wants-to-do-next/
https://www.forbes.com/sites/bruceupbin/2013/02/08/ibms-watson-gets-its-first-piece-of-business-in-healthcare/#4920c68c5402
https://medium.com/@froger_mcs/moravecs-paradox-c79bf638103f
https://books.google.com/books?hl=en&lr=&id=56mb7XuSx3QC&oi=fnd&pg=PA1&dq=hans+moravec+mind+children&ots=ijpLMTHIxG&sig=7Dv1MIrzMAbtQ4zllzD5NMd6ET0#v=onepage&q&f=false
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3368062/
https://www.nature.com/articles/nature03687
https://www.cell.com/neuron/fulltext/S0896-6273(15)00559-0
https://hackaday.com/2017/06/23/what-if-learning-new-things-made-you-forget-the-old/
https://arxiv.org/abs/1612.00796
https://rylanschaeffer.github.io/content/research/overcoming_catastrophic_forgetting/main.html
https://www.pnas.org/content/115/44/E10467.short
https://arxiv.org/abs/1611.05763
https://medium.com/explore-artificial-intelligence/an-introduction-to-recurrent-neural-networks-72c97bf0912
http://web.mst.edu/~gosavia/neural_networks_RL.pdf
https://deepmind.com/blog/alphago-zero-learning-scratch/
https://arxiv.org/abs/1611.05763
https://deepmind.com/blog/prefrontal-cortex-meta-reinforcement-learning-system/
https://arxiv.org/abs/1312.5602
https://arxiv.org/abs/1705.05363
https://www.quantamagazine.org/clever-machines-learn-how-to-be-curious-20170919/
http://robotics.sciencemag.org/content/4/26/eaau9354
https://www.quantamagazine.org/clever-machines-learn-how-to-be-curious-20170919/
Source
Cat-astrophic forgetting lol
All students would love having an AI doing their assignments.. Specially long ones
@SciShow the Go stones are misaligned on the Go board – each piece should be placed on an intersection
Saia I know you're reading this- we're going with 337
8:00 "In 2017 Berkley managed to [use machine learning] with super Mario Brothers.
Sethbling, here on YouTube did it a year earlier. Search for "Mari/o"
Isn't intelligence defined by the ability to learn? So if AI can learn by itself, wouldn't that solve the issue?
Is this the original Touring-defined AI, or the new fangled definition that means "the latest software"?
So we're teaching them how to overtake us?
We are gonna screw ourselves without even the courtesy of a reach around.
The moment AI learns to evolve probably will be the end for us 😀
It can if you trade it while it's holding an Upgrade.
Much of the reason we want computers is because they CAN remember what happened to object x three months ago.
Any sufficiently advanced biology is indistinguishable from magic.
Awesome. I love when SciShow covers tech.
the robot arm that learned by itself is terrifying.
Good delivery, much improved
Catastrophic Forgetting… Me, before taking any test.
You dont really give AI guidelines you simply choose a specific architecture for a specific problem. If youre training neural networks that need to recognize visual cues you will choose convolutional neural network for learning to make decisions you might choose actor-critic architecture or deep q-learning.
These architectures are kinda like modes of transportation. You wouldnt use a boat on a highway or a car in the ocean. You need to pick the right tool.
You will then most likely cheat and use whole bunch of preprocessing to make the problem easier for the AI.
But the beauty of AI is that once you do all that it will learn everything else from scratch. It will learn to recognize eyes and "dog" features. You will not guide it in any way. It will just learn it on its own.
Why don't they mimic humans more by allowing the AI to go into a sleep mode. And analyze/process everything it's learned for that day while sleeping. Maybe AI would learn to dream?
implying go and chess have been around only for ''decades''
That's why I like Chappie
It basically was an impressionable kid learning how the world works
Humans seemed to programmed to self-destruct. Ruining our one and only planet for convenience and wealth. Trying to create machines to replace people. Don't pretend you are surprised when the end comes.
“Tens of thousands of years”? More like millions!
Evolve just mean change. To doubt that things can change is just obtuse.
Equal parts fascinating and terrifying
For the love of god, please don't let AI play CoD online, or else AI will become a massively racist edgelord douche nozzle.
Last time on Battlestar Galactica……
We have differing reinforcement loops with different strengths and levels of strength.
Grow
Replicate
Learn
Intuition – pattern recognition and prediction without image or emotion
Evolving senses, touch, taste, smell, hear, see
Development of more predictive feedback loops with the ability to predict independently and vote
Instincts – Fight, flight, freeze
Emotions – fear, disgust, joy, sadness, anger
General intelligence – prediction with simulated realities from varying perspectives with the ability to vote on action.
What they're trying to do is teach a digital device to mimic analog computing, kind of like trying to force a square peg in a round hole. It can be done, but the results are never pretty
Since most games detect Bots and ban them, I need a cyborg to sit at my keyboard and get me past the hard stuff. Will we have cyborgs sitting at consoles all across the world competing with other cyborg and evolving at infinite speed. Talk about lag. Maybe thats how to combat out of control cyborgs, challenge them to a video game duel……………