Hacker News new | ask | show | jobs
by ppod 4424 days ago
>Anybody who has actually, or is currently, studying AI will know that there are fundamental differences in programming "AI" and some self aware magic computation which is almost entirely unfathomable. With current knowledge, at least, the only way to have any sort of learning is to design and implement algorithms to do so.

I think this is way too strong a statement. I have both studied and worked in AI for over 10 years and I definitely don't agree. Human beings are just learning machines too. Algorithms that both learn and learn how to improve their own learning already exist. In my opinion the major bottleneck at the moment is robotics - we have algorithms that can drive a car or win at jeopardy, but none that can regularly throw and catch a baseball in an open environment.
2 comments
mtrimpe 4424 days ago
Isn't the true bottleneck meaning though? As in knowing which goal to optimize for?

I believe that it's essentially an educational problem; where AI will be bound by it's ability to educate us and vice versa.

link
moe 4423 days ago
but none that can regularly throw and catch a baseball in an open environment.

Excuse me?

https://www.youtube.com/watch?v=pp89tTDxXuI

link
sampo 4423 days ago
An argument can be made, that is not an open environment, because (1) they use external cameras that have the whole room in their field of view, and (2) the room is totally white.

They could not do the same out in a park.

link
moe 4423 days ago
(1) they use external cameras that have the whole room in their field of view

http://en.wikipedia.org/wiki/Gorgon_Stare

(2) the room is totally white

http://www.irisa.fr/lagadic/pdf/2011_iros_teuliere.pdf

They could not do the same out in a park.

They (not ETH Zurich) can autonomously identify, track and chase targets in urban areas, without GPS. I'm almost sure juggling a ping-pong ball in a park would not be entirely out of the question.

link
ppod 4422 days ago
I'm well aware of the robotic research and I chose my example carefully. The extra weight of a baseball, the grasping motion necessary, and issues of bounce and spin on an uneven surface makes it a more difficult problem than the pole balancing or the ping ping quadrocoptors.
link