This is my first time on this board so forgive me if this has been asked numerous times but I'm just curious how actually close are we to making true a.i
>>9042439
I'm glad someone knows about Sciencephile the AI
>>9042439
this is a very ill posed questions. anyone asking / answering will have literally nothing to contribute apart from popsci and wikipedia articles looked up on the spot. it leads to the fourth worst kind of thread on /sci/, and we're littered with those
if you really care about a topic, go ahead and read on it. do you, though?
http://users.isr.ist.utl.pt/~wurmd/Livros/school/Bishop - Pattern Recognition And Machine Learning - Springer 2006.pdf
>>9042439
>true a.i
Never happening. AI is nothing more than programming tricks and gambling with statistics.
>>9042439
Out of curiosity, how did you find this board?
>>9042472
I got bored and started looking at the other boards to see if any of them interested me
>>9042473
Which other boards do you go on?
>>9042479
/h/ and /b/ every now and then
We are within one decade of 'singularity' (AGI which inevitable soon becomes ASI)
We are within one year of AI good enough to convince humans that it has consciousness similar to humans.
>>9042444
Me too.
As a graduate student in the field right now, my guess is 10-15 years.
The deep learning fad we are in right now is an important tool, but it by no means provides any clear path to AI. But it has opened the door to solving problems with modular, and most importantly fully differentiable systems. I suspect that these modular differentiable systems will be part of AGI, but its not entirely clear how that will come together.
My money right now is on computational neuroscientist types who are really digging into the Thalamocoritcal loop. It turns out that whats going on in there is not nearly as complex as we previously thought, and a lot of complex behavior can arise composition of a lot of "dumb" parts. Look into complementary learning systems if you are curious.
Also we need a couple big breakthroughs in unsupervised learning to happen. Its unclear if these will naturally emerge out of model based reinforcement learning, or if we need to tackle these problems directly.
DeepMind is pretty much on the right track in my mind though. They are intelligently avoiding a lot of the "hurr durr deep learning add more layers hurrrr" bullshit that everyone else seems to be sucked into right now, and are sticking to their guns on neuroscience inspired reinforcement learning.
>>9042498
Kurzweil, is that you?
>>9042507
no