Jürgen Schmidhuber

<p>Oct 08, 2016 02:10 PM &nbsp;talk in Brussels &nbsp;(Belgium) at ECS &nbsp;10 year &nbsp;anniv.</p>

<p>

&nbsp;

(video only shows &nbsp;L portion oh his slides)</p>

<p>

pub Jul 23 2016; &nbsp;talk in Brussels &nbsp;(Belgium)</p>

<p>

1908 &nbsp;(was it Haber process:) &nbsp;nitrogen extraction... most important invention!

extract nitrogen from thin air.</p>

<p>

1941 &nbsp;Konrad Zuse: &nbsp;first program controlled &nbsp;computer...</p>

<p>

Ukraine: &nbsp;first &nbsp;NN &nbsp;researcher.</p>

<p>

RNN &nbsp;recurrent neural networks... &nbsp;deepest and most powerful &nbsp;NN.</p>

<p>

thru learning and trial and error... they change....</p>

<p>

thru training synapses change...</p>

<p>

long short term memory &nbsp;LSTM &nbsp;developed in his lab in Munich and in Switzerland.</p>

<p>

widely used &nbsp;by Google and Apple;</p>

<p>

smartphone press on Google voice... &nbsp;based on his LSTM &nbsp;since...</p>

<p>

dev by his students... &nbsp;current form in 2006 &nbsp;is on your phone..</p>

<p>

learned &nbsp;from experience.</p>

<p>

in comes speech signals...</p>

<p>

go into input units....</p>

<p>

and outputs letters...</p>

<p>

until 2015 did not work v well...</p>

<p>

then replaced &nbsp;by LSTM... can recognize even in noisy restaurant...</p>

<p>

learns synatx...</p>

<p>

improved &nbsp;NLP by &nbsp;50% !</p>

<p>

pattern recog... kids do it...</p>

<p>

back in Deep Blue... &nbsp;cud not &nbsp;recog. &nbsp;signs</p>

<p>

now computers can... recog &nbsp;...</p>

<p>

at moment is scaling linearly...</p>

<p>

shows &nbsp;&nbsp;2012... winning pattern recog &nbsp;with &nbsp;cancer diagnosis...</p>

<p>

world GDP is 80 trillion; &nbsp;700B per year for health &nbsp;(world)</p>

<p>

shows his competition...</p>

<p>

hears of elephants &nbsp;walking across a dry grass field...</p>

<p>

LSTM &nbsp;outputs this!... &nbsp;no teacher... just lots of training examples...</p>

<p>

he jokes... one person in the audience... &nbsp;"i'm th enexty &nbsp;speaker"</p>

<p>

Deep Go... &nbsp;Deep Mind...</p>

<p>

heavily infl by his students...</p>

<p>

Google.. place better ads...</p>

<p>

taking half the ad. biz of the world! &nbsp;(Google and FB)</p>

<p>

2006 &nbsp;control &nbsp;surgery robots... to &nbsp;tie knots in artif pigs...</p>

<p>

creativity &nbsp;has a theory... &nbsp;has simple &nbsp;AI to do it...</p>

<p>

Stollenga &nbsp;AAAI 2013...</p>

<p>

how does gravity work &nbsp;if I push from the table...</p>

<p>

he thinks we will have something like a little monkey &nbsp;(say within a couple of decades)

like a capuchin or a crow... &nbsp;then &nbsp;moving toward human &nbsp;won't take that long.</p>

<p>

tech evol... is 1M X faster</p>

<p>

NNAIsence.... &nbsp;our company...</p>

<p>

eventually &nbsp;AIs will be smarter than humans...</p>

<p>

they will not stick to the biosphere... they will take off for

resources... they will move out there...</p>

<p>

Bs of self replic. &nbsp;robots...</p>

<p>

space &nbsp;is friendly to robots.</p>

<p>

within &nbsp;lightspeed... &nbsp;&nbsp;AIs &nbsp;will travel by lightspeed by radio...</p>

<p>

a new step &nbsp;on the path of Universe.</p>

<p>

last step &nbsp;3.5B years ago with invention of life!</p>

<p>

privilege to be part of that!</p>

<p>

it is not US vs THEM...</p>

<p>

view all of us... one grand scheme from &nbsp;simple to more complex.</p>

<p>

&nbsp;

(his view agrees completely with mine)</p>

<p>

Q &nbsp;what can we already see...</p>

<p>

A &nbsp;pure pattern recog</p>

<p>

but soon will see v sophisticated &nbsp;robots</p>

<p>

including &nbsp;strawberry plicking...no robot can do that!</p>

<p>

also &nbsp;get &nbsp;a stream of data that you (the AI) &nbsp;directs...</p>

<p>

stock market...</p>

<p>

what do you want to do next...</p>

<p>

(who is the blonde moderator... asking Qs at end)</p>

<p>

Q &nbsp;will &nbsp;wordy &nbsp;professionals be &nbsp;replaced...</p>

<p>

A &nbsp;perhaps &nbsp;taxi drivers will be replaced...</p>

<p>

Homo Ludens... the playing man...</p>

<p>

hard to predict &nbsp;new jobs...</p>

<p>

bloggers, YouTubers</p>

<p>

Usain Bolt... races against &nbsp;other humans and makes &nbsp;millions...</p>

<p>

Q &nbsp;comarison to teaching machines and parents...</p>

<p>

might parents teach the wrong things...</p>

<p>

how abt &nbsp;self driving land mines...

how &nbsp;about &nbsp;teaching &nbsp;kids &nbsp;to do bad things... &nbsp;</p>

<p>

more research in AI &nbsp;is devoted to making humans beter</p>

<p>

health care; &nbsp;better friend;</p>

<p>

Q &nbsp;a base salary... &nbsp;(</p>

<p>

A &nbsp;I am not Swiss &nbsp;but I live there</p>

<p>

(1/3 &nbsp;of people think &nbsp;base income is good)</p>

<p>

how abt having robots &nbsp;pay taxes.</p>

<p>

Q &nbsp;prediction &nbsp;abt &nbsp;future computers</p>

<p>

A &nbsp;will robots &nbsp;take over the Earth...</p>

<p>

comment inspired by Schwarzenneger...</p>

<p>

but &nbsp;taken over by beings that share the same &nbsp;goals...</p>

<p>

but we don't quarrel with kangaroos.</p>

<p>

but ants &nbsp;happily live in forest &nbsp;(perhaps &nbsp;no goal conflict...</p>

<p>

same with super smart robots)</p>

<p>

how about &nbsp;2001 &nbsp;movie &nbsp;audience A</p>

<p>

perhaps kill 2B people so thaa the remaining &nbsp;1B can survivie...

fighting over land and resources...</p>

<p>

J SH... &nbsp;always about &nbsp;goal conflict...as with Hitler</p>

<p>

no interest &nbsp;in fighting &nbsp;if difft goals</p>

<p>

don't have to worry &nbsp;if totally &nbsp;difft goals...</p>

<p>

like Matrix &nbsp;silly goal conflict</p>

<p>

(brain perhaps &nbsp;takes &nbsp;20 &nbsp;(he asy 30) watts &nbsp;but

powerplant for human takes &nbsp;kwatts.</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>