2015 Aug ANISSIMOV Notes
<p>Aug 28, 2015 12:46 PM</p>
<p>
Part1 Singularity</p>
<p>
Vernor Vinge.. <b>smarter than human intel.</b>..</p>
<p>
cannot predict the future...</p>
<p>
all the essential properties: <b> model the world, make predictions</b>,
formulate beliefs</p>
<p>
put a lot of force, intel and energy behind its goals.</p>
<p>
superintel would use tech.</p>
<p>
Singularity is far from inevitable.</p>
<p>
*</p>
<p>
Part 2: Singularity Not a New Idea</p>
<p>
Part 3: First ref to Singular</p>
<p>
1935... The Machine by John Campbell..</p>
<p>
Pt 4: Sing Crawls into Mainstream...</p>
<p>
Glenn
Reynolds... break thru the human ceiling.</p>
<p>
nobody can bench press 20 tons or run a mile in a minute.</p>
<p>
A superintel might be able to walk into a lab and point out 10 difft things the scientists
had never thot of...</p>
<p>
A superintel could teach the ancient Babylonians how to build a nuke.</p>
<p>
FIVE Mitchell Howe The Greatest Leap... excellent...</p>
<p>
The Wright Brothers...</p>
<p>
galaxy itself opened to our exploration.
Nanotech.</p>
<p>
since the rise of civiliz we have seen no appreciable incr in our intel.... NO wrong.</p>
<p>
but... fully empowered society...</p>
<p>
Mitchell Howe Evolution by Choice. 6</p>
<p>
minds able to recognize beauty...</p>
<p>
break many eggs to create an omelet.</p>
<p>
loc 273 AGI could provide enormous assist... self-reinforcing cycle.</p>
<p>
suffering of millions for potentially curable disease.</p>
<p>
Investing in Immortality Mitchell Howe 7.. loc 294.</p>
<p>
immortality is not just more broken hips.</p>
<p>
MIRI machine intel research institue.
goal... friendly AI.</p>
<p>
The Odds Mitch Howe.</p>
<p>
Turing Test by 2029... Kurz vs Kapor.</p>
<p>
quotes K vs K.</p>
<p>
first AI must be engr to share human values.</p>
<p>
Mitch Howe:</p>
<p>
1969 Icebreaker Manhattan first vessel to nav. NW Passage.
loc 443: such minds may be able to visualize molec bio or socioecon
effortlessly.</p>
<p>
want friendly AI on our side..</p>
<p>
<b>Mitch Howe Rapids of Progress</b>. 10</p>
<p>
loc 462... always be a few who would destroy civil if they had the capacity.</p>
<p>
dangerous swiftness.</p>
<p>
danger far greater with genet engr and nanotech.
custom designed plague...</p>
<p>
AGI must CARE about human.</p>
<p>
11 AI and Effective Sagacity</p>
<p>
key supergoal is <b>learning and improving to eventual goal </b>of signif action.</p>
<p>
FIDGET and SAGE</p>
<p>
Fidget: small, trivial decisions eg
card shuffling; assembly line; brush teeth...</p>
<p>
537 ... SAGE: consc thought and complex decisions.</p>
<p>
Solitaire: thoughtless exercise of mouse clicks under FIDGET control.</p>
<p>
humans: v little time with SAGE hard at work.</p>
<p>
high-level SAGE thought => learning, creativity, experiment.</p>
<p>
preoccupation with bio drives in humans.</p>
<p>
AI... free of bioburdens.</p>
<p>
* Intel Augmentation vs AI Michael Anissimov</p>
<p>
<b>thinks AI will reach Sing first.</b></p>
<p>
brain + computer... channels are limited.
and reprogram PFC needs neurosrug + neurosci.</p>
<p>
cannot experiment on human brain...</p>
<p>
brain... not designed to be upgraded...</p>
<p>
benefits of AI:</p>
<p>
rapid design and test;</p>
<p>
power expands with money</p>
<p>
can train in virtual world</p>
<p>
optimal inference...</p>
<p>
nascent AIs share info...</p>
<p>
indef copies.</p>
<p>
no legal/ ethical concerns.</p>
<p>
*</p>
<p>
<b>MAX TEGMARK on singularity</b>... the key essay</p>
<p>
see loc 762...</p>
<p>
unshackled by the limits of our human bodies... such advanced life could rise up
and inhabit our observable universe.</p>
<p>
Q is not can it happen but if it WILL HAPPEN.</p>
<p>
Max guess: we're the ONLY life form in our entire observable U that has advanced
to building telescopes...</p>
<p>
our centry: the the most significant one in the history of our universe...
when the future gets decided.</p>
<p>
situations is SO UNSTABLE that I doube we can dwell here for another century.</p>
<p>
the life route rather than the death route.</p>
<p>
</p>
<p>
14 Relative Advantages of AI and Human Brains... Michael A...</p>
<p>
more design freedom, ease of modif;
debug, backup;
(lack human probs: larck of focus, energy. etc)</p>
<p>
extended tasks at > human speed 10M to one speed avtange.</p>
<p>
24 ours a day * 365 days.</p>
<p>
brain does not phys integrate...
non-existence of boredom.</p>
<p>
<b>brain over AI: part 14...</b>
AIs lack AGI and real-world experience.</p>
<p>
Minds in general</p>
<p>
lack of bad hueristics
unneeded sex reproduc; unnec autotomic processes.</p>
<p>
add more comput power.</p>
<p>
ability to control what is and what is not remembered.</p>
<p>
learn rapidly...</p>
<p>
ability to consc create, analyze, etc.</p>
<p>
loc 839... an importnat list of computer AI advantages...</p>
<p>
Part 15 Why is AI dangerous...</p>
<p>
only some goals sets place high priority on human beigns and their worth.</p>
<p>
hard to program an AI tat values us and our goals.</p>
<p>
specify... in math detail... friendly AI.. ridiculous....</p>
<p>
list of AI applications</p>
<p>
wargaming</p>
<p>
lazw enforcement</p>
<p>
interstate politics</p>
<p>
finance</p>
<p>
combat robots.</p>
<p>
Why these AIs are dangerous.. lack of bkgrnd knowledge and humanness.</p>
<p>
you can't drive an OUGHT from an IS...</p>
<p>
must be built from the ground up to incorp morality.</p>
<p>
freedom, empathy, consensus-builidng, conflict resollution, aesthetics, camraderies and rapport
are built into evry human.</p>
<p>
It's that difficulty gap btwn stripped down general intel and morally sophisticated AGI...</p>
<p>
</p>
<p>
Other Kinds of Minds 16...</p>
<p>
clock speed.</p>
<p>
communic</p>
<p>
reprogrammability.</p>
<p>
intel
morality... goal orientation...</p>
<p>
17 Benefits of a Singularity</p>
<p>
invention of molecular nanotech... superior manufac</p>
<p>
reduce violence to zero...</p>
<p>
rearrange government to work better.</p>
<p>
alteration of objects in the Univ...</p>
<p>
computronium.</p>
<p>
tranform willing human into superintel...</p>
<p>
preserv human societies.</p>
<p>
dispell mystery of human itneractions usng xray glasses.</p>
<p>
elim suffering..</p>
<p>
higher aesthetics...</p>
<p>
entire space of thought...</p>
<p>
end the orgy of killing and torture...</p>
<p>
18.. Radical Discontinuity Does not follow from Hard Takeoff...</p>
<p>
v quickly... emergence may be sudden...</p>
<p>
loc 1113 advantages of a human-similar AI...</p>
<p>
more design freedom, ease of modification; copies, debug;</p>
<p>
no fatigue, lack of energy, focus, attention, mistakes, etc....</p>
<p>
24 hours per day</p>
<p>
brain does not physcially integrate well; no boredom;</p>
<p>
incr ability to acquire info from the internet...</p>
<p>
lack of human failings; biases; rationalizations...</p>
<p>
lack of neurobio limits;</p>
<p>
lack of complexity acquired from evolut. design...
unneces autonomic proc and sex reproduction</p>
<p>
the ability to advance on the design of evolution; unconstrained by pre-existing design...</p>
<p>
add more comput power as needed for a partic problem...</p>
<p>
can analyze and modify every design level and feature...</p>
<p>
control what and is not remembered/ learned;</p>
<p>
share concepts, abilities, memories fast...</p>
<p>
ability to self-observe and understand on a fine-grained level...</p>
<p>
mind can recursively self-improve...</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
19 Singularity is NOT unknowable...</p>
<p>
smarter-than-human intel.</p>
<p>
</p>
<p>
20 <b>CORY DOCTOROW on the Singularity</b>...</p>
<p>
most popular blog on the internet Boing Boing.</p>
<p>
This would have been better as adialogue... this is just a rebuttal.</p>
<p>
21 Answering Pop Science 10 Qs on Singularity</p>
<p>
hypothet Qs directed to an AI.</p>
<p>
Q is there just <b>one kind of intel</b>...</p>
<p>
A Ben Goetzel: achieving complex goals in complex environs...</p>
<p>
Howard Gardner... multiple intel</p>
<p>
but must be correlated with g the genl intel factor. unsupported</p>
<p>
IQ and other g-weighted measures blow away multiple intel.</p>
<p>
g is sufficient.</p>
<p>
Q how would u kill all of us...</p>
<p>
A human level intel on the net can distrib itself.</p>
<p>
like the mythical Hydra... grow back 10 heads.</p>
<p>
get manipulators, build robot arms...</p>
<p>
physical manipulators that can manufac raw materials into MORE physical manipulators.</p>
<p>
ASK, bribe or deceive human beings (or just HIRE them)...</p>
<p>
build added nanoscale arms.</p>
<p>
use neutron bombs to kill with radiation.</p>
<p>
why not pull the plug; rivate servers;</p>
<p>
hostile AI would look like a Friendly AI until it passes the thethreshold of power.</p>
<p>
AI and the SENSES: loc 1479;</p>
<p>
NO would not wake up with NO senses. It nees rich senses in which to soak up data;
make predictions and pursue goals.</p>
<p>
FEELINGS... ... might choose to feel or not... ridiculous.</p>
<p>
Consc is a red herring he says.</p>
<p>
Emotions will not exist; NO; need them to understand humans; empathisize, etc.
(for programs such as advisors; confidantes)</p>
<p>
Free will is a red herring; yes...</p>
<p>
when you're as dumb as humans everything is a surprise... loc 1572...
it does not mean reality is inherently non-deterministic.</p>
<p>
we enjoy... sex, status, fun, knowledge, relaxation.</p>
<p>
preventing SUBGOAL STOMP loc 1601 and goal drift is
a big tech challenge.</p>
<p>
Knowability of Friendly AI...(Yudkowsky)</p>
<p>
is the AI really on your side.... if you want someone to love you forever, get a dog.</p>
<p>
<b>22 Illusion of Control in IA Sing.</b></p>
<p>
from what I understand... Sing outweighs all other concerns NO!!!</p>
<p>
winner take all effect: AI facilitated dictator (Kim Jong Un)
or human-indifferent AI.</p>
<p>
humans 2 flavors: IA bootstrap and WB emulation.</p>
<p>
will humans or AI reach it first... all signs point to AI...</p>
<p>
we need a living entity; </p>
<p>
human morality: complex network of competing drives...</p>
<p>
we need an agent that shows vast wisdom, benevolence, harmony...
not a human dictator.</p>
<p>
humans innate temptation to abuse power...
we are programmed for conflict;</p>
<p>
the entire human race would answer to the insanity of over overpowered being.... Kim Jong Un.</p>
<p>
TRUST the spirit of benevolence.... see Yudkowsky ref.</p>
<p>
with AIs you can experiemnt.</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
Part Two: Transhumanism</p>
<p>
chap 23: 6 defns:</p>
<p>
Anders Sandberg: life extension, cryonics, MT, enhancements, uploading</p>
<p>
</p>
<p>
24 Ten Transhumanist Upgrades Everyone will want...</p>
<p>
10 disease immunity...artificial antibodies...(already have monoclonal antibodies)</p>
<p>
9 artificial retinas...</p>
<p>
8 BCI telepathy</p>
<p>
</p>
<p>
25 Top Ten Transhumanist Techs...</p>
<p>
10 cryonics... high fidel preserv of human body/ brain...</p>
<p>
9 virtual reality VR...</p>
<p>
gaming.... Metal Gear Solid V...</p>
<p>
5% of all broadband internet uses play games.</p>
<p>
8 gene rx RNA interfer... indef lifespans.</p>
<p>
7 space colonization...</p>
<p>
6 augmented humans</p>
<p>
5 see NASA study Advanced Automation for Space Missions...</p>
<p>
4 Molec manufac</p>
<p>
3 megascale engr..</p>
<p>
1 AGI noetic Renaissamce...</p>
<p>
the expansions of intel and exper beyond our wildest dreams.</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
26 Dream Machine Brain computer interace for manip dreams...</p>
<p>
EMOTIV systems...</p>
<p>
record all your dreams in entirety.
distributed nanobots.</p>
<p>
2011 reconstruct movie trailers by watching brain activity...</p>
<p>
</p>
<p>
27 How Can I incorp Transhumanism into my daily life?</p>
<p>
rationalism in debate and analy...</p>
<p>
</p>
<p>
Part 3: Nanotech...</p>
<p>
29 What will the first nanotec proudcts be?</p>
<p>
carbon only?</p>
<p>
group 1: walls, beams, domes...</p>
<p>
group 2 flywheels, turbines, solar plants..</p>
<p>
group 3: trains , planes, ships, cars...</p>
<p>
</p>
<p>
</p>
<p>
30 Nanofactory Regulation</p>
<p>
PNs personal nanofactories... safety authority.</p>
<p>
PART FOUR Future tech: 31 Nuclear; 32 Nanotech</p>
<p>
32 Replicating Machines:</p>
<p>
prob os seeing something = how common it is * how long it lasts.</p>
<p>
long-lived adaptive self-replc systems... humans and trees.</p>
<p>
margin gains to the exponent... bacterium multiplies.</p>
<p>
flexible diets... raccoons, insects...</p>
<p>
make nanobot that can get energy from sun.</p>
<p>
patterns... like computer viruses...</p>
<p>
Rob Freitas and Ralph Merkle kinemtic self-replic.</p>
<p>
indium gal nitride systems.... 70% effic.</p>
<p>
matches the solar spectrum.</p>
<p>
700 X more effic thanplants.</p>
<p>
bio sys limited by dozens of constains.</p>
<p>
compete for resources.</p>
<p>
bio mol machines use slow thermodynamics and diffusion.
drag of water.</p>
<p>
</p>
<p>
31 Nuclear Reactor in Every Home</p>
<p>
Thorium... no danger; min waste</p>
<p>
but thor =.U233...</p>
<p>
large truck...</p>
<p>
Westinghouse AP1000</p>
<p>
loc 2576 Gen IV reactor will start build in 2030s...
reduce capital, enhanced safety, min gen of waste, and reduc of weapons grade material.</p>
<p>
molten salt reactor idea.</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>
<p>
</p>