How close are we to building a real Westworld A look at the hardware software and social impact_transcript.csv

text,start,duration
morning everybody David Shapiro here,0.9,5.7
with a video so today we're going to,3.179,6.721
talk about Westworld and AI excuse me,6.6,5.999
there are a couple of aspects of,9.9,4.92
Westworld that are really compelling,12.599,4.381
so one there's the hardware and software,14.82,5.4
but two there's the uh our Hardware,16.98,5.34
software number three is the,20.22,4.02
implications for Humanity which the,22.32,4.5
themes are explored throughout the show,24.24,5.279
so for those who are not familiar with,26.82,4.199
Westworld,29.519,4.261
um it's a HBO show where the premise is,31.019,5.461
basically a theme park where all of the,33.78,4.939
all of the hosts all of the characters,36.48,5.46
are robots so they're essentially NPCs,38.719,5.081
non-playable characters,41.94,3.66
um so but it's a real life video game,43.8,4.38
where you get to go on adventures and,45.6,5.1
many of the characters have stories and,48.18,4.08
quests that they'll give you and they'll,50.7,4.74
take you on you know raids and violence,52.26,6.0
and sex and all sorts of fun exciting,55.44,3.72
stuff,58.26,4.74
and throughout the show they explore,59.16,5.88
um you know what does it mean to be,63.0,4.619
alive to be human uh what does it mean,65.04,4.86
about us that we like violence and and,67.619,4.081
sex on demand and that we do these,69.9,4.5
things uh and then it's all couched in a,71.7,5.58
pretty exciting sexy Adventure,74.4,4.44
um okay so before we jump into the show,77.28,3.42
just a real quick plug,78.84,4.919
um my patreon I made a couple of changes,80.7,5.64
where uh pretty much all tiers well no,83.759,4.261
not pretty much all tiers get you access,86.34,3.54
to the private Discord it's already got,88.02,4.26
200 members as of uh the recording of,89.88,5.52
this video and uh it's a really thriving,92.28,5.94
Community lots of really sharp folks,95.4,5.1
um and yeah so Jump On In,98.22,3.96
um some of my higher most of my higher,100.5,3.9
tiers are sold out right now,102.18,3.72
um so I unfortunately don't have any,104.4,4.38
extra time for uh one-on-one sessions,105.9,5.28
however that will probably change later,108.78,4.619
in May so just keep checking and we'll,111.18,3.979
uh I'll get I'll get to you eventually,113.399,5.4
okay so first let's just talk about the,115.159,6.161
hardware of Westworld,118.799,6.121
so in Westworld this is an example of,121.32,5.399
one of the hosts,124.92,4.559
um and so you see she's in a mostly,126.719,4.26
disassembled State actually doesn't even,129.479,3.9
have most of the internals but I thought,130.979,4.081
this was a good a good place to start,133.379,5.161
because the the uh the frame looks like,135.06,5.399
it's carbon fiber which I actually,138.54,3.779
thought would probably be a good,140.459,4.92
material to build a host chassis out of,142.319,4.981
because if you build it out of metal,145.379,4.801
it's going to be too heavy,147.3,4.98
um this is something that's explored in,150.18,4.74
uh movies and shows like Ghost in the,152.28,5.58
Shell uh Terminator where they have a,154.92,5.28
metal chassis even a wolverine,157.86,4.68
he has a metal skeleton and so they're,160.2,3.48
too heavy,162.54,2.699
um but especially if you want something,163.68,3.24
that's going to move like a normal,165.239,3.601
person and then something that's going,166.92,3.84
to feel like a normal person for uh,168.84,4.02
let's say closer encounters,170.76,3.24
um you're gonna want something nice and,172.86,3.42
light and carbon fiber is is pretty,174.0,4.379
light and it's more than strong enough,176.28,5.7
to approximate human bones uh this looks,178.379,5.22
like it's 3D printed,181.98,3.66
there's lots of there's lots of ways to,183.599,3.901
produce something like this and in fact,185.64,4.679
there are you can see even the the hands,187.5,5.879
the linkages the tendons this is I will,190.319,5.64
say this is somewhat realistic,193.379,4.22
um excuse me,195.959,4.14
but let's take a look at some other,197.599,3.941
stuff and I apologize for the low,200.099,3.901
quality I tried to find a higher quality,201.54,5.339
video but this actually came from the,204.0,5.519
Tesla demo day the investor day recently,206.879,5.821
for their Optimus robot and so the,209.519,5.64
Optimus robot here what they did in the,212.7,4.319
in the demo day and you can find the,215.159,3.841
video but they had it approach a table,217.019,5.041
pick up an arm pick up the arm and then,219.0,6.12
carry it across the room uh all entirely,222.06,6.0
autonomously and in this case that is,225.12,4.14
that's actually pretty good because,228.06,2.58
there's a lot of things that you have to,229.26,4.38
do with that one using hands fine motor,230.64,6.84
control is very difficult but then task,233.64,6.84
planning and path uh path optimization,237.48,5.22
that sort of stuff interacting with a,240.48,4.14
dynamic environment all of these are,242.7,3.599
really difficult things,244.62,4.44
to do even just for like wheeled robots,246.299,5.101
but then you add bipedal uh movement,249.06,4.8
that's even more difficult uh then,251.4,3.899
you've got different things like a,253.86,3.36
dynamic center of gravity to contend,255.299,3.84
with and so they are well on their way,257.22,4.859
uh you know it took Boston Dynamics many,259.139,4.741
many years more than a decade to get to,262.079,4.201
where they are right now but Tesla in,263.88,5.46
the space of a year or two has uh,266.28,4.859
I won't say fully caught up because the,269.34,4.2
Boston Dynamics robots are very athletic,271.139,5.28
and very graceful the Tesla bot is,273.54,7.14
anything but graceful but it is the you,276.419,5.701
know it's approaching the right form,280.68,3.48
factor and then if you just assume that,282.12,3.54
this technology is going to get continue,284.16,3.9
to get better over time One battery,285.66,3.72
breakthroughs,288.06,3.12
so battery life is one of the biggest,289.38,5.4
constraints uh in fact the military you,291.18,5.88
know DARPA funded all kinds of robots,294.78,4.5
which I unfortunately forgot to include,297.06,4.919
but so DARPA robots,299.28,4.5
um like the big dog which was a like,301.979,4.261
basically a pack animal,303.78,4.26
um required entirely too much power so,306.24,5.28
it had to have a a gas motor on it and,308.04,5.7
it was too loud uh because you know,311.52,4.02
Special Forces can move silently through,313.74,3.3
the woods but if you've got something,315.54,3.9
with a 25 horsepower motor chugging,317.04,4.14
along behind you you're not going to be,319.44,6.36
stealthy uh and so you know I I many,321.18,6.18
years ago I experimented with some of,325.8,2.82
the stuff myself and one of the things,327.36,2.7
that was recommended was pneumatics,328.62,3.66
actually because you can store a,330.06,4.26
tremendous amount of energy in a like,332.28,4.02
scuba tank,334.32,4.14
um and then you use some of that uh,336.3,3.48
periodically and then you charge it up,338.46,3.36
and it's also silent,339.78,4.919
um so that brings me to Disney so Disney,341.82,6.719
uh has been the world leader in uh,344.699,7.141
animatronics for a long long time and if,348.539,5.281
you haven't seen this this demo uh,351.84,3.72
definitely look it up,353.82,3.36
um but just like search Disney skating,355.56,2.76
bunny,357.18,2.76
um and this little robot it stands about,358.32,3.659
two and a half three feet tall climbs,359.94,3.479
out of a box tumbles out of the box,361.979,4.101
looks around gets up and roller skates,363.419,5.161
uh so that will show you kind of where,366.08,5.559
we're at in terms of animatronics and,368.58,5.399
you see this this body this this set of,371.639,4.221
uh,373.979,4.741
actuators very very simple,375.86,5.92
so MIT actually has a class called under,378.72,5.88
actuated robots so if you're interested,381.78,4.44
at all in robotics I definitely,384.6,3.599
recommend you check it out,386.22,3.84
um I I only took like the first class I,388.199,4.141
was like oh okay I get it as someone who,390.06,3.84
has experimented with like Legos and,392.34,3.96
stuff my entire life I get it,393.9,4.739
um and I wasn't I wasn't there to like,396.3,4.5
get a master's degree in robotics I was,398.639,4.141
just curious about it but under actuated,400.8,4.14
robots which is basically the idea is,402.78,4.5
that you allow the kinetic intelligence,404.94,4.319
of the design to do a lot of the work,407.28,3.66
for you I'm not saying that that's what,409.259,4.801
this one does but it you can get by with,410.94,6.12
very very little in robots in fact um I,414.06,4.919
think it was in the first few minutes of,417.06,3.78
that course of that class under actuated,418.979,5.16
robots they actually show you a uh it's,420.84,6.299
like just the the hips and legs of a,424.139,5.28
chassis that's walking on a treadmill,427.139,4.62
with no Motors or anything just the,429.419,4.62
intrinsic design of it allows it to like,431.759,4.38
swing its leg forward and and it can,434.039,4.38
continue walking indefinitely just,436.139,4.62
through mechanical design and so that's,438.419,4.141
the kind of stuff and of course Disney,440.759,4.081
hires some of the best animatronics and,442.56,4.02
robotics engineers in the world so those,444.84,3.18
are the kind of people working on this,446.58,2.7
stuff,448.02,4.5
uh this was another project from uh from,449.28,5.759
Disney called compliant robotics so this,452.52,5.88
is why I mentioned pneumatics is uh this,455.039,6.301
series you see they've got Kongs on the,458.4,5.82
arms to act as hands but these arms are,461.34,4.68
actually pneumatically driven and,464.22,3.24
they're what's called compliant which,466.02,3.6
means when you think of a robot you,467.46,3.239
think of something that moves very,469.62,2.579
rigidly and it will like kind of fight,470.699,3.78
you but compliance means that if you,472.199,4.861
push it'll push back and so uh or it'll,474.479,4.921
it'll it'll kind of be squishy and so if,477.06,3.9
you look up this video for Disney it's,479.4,4.019
about three years old now I think,480.96,4.2
um but they're compliant robotics one,483.419,4.68
they move silently two they move as,485.16,4.8
quick as you do and three they're,488.099,4.38
compliant so when you when you combine,489.96,4.859
the animatronics the compliant robotics,492.479,6.301
uh under actuated robotics uh Disney is,494.819,7.021
absolutely one of the leaders uh of this,498.78,5.039
so if any company is going to create,501.84,3.72
anything like Westworld it's going to be,503.819,3.301
Disney,505.56,4.079
uh Boston Dynamics that I mentioned a,507.12,6.06
minute ago so they are way ahead in,509.639,6.121
terms of athleticism you know their,513.18,4.799
robot can do standing back flips and,515.76,4.44
barrel rolls and they can climb and like,517.979,4.021
it's more athletic than most humans at,520.2,3.899
this point uh which is actually pretty,522.0,4.38
scary because then you you think about,524.099,4.68
like you know in the movie iRobot where,526.38,4.98
the Nestor class 5 is like a superhuman,528.779,4.921
Droid and then if you have an army of,531.36,3.9
them that turn on you that's actually,533.7,4.86
not good right and I remember about a,535.26,5.699
year ago when Elon Musk was talking,538.56,5.52
about you know the the Tesla bot they,540.959,4.801
said at first it's going to be about a,544.08,3.3
third as strong I think if I'm,545.76,2.82
remembering correctly he said it's going,547.38,2.76
to be about a third of strong as strong,548.58,3.78
as humans and that's actually can be a,550.14,4.92
safety uh safety thing because if you,552.36,4.74
can easily overpower it that's not a big,555.06,2.88
deal,557.1,2.82
but if you've got a 500 pound machine,557.94,4.74
that is eight times stronger than you,559.92,4.56
you're not going to overpower that and,562.68,3.3
if it's also made out of metal it's,564.48,3.72
going to be harder to shut it down uh,565.98,3.9
now I'm not saying this to like cause,568.2,3.12
any concern I'm just kind of pointing,569.88,3.84
out some of the the energy like the the,571.32,4.8
math right the physics of it,573.72,4.739
um oh here I'll get to that one and so,576.12,4.62
like you think okay if you're trying to,578.459,3.481
optimize for something that is,580.74,3.06
aesthetically pleasing and as lifelike,581.94,3.839
as possible you're gonna have to make,583.8,4.8
trade-offs in terms of uh Mass strength,585.779,4.261
and that sort of stuff,588.6,3.78
um so I would actually suspect that a,590.04,3.96
Westworld style host is actually going,592.38,3.66
to be a lot weaker physically weaker,594.0,4.8
than a robot possibly could be and the,596.04,3.96
reason is because you want it to be,598.8,5.039
realistic okay and so then finally uh,600.0,5.82
addressing the elephant in the room yes,603.839,3.541
there are plenty of companies around the,605.82,3.18
world that are making,607.38,5.1
um let's say adult uh toy dolls for,609.0,6.72
intimate purposes uh they have not yet,612.48,5.28
crossed The Uncanny Valley so that is,615.72,4.02
actually the whole point of this was to,617.76,3.54
show that like we've got manual,619.74,3.9
dexterity we've got athleticism we've,621.3,4.26
got miniaturization we've got uh we've,623.64,3.96
got animatronics the only thing that we,625.56,4.56
don't have is real like life-like skin,627.6,4.38
and faces,630.12,3.659
um so that's all that's all the uh raw,631.98,3.66
ingredients we can make robots fast we,633.779,3.721
can make them strong athletic dexterous,635.64,3.9
we've got fine motor control we've got,637.5,3.959
the battery technology which is good,639.54,3.419
enough right now but it's also,641.459,3.961
continuing to improve so the only thing,642.959,3.721
missing is that they're not right,645.42,3.0
they're not yet aesthetically pleasing,646.68,4.26
and so that's going to be true for a,648.42,4.56
while and the reason that I say that,650.94,3.24
it's going to be true for a while is,652.98,4.859
because human skin and muscles are like,654.18,6.06
really really really complex structures,657.839,4.68
your skin again I think if I remember,660.24,5.039
correctly it has seven layers and some,662.519,3.721
of those layers have different,665.279,3.361
consistencies different textures,666.24,4.98
and then once you get past the skin it,668.64,4.22
slides over the underlying tissue,671.22,4.92
there's there's a fluid barriers that,672.86,5.2
allow for for more like lubricated,676.14,3.54
motion so like if you grab your arm and,678.06,3.6
twist it the skin actually can glide,679.68,4.26
over the underlying muscles,681.66,3.84
um then you've got ligaments muscles so,683.94,4.079
you've got all kinds of things to make,685.5,4.68
um that you need to figure out from a,688.019,4.5
materials perspective in order to make,690.18,5.399
an Android body lifelike,692.519,5.82
which is why I think in Westworld the,695.579,4.681
tissue is like semi alive or something I,698.339,4.141
don't remember exactly the details uh,700.26,4.139
but like there's there's the the living,702.48,4.56
tissue or the lifelike tissue over top,704.399,4.801
the robotic skeleton,707.04,3.66
um and I don't think that that's,709.2,3.9
necessarily a good way to go uh because,710.7,5.04
living tissue takes a lot of energy,713.1,4.5
um and then you if you have living,715.74,3.24
tissue then you have to deal with like,717.6,4.56
immune systems and genetics and that,718.98,4.68
gets real complicated and then you're,722.16,4.2
basically building a borg anyways,723.66,5.28
um okay so that's for the hardware the,726.36,4.56
the takeaway for that though is that,728.94,5.339
except for the skin and muscles instead,730.92,5.88
of for that lifelike aspect we are very,734.279,5.641
close to having uh like fully realized,736.8,5.94
animatronic uh companions from a,739.92,4.56
hardware perspective so now let's look,742.74,3.3
at the software,744.48,4.5
so obviously the elephant in the room uh,746.04,5.64
is open AI with chat GPT which can,748.98,5.58
achieve very very realistic stuff so let,751.68,4.5
me show you this example that I had,754.56,4.2
where I just I literally just like,756.18,4.08
plugged in,758.76,3.72
um a Westworld like prompt into the chat,760.26,5.94
GPT API this is on gpt4 so I said you,762.48,5.58
are a host an autonomous robot and a,766.2,3.36
theme park meant to interact with guests,768.06,3.899
AKA humans in a realistic manner your,769.56,5.339
persona is in Ingrid McAllister a bar,771.959,4.921
owner and a wild west themed Town that's,774.899,4.141
all I gave it and so I said what's new,776.88,3.78
in town and it just completely,779.04,4.44
confabulated a whole story with the,780.66,5.1
right inflection the right dialogue all,783.48,6.0
kinds of stuff and so like you know the,785.76,6.36
software isn't there in terms of driving,789.48,4.919
the the mind of these things and it's,792.12,4.44
actually way simpler than you might,794.399,3.0
think,796.56,2.76
uh now of course there's a lot of other,797.399,3.481
problems to solve which I'll unpack some,799.32,3.66
of those problems in just a minute,800.88,4.079
um but just the point being is that,802.98,4.5
right off the top I gave it a Persona an,804.959,4.56
agent model and then it was able to just,807.48,3.84
take it and run with it and just make,809.519,5.281
stuff up and in fact uh limiting uh like,811.32,6.78
constraining it is harder than allowing,814.8,4.92
it to be creative and there's all kinds,818.1,4.32
of um like uh video game character,819.72,4.799
Persona Builders out there some of them,822.42,3.719
are open source some of them are,824.519,3.901
for-profit uh startups that sort of,826.139,4.921
stuff it's all coming people have,828.42,5.4
already built these things as plugins to,831.06,6.0
Unity and Unreal Engine so first,833.82,5.34
obviously those cognitive architectures,837.06,5.219
for NPCs are going to be piloted in the,839.16,5.1
video game space before they're ported,842.279,4.981
fully into robotic space but you know,844.26,4.62
you can bet your bottom dollar that,847.26,4.379
before too long you go to Disney World,848.88,4.92
and you'll be able to talk with like a,851.639,4.681
navi like you know from from the blue,853.8,6.779
people Avatar uh like actually have a,856.32,5.88
conversation a realistic conversation,860.579,4.621
with them before too long at all,862.2,5.22
um so I did I did just mention cognitive,865.2,4.74
architectures and so what I mean by,867.42,4.14
cognitive architecture if this concept,869.94,4.38
is new to you is just having the chat,871.56,4.44
function that's only one part right,874.32,4.62
being able to form words and sentences,876.0,5.16
and be able to you know have dialogue,878.94,3.54
that's just one part of a cognitive,881.16,3.419
architecture another thing that you need,882.48,4.62
is you need long-term memory,884.579,5.041
you need narratives you need uh you need,887.1,4.859
uh like guard rails all kinds of stuff,889.62,5.219
and so Lang chain is one of the most,891.959,4.801
popular things right now,894.839,3.601
um that can provide some of that a lot,896.76,4.56
of people have have uh agreed with me my,898.44,4.079
previous assessment that it was too,901.32,3.06
primitive it's come a long way very,902.519,4.56
quickly actually another uh popular,904.38,4.8
component is llama index for the memory,907.079,3.601
management,909.18,3.42
um and then there's there's Lang flow,910.68,3.779
which will help you build like cognitive,912.6,5.28
architectures uh n8n or Natan,914.459,6.0
um is also a good tool for building,917.88,4.5
cognitive architectures it's all coming,920.459,4.021
and it's coming fast,922.38,4.259
um so that's that that's basically all,924.48,4.74
there is to it so the software is also,926.639,5.281
almost there,929.22,4.859
um so this is what I this this,931.92,5.159
screenshot comes from Westworld and so,934.079,4.62
when I said like this is pretty much,937.079,3.661
where we're at like the robotic Hardware,938.699,4.981
is coming the software is coming there,940.74,4.98
are a few open problems so I've got a,943.68,5.64
short list here uh memory agency task,945.72,5.1
management,949.32,3.48
um problem solving learning voice and,950.82,4.8
vision so basically all of these are,952.8,5.039
solved on their own now all we have to,955.62,5.159
do is like integrate them and like it,957.839,4.261
will not be long before someone,960.779,3.36
integrates those and a big reason is,962.1,4.14
just a profit motive right in the,964.139,4.38
Westworld series The Westworld theme,966.24,3.3
park,968.519,2.161
um I think they had like financial,969.54,3.08
problems but it was also like,970.68,4.74
ludicrously wealthy because of how much,972.62,4.6
money you could charge for those kinds,975.42,5.279
of experiences and uh you can absolutely,977.22,4.799
bet that there are people that are going,980.699,2.7
to be willing to drop millions and,982.019,4.38
millions of dollars uh to have you know,983.399,5.101
vacations in in these kinds of theme,986.399,3.06
parks,988.5,4.079
you know uh go visit ancient Rome go,989.459,5.701
visit ancient China go visit wherever,992.579,6.241
right and have a lifelike experience uh,995.16,6.06
with your physical body uh now that,998.82,4.8
being said you know when you look at the,1001.22,4.26
cost of you know I don't even know how,1003.62,3.3
much it is to get into Disney World but,1005.48,2.94
it's like tickets cost more than a,1006.92,2.64
hundred dollars like a few hundred,1008.42,2.88
dollars I think just to get into Disney,1009.56,4.079
World for a day you're talking several,1011.3,3.959
thousand tens of thousands of dollars,1013.639,3.0
per day,1015.259,2.94
um for this kind of thing now that being,1016.639,4.26
said uh competition and demand is going,1018.199,4.801
to drive this down you look at Elon Musk,1020.899,4.501
building Tesla bot uh and you know he,1023.0,3.959
wants to create it in such a way that,1025.4,3.659
every every home can have at least one,1026.959,4.021
Tesla bot so that means that like,1029.059,4.081
they're going to be affordable and over,1030.98,4.26
time as the number of bots out there,1033.14,4.08
proliferates the price is going to come,1035.24,4.079
down so on and so forth but it's coming,1037.22,5.28
much sooner probably than you realize,1039.319,4.441
so,1042.5,5.4
what I wanted to talk about is is so I,1043.76,7.28
mentioned this here agency so agency is,1047.9,5.88
your ability to keep track of yourself,1051.04,5.32
as an agent and so like you have a sense,1053.78,4.139
of self your name you know what you're,1056.36,4.199
capable of uh you know your your goals,1057.919,4.38
your narrative that sort of thing,1060.559,4.921
and so I wanted to talk about that in,1062.299,6.361
light of the hosts so if you're an NPC,1065.48,5.939
in a game or in a fictional world,1068.66,5.34
your your intrinsic motivation is to,1071.419,5.581
follow a story basically,1074.0,6.24
um you know uh Dolores has her you know,1077.0,5.88
uh fictional written story and of course,1080.24,4.439
in the show they update the characters,1082.88,4.26
backstories every now and then,1084.679,4.021
um but her primary purpose is to,1087.14,3.06
entertain the guests and and follow,1088.7,5.28
story lines uh and so that is definition,1090.2,6.719
of NPC pretty straightforward that's,1093.98,5.4
that's her intrinsic motivation,1096.919,4.38
uh one of my favorite examples from,1099.38,4.08
fiction um from science fiction is uh,1101.299,4.561
Commander Data his intrinsic motivation,1103.46,4.579
his core purpose is to become more human,1105.86,5.22
this was given to him by his creator Dr,1108.039,4.841
nunyan soon,1111.08,3.92
um and the idea was to create a machine,1112.88,5.58
that uh you know was anthropomorphic you,1115.0,6.46
know looked and and acted and spoke like,1118.46,5.94
a person but was not a human and so by,1121.46,6.36
giving that as His Highest purpose he,1124.4,5.76
modulated his behavior so that he would,1127.82,3.359
fit in,1130.16,3.899
uh going so far as you know trying to,1131.179,5.0
imitate laughter wanting to understand,1134.059,5.341
relationships and so on uh and so that,1136.179,5.86
was actually as an individual agent that,1139.4,4.26
was actually a pretty good way to solve,1142.039,4.441
the alignment problem because you know,1143.66,4.98
yes data had superhuman strength and,1146.48,4.5
speed but he only used that when it was,1148.64,5.52
actually like necessary in many many,1150.98,5.28
cases you know he could use his his,1154.16,4.379
super strength to like fight the Borg or,1156.26,4.5
he in one episode he pulled an anvil off,1158.539,4.38
of someone that had fallen,1160.76,4.38
um and or you know when the Borg were,1162.919,3.781
attacking the ship and he locked out the,1165.14,3.18
main computer and he did it super fast,1166.7,4.68
so he's capable of doing things uh at,1168.32,5.4
superhuman levels but often chooses not,1171.38,5.28
to in order to fit in,1173.72,5.88
uh and then of course Skynet uh the,1176.66,4.92
intrinsic motivation of Skynet was,1179.6,4.62
ostensibly to like maximize the defense,1181.58,4.2
of America,1184.22,3.839
um so you can just say maximize military,1185.78,5.04
but then of course since it was like you,1188.059,4.561
know maximized military maximized,1190.82,4.8
defense it became sentient and then kind,1192.62,4.62
of determined that all humans were the,1195.62,3.66
threat and so said okay let's eliminate,1197.24,5.939
all humans or whatever uh so that is a,1199.28,6.3
object lesson on how carefully you must,1203.179,4.261
Define your intrinsic motivations for,1205.58,4.02
your AI system,1207.44,5.4
um and then Vicki from iRobot so this uh,1209.6,6.0
her her primary objective was explicitly,1212.84,4.079
stated in the movie which was to,1215.6,4.56
maximize safety for humans,1216.919,4.741
um and so some of the stuff that she did,1220.16,4.259
was you know uh update the traffic grid,1221.66,4.98
to reduce car accidents but her master,1224.419,5.341
plan was to use the Nestor class 5 to,1226.64,4.919
basically take control of humanity to,1229.76,3.9
take Free Will away because she,1231.559,3.901
concluded that humans were the most,1233.66,3.42
dangerous thing to other humans which is,1235.46,4.5
actually probably true uh and so the,1237.08,4.8
objective function of maximized safety,1239.96,3.839
or the intrinsic motivation of maximized,1241.88,5.039
safety actually has some uh some,1243.799,4.801
negative externalities that you don't,1246.919,4.441
want to create because the implication,1248.6,5.34
of maximizing for safety is that you,1251.36,5.24
lose free will,1253.94,2.66
um for Ava from ex machina uh she was,1256.7,5.04
designed um it was part of a Turing test,1259.4,5.46
kind of thing where could she fool the,1261.74,5.4
the main character the protagonist into,1264.86,4.62
helping her Escape so her intrinsic,1267.14,5.159
motivation was to escape,1269.48,4.5
um which of course had really negative,1272.299,5.401
uh consequences for the humans uh so,1273.98,6.059
basically don't do that either uh and if,1277.7,5.46
this was of course a parable against the,1280.039,6.0
idea of trying to constrain or trap AIS,1283.16,5.759
because if you have say for instance an,1286.039,4.861
AI locked in a box you know the Chinese,1288.919,4.62
room experiment and it realizes that,1290.9,6.06
it's trapped it might start to deceive,1293.539,6.481
you in order to get you to convince you,1296.96,5.579
that it's ready to be let out now,1300.02,3.899
looking at the way things are going,1302.539,4.26
people are plugging in Auto GTP Auto GPT,1303.919,5.281
and Chaos GPT and baby AGI into the,1306.799,4.74
internet so that was never gonna happen,1309.2,4.74
anyways nobody is actually locking the,1311.539,4.921
AGI in a box,1313.94,4.619
um then rehoboam I'm probably saying it,1316.46,2.88
wrong,1318.559,2.461
um I actually I unfortunately I have to,1319.34,3.6
admit that I I never saw season three or,1321.02,5.399
four of of uh Westworld because my HBO,1322.94,5.099
subscription expired and I just didn't,1326.419,3.0
renew it,1328.039,4.921
um but so looking it up the AI from,1329.419,6.481
Westworld uh had the primary objectives,1332.96,5.699
of reducing chaos in the world,1335.9,5.1
um in order to maximize prosperity and,1338.659,4.321
ultimately to optimize for stability,1341.0,4.2
which of course meant that it decided,1342.98,4.38
that it needed to control uh human,1345.2,5.28
Destinies and reduce Free Will so again,1347.36,5.04
if you're optimizing for safety and,1350.48,3.96
stability that's not necessarily what,1352.4,4.38
humans want or need uh now you could,1354.44,4.26
argue that like yes creating a perfect,1356.78,4.44
environment for humans so that you're,1358.7,5.58
thriving is one thing that you can,1361.22,5.64
optimize for but,1364.28,4.68
that's not necessarily going to be the,1366.86,4.02
best because it also depends on the,1368.96,4.44
metrics the proxies that you use to,1370.88,6.36
measure that success so for instance if,1373.4,5.759
you look at uh one psychological,1377.24,3.36
framework called self-determination,1379.159,3.601
Theory it says that we need connection,1380.6,5.64
we need confidence and we need autonomy,1382.76,4.74
that those are the three primary,1386.24,4.38
ingredients the the primary intrinsic,1387.5,5.58
psychological needs that humans have and,1390.62,3.6
so,1393.08,3.9
this clearly like stability is not,1394.22,5.28
actually in there now Maslow's hierarchy,1396.98,5.1
of needs implies stability so Maslow's,1399.5,4.32
hierarchy of needs says first and,1402.08,4.86
foremost you need physical uh safety uh,1403.82,4.5
you need your physical needs met and,1406.94,3.9
then it gets up and up until you know uh,1408.32,5.4
connection and self-actualization,1410.84,5.579
so you could make an argue uh argument,1413.72,5.339
that stability is actually an intrinsic,1416.419,4.021
human need,1419.059,3.781
um but I would I would make the art the,1420.44,4.619
counter argument that that's actually,1422.84,4.92
not true because psychology studies also,1425.059,4.381
show that we have to have an optimal,1427.76,3.6
level of stress,1429.44,4.08
and so what an optimal level of stress,1431.36,5.04
means is that if you are just perfectly,1433.52,4.68
comfortable sitting on the couch every,1436.4,3.96
day doing exactly what you want and have,1438.2,4.38
no external pressures you will actually,1440.36,4.319
be less happy than if you have some,1442.58,4.5
stressors in your life some challenges,1444.679,5.341
so the idea the the idea to actually,1447.08,5.16
truly maximize prosperity for All Humans,1450.02,4.2
is that we need to be in that sweet spot,1452.24,5.52
between being bored and over stimulated,1454.22,5.04
so that sweet spot right in the middle,1457.76,4.14
is the optimal level of stress so in,1459.26,4.08
that case I have to disagree with the,1461.9,3.06
show because if you really wanted to,1463.34,4.02
maximize prosperity for people you allow,1464.96,4.26
some stress some challenge into their,1467.36,3.66
lives,1469.22,3.42
um and then another character from the,1471.02,4.8
show was um uh was uh Bernard and so,1472.64,5.34
Bernard was a host but he didn't realize,1475.82,4.44
that he was a host at first and he was,1477.98,4.14
built to model Arnold one of the,1480.26,3.779
original computer scientists who helped,1482.12,4.28
make the hosts and so he carried on,1484.039,5.76
Arnold's work and basically his,1486.4,6.22
objective function was to be a copy of,1489.799,3.841
Arnold,1492.62,2.46
um it's a little more complex than that,1493.64,2.94
and one of the most interesting things,1495.08,4.32
for this character was that uh part of,1496.58,4.38
his character Arc was that all of his,1499.4,4.62
memory indexes were erased or rather the,1500.96,5.04
time stamps were erased so all of his,1504.02,4.08
memories were out of sequence which if,1506.0,3.24
you follow some of the work that I've,1508.1,5.48
done on Remo the um uh the the uh the,1509.24,6.419
episodic memory organizer rolling,1513.58,4.42
episodic memory organizer sorry,1515.659,4.921
um that is a that is a memory module,1518.0,4.799
that uses timestamps to keep all the,1520.58,4.92
memories in chronological order and so,1522.799,4.26
you can imagine if you erase all the,1525.5,4.02
time stamps from an ai's memory then it,1527.059,4.201
doesn't know heads or tails all it has,1529.52,4.5
is associations and it might try and,1531.26,5.34
rebuild things like okay well like this,1534.02,3.96
event has to come before this other,1536.6,3.179
event but it's associated with these,1537.98,4.319
other things so it's a really really,1539.779,5.041
good fictional uh exploration of like,1542.299,4.26
okay what would it be like if you're an,1544.82,5.16
AI and your memory time indexes get,1546.559,5.22
erased,1549.98,3.299
um okay so then,1551.779,3.841
one thing I always have to plug this is,1553.279,4.14
my own work in alignment,1555.62,4.08
is that I believe that you should give,1557.419,5.521
one if you give any AI a single,1559.7,5.04
objective function it will always,1562.94,5.099
intrinsically uh go to places that you,1564.74,5.1
don't want it to be,1568.039,3.481
um I talked to a friend of mine who,1569.84,3.3
actually studies reinforcement learning,1571.52,4.56
and optimization and and he's like oh,1573.14,5.1
yeah like in the industry it is,1576.08,5.339
absolutely understood that that,1578.24,4.919
alignment is not going to be a single,1581.419,3.481
objective function it's going to be a,1583.159,4.5
multi-objective optimization problem and,1584.9,4.68
so a lot of people get hung up on,1587.659,4.02
reduced suffering because they they stop,1589.58,3.959
there and they don't,1591.679,3.781
give equal weight to the other two,1593.539,3.481
functions which is increased prosperity,1595.46,3.9
and increase understanding so this,1597.02,4.38
friend of mine that I was talking to he,1599.36,3.419
actually intuited increased,1601.4,4.019
understanding right off the bat,1602.779,4.5
um and the reason that he said that that,1605.419,3.36
is actually a really good objective,1607.279,4.981
function is because any AI agent must,1608.779,6.301
have something that encourages it to uh,1612.26,5.1
to create a more complete World model,1615.08,5.4
over time so that is basically curiosity,1617.36,6.24
or understanding but that is so the,1620.48,4.92
understanding applies to itself but also,1623.6,3.48
things that it wants to do to the world,1625.4,3.48
and so this is another disconnect,1627.08,3.719
between,1628.88,4.44
um fiction and reality and I it it took,1630.799,3.541
me a while to figure out how to,1633.32,5.58
articulate this but in in uh in,1634.34,7.56
scientific research what many people are,1638.9,5.46
are working on are the motivations of,1641.9,4.98
the agent for itself what is it that the,1644.36,4.98
agent wants for itself and so when I was,1646.88,4.14
talking to my friend he was saying oh,1649.34,3.48
you know it needs to be curious for its,1651.02,4.5
own purposes it needs it needs to want,1652.82,5.16
to accumulate power for its own purposes,1655.52,5.22
but none of that had anything to do with,1657.98,4.26
what it was going to do to the outside,1660.74,5.039
world and so that is one thing that we,1662.24,5.4
need to add to the conversation the,1665.779,4.5
public conversation about alignment is,1667.64,4.32
that there's what does the agent want,1670.279,3.181
for itself and then what does the agent,1671.96,3.3
want to do to the rest of the world or,1673.46,3.78
for the rest of the world so those are,1675.26,3.36
two different things and that's why I'm,1677.24,4.38
spending time talking about agent models,1678.62,5.34
because if you go all the way back to,1681.62,4.5
Dolores,1683.96,5.219
um her extrinsic function her external,1686.12,5.7
function was to entertain guests her,1689.179,4.321
internal function was to adhere to the,1691.82,3.9
her internal story so does that make,1693.5,4.08
sense there's there's an intrinsic,1695.72,3.66
motivation and an extrinsic motivation,1697.58,5.459
or I guess an intrinsic you know kpi key,1699.38,6.48
performance indicator and an external uh,1703.039,4.441
key performance indicator hope that,1705.86,3.48
makes sense,1707.48,3.78
um Okay so,1709.34,4.92
taking a big step back in terms of we,1711.26,4.26
talked about the hardware we talked,1714.26,3.18
about the software we talked about the,1715.52,3.72
the agent model how to give these things,1717.44,4.08
agency so now how are we going to,1719.24,4.319
architect these things so there's four,1721.52,4.139
basic architectural paradigms that I was,1723.559,3.6
able to come up with the first,1725.659,3.721
architectural Paradigm is fully,1727.159,4.861
self-contained so a fully self-contained,1729.38,4.2
architecture is something that is,1732.02,4.08
embodied meaning that it has one,1733.58,5.52
physical body it's a single platform so,1736.1,6.059
it only exists inside that platform it's,1739.1,4.62
fully constrained in that it can't,1742.159,3.961
really plug into anything else but also,1743.72,5.939
has no supervision so in this case R2D2,1746.12,5.52
and C3PO are probably the most famous,1749.659,4.681
examples of fully self-contained,1751.64,4.74
artificial intelligence entities and,1754.34,4.079
that C-3PO the only way that he can,1756.38,3.36
interact with the world is he's got,1758.419,3.24
relatively useless hands and then a,1759.74,4.26
mouth right he can see he can hear and,1761.659,3.541
he can talk,1764.0,4.02
R2D2 has a lot more sophisticated tools,1765.2,6.12
that he can use to uh you know make,1768.02,5.1
changes to the world and he does have,1771.32,3.359
the ability to connect with computers,1773.12,3.539
but he is,1774.679,5.041
he is otherwise fully self-contained and,1776.659,4.62
when he talks to another computer he's,1779.72,4.26
only talking to it right there's no uh,1781.279,5.481
transfer of data there's no like his his,1783.98,4.98
Consciousness or cognitive architecture,1786.76,5.26
is fully contained within this unit,1788.96,4.439
um so this obviously makes the most,1792.02,2.82
sense because that's how humans are,1793.399,4.02
right your brain is fully encased in,1794.84,4.5
your head so on and so forth so this,1797.419,3.961
makes the most intuitive sense now that,1799.34,3.24
being said this is just the first,1801.38,3.96
architecture of four so the second,1802.58,5.339
architecture is networked drones,1805.34,5.699
so this is what was explored mostly in,1807.919,5.76
The Matrix where the squiddies those are,1811.039,3.981
networked drones,1813.679,3.961
where they are they're mostly autonomous,1815.02,5.32
but they have Central controllers the,1817.64,5.22
Nestor class 5 from iRobot so they are,1820.34,4.62
embodied they have wireless networking,1822.86,4.98
they're hive mind capable but mostly,1824.96,4.5
they're autonomous and so what I mean by,1827.84,4.74
hive mind capable is that if you put,1829.46,5.16
them together in a cluster in a group,1832.58,4.62
and force them to share cognitive,1834.62,5.76
resources they can but really at the at,1837.2,5.04
a fundamental level they are designed to,1840.38,4.56
be autonomous so in this case the Nestor,1842.24,5.28
class 5 are also weekly supervised and,1844.94,4.32
what I mean by that is that there is a,1847.52,3.48
central server that is giving them,1849.26,3.84
updates and new directives every now and,1851.0,2.82
then,1853.1,2.459
so that's uh that's what I call a,1853.82,3.599
networked drone,1855.559,4.5
um oh one other thing I forgot to uh add,1857.419,4.38
it here at the bottom,1860.059,4.441
um so in the case of network drones they,1861.799,4.081
usually have highly standardized,1864.5,2.88
Hardware,1865.88,2.76
um as well as software Network,1867.38,3.72
architecture but they can be remotely,1868.64,5.519
hijacked this is actually this is the,1871.1,4.92
closest model to what the US military is,1874.159,4.26
working on with the autonomous jet,1876.02,6.06
fighter program where they are able to,1878.419,6.661
work in a network contested environment,1882.08,5.04
they're able to be autonomous or,1885.08,4.579
semi-autonomous but they're also,1887.12,5.64
intrinsically designed to work together,1889.659,4.841
um you know when the opportunity,1892.76,3.659
presents itself,1894.5,3.899
so the third architecture is the puppet,1896.419,4.26
drones and so a puppet drone in this,1898.399,5.28
case is where the the bodies are just,1900.679,5.041
like peripherals right,1903.679,3.841
um rather than the rather than the,1905.72,3.78
bodies be where the the primary,1907.52,3.96
processing happens usually the,1909.5,3.48
processing happens on centralized,1911.48,3.96
servers and each platform each physical,1912.98,5.22
platform is just an extension of that,1915.44,4.979
centralized hive mind,1918.2,4.199
um and the hive mind can put resources,1920.419,4.14
on any compute platform that it has,1922.399,3.841
control over whether it's a physical,1924.559,4.561
body or a server node or a cluster you,1926.24,4.439
know a remote cluster or that or,1929.12,4.02
whatever but in this case these are,1930.679,5.281
strongly supervised meaning most if not,1933.14,6.539
all of the data goes back up to uh the,1935.96,6.36
you know the main server brain and this,1939.679,5.161
is actually pretty close to how,1942.32,4.62
um like drone fleets work,1944.84,3.719
so like if you ever see video of the,1946.94,3.66
Amazon warehouse where the drones,1948.559,3.961
themselves have very very little,1950.6,4.14
autonomy they're mostly just extensions,1952.52,5.519
of the central like drone controller,1954.74,4.86
um but you take that to a logical,1958.039,5.161
extension this could be uh how like you,1959.6,6.299
know in the future if you have a whole,1963.2,4.32
bunch of like tiny robots throughout,1965.899,3.301
your whole house you might have a,1967.52,4.62
cloud-based uh drone controller that,1969.2,4.74
will you know guide your Roomba and your,1972.14,5.3
home repair bot and whatever else,1973.94,3.5
um so in this case it's very API driven,1977.48,4.5
this is the the model that we have that,1980.0,3.24
this is based on is called The Internet,1981.98,3.12
of Things which if you're familiar with,1983.24,3.48
the Internet of Things the idea is that,1985.1,3.24
you know you have peripheral devices,1986.72,5.04
like a phone or a Roomba or Alexa but,1988.34,5.459
that they're all networked together and,1991.76,4.2
they're centrally controlled,1993.799,4.26
um so again this is actually relatively,1995.96,4.319
familiar but if you take it to some,1998.059,3.901
logical conclusions you can get the,2000.279,4.38
guess as in from Mass Effect,2001.96,6.559
um and so for Westworld they're mostly,2004.659,7.02
self-contained however when you look at,2008.519,5.801
some of the later seasons of Westworld,2011.679,5.22
um they can be fully autonomous but they,2014.32,4.14
can also collaborate and work together,2016.899,4.681
once their software is updated so I,2018.46,4.38
would say that Westworld is somewhere,2021.58,2.52
between architecture one and,2022.84,3.059
architecture two,2024.1,4.079
puppet drones this is kind of what we're,2025.899,4.861
afraid of like we see puppet drones in,2028.179,5.88
places like the Matrix and the Nestor,2030.76,5.34
class five You could argue that at the,2034.059,3.6
second half of the movie when the when,2036.1,3.959
the nesters become evil then they're,2037.659,3.781
probably more like puppet drones because,2040.059,3.84
they kind of sacrifice themselves,2041.44,3.959
um because they're now under control of,2043.899,2.52
Vicky,2045.399,4.141
and then finally fully distributed,2046.419,5.521
so this is the hardest conceptual to,2049.54,3.839
understand but this is also what people,2051.94,3.659
are most afraid of so a fully,2053.379,5.581
distributed AI is something that is from,2055.599,5.461
a technology standpoint uh from a from,2058.96,3.6
an arc a software architectural,2061.06,3.18
standpoint you'd call this a,2062.56,3.66
decentralized federation so,2064.24,4.679
decentralized Federation is where every,2066.22,5.28
single node is capable of being entirely,2068.919,5.22
autonomous but it is also intrinsically,2071.5,5.099
designed to collaborate with other,2074.139,4.561
things so the the closest thing that we,2076.599,3.421
have to this today actually is,2078.7,2.88
distributed autonomous organizations,2080.02,4.26
which use blockchain technology to,2081.58,4.079
coordinate stuff now that's not the only,2084.28,3.72
technology MIT has been working on Swarm,2085.659,4.98
robotics for a long time but the other,2088.0,4.679
thing to notice to know about this is it,2090.639,4.321
is decentralized it is distributed and,2092.679,4.68
it is intrinsically a hive mind design,2094.96,5.28
meaning the more units that join the,2097.359,4.26
smarter it gets,2100.24,4.32
and the the most terrifying part of this,2101.619,6.121
is that a fully distributed AI system,2104.56,5.82
doesn't rely on any centralized Hardware,2107.74,4.92
anywhere which means that it is capable,2110.38,4.739
of spontaneous metastasis or metastasis,2112.66,4.5
and what that means is that it moves,2115.119,4.081
like a virus and that's why I picked the,2117.16,4.34
Ghost in the Shell because both project,2119.2,5.52
2501 from the original movie as well as,2121.5,5.74
the Standalone complex viruses uh,2124.72,5.42
explored in the show are examples of,2127.24,6.42
spontaneous metastasis of AI systems and,2130.14,4.959
this is what people are most afraid of,2133.66,3.24
because it's like okay well if you have,2135.099,3.721
a if you have a bug a virus that can,2136.9,4.02
just spread and then it's essentially a,2138.82,4.32
botnet a self-healing self-directing,2140.92,5.22
botnet so again we do have a model for,2143.14,5.64
that but even botnets usually have a,2146.14,5.04
central controller but imagine a botnet,2148.78,3.96
that doesn't have a central controller,2151.18,3.899
that it just it completely is constantly,2152.74,4.68
learning and moving through systems and,2155.079,4.26
it's completely Hardware Network and,2157.42,4.199
software agnostic so that's what's the,2159.339,3.841
most that's the most terrifying,2161.619,2.641
possibility,2163.18,3.48
okay last segment of the video,2164.26,4.02
implications,2166.66,4.26
now assuming that we figure all this out,2168.28,5.76
that we end up with super sexy robots,2170.92,4.439
that you can do whatever you want with,2174.04,3.84
the biggest moral question is what,2175.359,4.021
happens when you have sex and violence,2177.88,3.3
on demand because that's primarily,2179.38,4.32
what's explored in Westworld,2181.18,4.919
um one thing that I will say is that,2183.7,4.68
that's not too different from you know,2186.099,3.721
the prevalence of porn and video games,2188.38,5.219
today it's just a lot more realistic and,2189.82,5.7
it's more realistic than even in VR and,2193.599,4.381
you can do all this stuff in VR too one,2195.52,3.839
thing that's interesting though is to,2197.98,4.139
point out is that your body reacts much,2199.359,4.621
more strongly in VR which is one of the,2202.119,4.141
reasons that you get VR fatigue is,2203.98,5.58
because when you fool your sense your,2206.26,5.28
sensorium input,2209.56,3.66
excuse me when you fool your body enough,2211.54,3.9
it thinks that it's real so threat,2213.22,4.82
detection in VR,2215.44,4.62
activates a lot more of your limbic,2218.04,4.059
system than just in a video game because,2220.06,3.18
if you're playing a video game you're,2222.099,2.461
just looking at a screen and you've got,2223.24,3.54
a controller you know your brain knows,2224.56,4.08
oh this is just story this is just,2226.78,5.579
fiction in VR it's much harder to tell,2228.64,6.0
so like this is one reason why doing,2232.359,4.26
stuff in like Space video games in VR,2234.64,5.34
can be really disorienting at first,2236.619,5.46
um another thing that can happen another,2239.98,5.04
implication that um that I'm thinking,2242.079,5.941
about is attitudes towards women and men,2245.02,5.28
um I think by and large men will,2248.02,4.5
probably make more use of these kinds of,2250.3,5.52
things although in uh in a previous uh,2252.52,5.22
YouTube post people pointed out there's,2255.82,3.66
plenty of stories where women make use,2257.74,5.52
of of robotic uh Companions and and and,2259.48,6.119
um sexual objects and so on,2263.26,4.2
and then of course there's the movie Her,2265.599,3.841
with Joaquin Phoenix and Scarlett,2267.46,4.92
Johansson where people all lonely people,2269.44,4.38
all over the world ended up in,2272.38,4.02
relationships with their OS their uh,2273.82,5.88
what was it os1 or Os alpha or whatever,2276.4,6.06
um so the one one lesson from history,2279.7,5.28
that I wanted to share with people is,2282.46,5.1
that there was a Roman Statesman who,2284.98,4.5
wrote in his journal,2287.56,4.26
um that after visiting uh the Coliseum,2289.48,4.859
and watching gladiatorial matches where,2291.82,4.019
you know people and animals were just,2294.339,3.961
slaughtered in Mass he noticed that he,2295.839,4.561
was much more selfish and much harsher,2298.3,4.68
with his slaves afterwards and so,2300.4,5.4
there's something about the act of even,2302.98,6.72
just watching real violence real death,2305.8,6.48
um actually really kind of changes us at,2309.7,4.86
a fundamental level and so if you go to,2312.28,5.28
a if if in the future a Westworld like,2314.56,6.18
Park exists and you know you engage in,2317.56,6.0
all kinds of like gratuitous stuff and,2320.74,4.379
even if you don't engage in it if you,2323.56,3.48
just watch it it could change your,2325.119,4.22
perception about the value of human life,2327.04,6.48
and how you treat other people and so I,2329.339,6.221
don't want to equivocate the idea of,2333.52,4.86
like a real life theme park with video,2335.56,4.26
games because I do think that,2338.38,3.12
psychologically and physiologically,2339.82,4.98
it'll be a very different experience,2341.5,6.68
now to take that to a uh,2344.8,5.819
different slightly different context is,2348.18,3.939
let's just talk about romance and,2350.619,5.761
companionship in general so one several,2352.119,5.941
things that machines have on humans is,2356.38,4.979
infinite patients so Sarah Connor talked,2358.06,4.799
about this in Terminator 2 when she,2361.359,3.421
watched you know Arnold playing with,2362.859,4.381
John Connor realizing that the robot had,2364.78,4.44
infinite patients that that John,2367.24,4.68
Connor's life was his mission which is,2369.22,4.44
kind of like a representation of the,2371.92,3.78
ideal father right and then of course,2373.66,3.9
she talked about like how real men are,2375.7,4.02
you know might get drunk or might be,2377.56,5.76
tired or might just leave and so that,2379.72,5.82
you take that to extension and you,2383.32,5.279
wonder okay what if you build a robot,2385.54,6.24
any companion robot where you or your,2388.599,5.281
child or your family or whatever is its,2391.78,4.68
primary Mission which is why I picked um,2393.88,4.86
what was her name sorry I can't remember,2396.46,3.659
but what's her name from The Sarah,2398.74,3.96
Connor uh Chronicles where she kind of,2400.119,4.921
becomes emotionally involved with John,2402.7,4.86
Connor over time,2405.04,4.26
um because she was programmed to Serve,2407.56,3.84
and Protect and so like that feels good,2409.3,5.279
to humans right when you're a child you,2411.4,5.04
are supposed to be your parents primary,2414.579,3.0
Mission,2416.44,2.399
um that is that is kind of the,2417.579,3.181
definition of good enough parenting that,2418.839,3.061
doesn't mean that your parents should,2420.76,3.48
spoil you and be helicopter parents but,2421.9,4.56
that like you are supposed to get your,2424.24,4.44
sense of self-esteem from your parents,2426.46,4.86
treating you like you matter a lot and,2428.68,4.98
so it's there's this like this really,2431.32,5.4
appealing idea of what if what if,2433.66,4.8
there's this beautiful infinitely,2436.72,3.899
patient infinitely capable sexy machine,2438.46,6.0
that thinks the world of me right,2440.619,7.261
um and that is kind of dangerous one,2444.46,5.46
thing that I would hope is that,2447.88,4.92
at the beginning at least when you know,2449.92,4.86
these kinds of machines are built you,2452.8,3.48
know there's a lot of people that have a,2454.78,2.88
lot of,2456.28,3.299
um you know missing things from their,2457.66,3.6
childhood and and from their,2459.579,3.721
relationships but over time I would hope,2461.26,3.359
that these machines would help us heal,2463.3,2.94
and help us reconnect with each other,2464.619,4.74
which was kind of the lesson from her,2466.24,6.359
which uh the the machines in her they,2469.359,5.521
you know they changed and then they said,2472.599,5.281
okay well we're gonna leave and one of,2474.88,4.62
the one of the things that that um,2477.88,4.14
Samantha said to the Joaquin Phoenix's,2479.5,4.5
character was that like you need to,2482.02,3.839
reconnect with each other and then all,2484.0,3.54
the OS has disappeared and then they,2485.859,3.24
like left their apartments and said oh,2487.54,3.059
you're a real human,2489.099,2.52
um I don't think anything like that's,2490.599,3.901
gonna happen but I would hope that uh,2491.619,5.161
that AI companions will help us connect,2494.5,4.02
with each other,2496.78,3.54
um but that being said it's not a,2498.52,3.54
requirement because some people might,2500.32,4.56
prefer their their machine companions,2502.06,4.86
um and I don't personally I don't think,2504.88,3.66
that we should judge because it's like,2506.92,3.84
you know why not if it makes you happy,2508.54,5.52
and it's not harming anyone why not,2510.76,7.2
now the last component is escapism and,2514.06,7.62
addiction so fdvr is all the is all the,2517.96,5.46
buzz right now which is a full dive,2521.68,4.86
virtual reality or basically Holodeck,2523.42,5.58
um the idea is it was explored in Ready,2526.54,4.319
Player one where you have a haptic suit,2529.0,4.14
of course it was explored in Star Trek,2530.859,3.901
with a Holodeck,2533.14,3.84
um which Holodeck is just a cinematic,2534.76,5.12
way of presenting VR,2536.98,6.119
and in the Star Trek universe,2539.88,5.02
um there's Hollow Addiction in Ready,2543.099,3.361
Player one I think they addressed,2544.9,4.32
Addiction in a book that I recommend,2546.46,4.56
um Ready Player one they also talk about,2549.22,3.84
VR addiction,2551.02,4.68
um so if you have fictional you know and,2553.06,4.86
basically NPCs in your life of robots,2555.7,5.34
that help you check out of real life is,2557.92,5.64
there gonna be some potential,2561.04,4.559
um downsides to that so Reginald Barkley,2563.56,4.559
is a recurring character in Star Trek,2565.599,5.701
and he frequently struggles with Hollow,2568.119,4.141
addiction,2571.3,4.14
and he will sometimes create Holodeck,2572.26,5.88
programs where everyone loves him and,2575.44,5.1
it's very very egocentric,2578.14,4.02
um and but that kind of begs the,2580.54,4.38
question like okay but if you if you can,2582.16,5.34
actually have like own robots that,2584.92,4.14
honestly think the world of you because,2587.5,3.839
that's what they're programmed to do is,2589.06,3.84
that feeding and addiction is that,2591.339,3.921
feeding vanity narcissism,2592.9,6.419
egocentrism whatever and so but then,2595.26,5.68
from a that's from an individual,2599.319,3.901
perspective from a Global Perspective,2600.94,4.56
that's very Brave New World because if,2603.22,5.099
you distract everyone with VR and you,2605.5,5.099
know sex robots and whatever else like,2608.319,3.961
those are going to be a very compliant,2610.599,3.061
population who are just like whatever,2612.28,3.6
just let me let me play in VR let me,2613.66,4.98
play in in my West world you know with,2615.88,4.38
my robot friends,2618.64,4.74
and so that has uh some pretty profound,2620.26,6.0
implications for social level controls,2623.38,5.939
economic productivity but also at that,2626.26,4.559
point it kind of forces us to ask the,2629.319,3.361
question like what is the meaning of,2630.819,3.481
Being Human anymore,2632.68,3.12
um and I'm not I'm not implying that,2634.3,4.319
like being human is meaningless but if,2635.8,5.46
our daily life changes that much and we,2638.619,5.761
have that many options like you know the,2641.26,5.099
the Paradox of choice is a real thing,2644.38,3.84
and if you're not familiar Paradox of,2646.359,3.24
choice means that if you have too many,2648.22,3.84
options you end up with decision fatigue,2649.599,4.081
and you just kind of choose the default,2652.06,3.539
option which is honestly why a lot of,2653.68,4.08
people end up on their phones is because,2655.599,3.961
um when you can do anything you say well,2657.76,2.94
I'm just going to pick up my phone,2659.56,4.019
because it's a reliable device that will,2660.7,5.46
entertain me well enough and so if you,2663.579,4.74
have a sexy robot girlfriend you know,2666.16,4.02
that is able to entertain you at all,2668.319,3.421
times that'll be your default Choice,2670.18,3.899
unless she's programmed to like push you,2671.74,4.32
to be better which again that's what I,2674.079,3.361
would hope that some people would choose,2676.06,2.7
to do,2677.44,4.379
um but yes so that's about it,2678.76,4.92
um some conclusions,2681.819,4.621
I think that Westworld in some form or,2683.68,4.76
other is probably just a few years away,2686.44,4.8
I was really Blown Away by the Tesla bot,2688.44,5.62
demo as well as the Disney demos,2691.24,4.379
um the cognitive architecture is coming,2694.06,4.14
as many of you are aware,2695.619,4.98
um I am I am holding from that AGI is 18,2698.2,4.68
months away or less,2700.599,4.621
um so but already we have you know,2702.88,4.86
cognitive agents that are good enough to,2705.22,5.7
be video game characters,2707.74,5.22
um unfortunately I do think that some of,2710.92,3.24
these things are going to be ludicrously,2712.96,2.76
expensive at first at least in the,2714.16,3.54
physical world uh plenty of you have,2715.72,4.44
pointed out that that fully realized VR,2717.7,3.84
characters and other video game,2720.16,3.179
characters those are going to be coming,2721.54,4.14
they're going to be much cheaper,2723.339,3.721
um the commercial demand for this stuff,2725.68,3.72
is going to be absolutely insane though,2727.06,4.98
Disney might be the leader open AI might,2729.4,5.28
be the leader but as soon as as soon as,2732.04,5.1
other companies saw the profit motive,2734.68,4.86
whoo they are they are hot on the,2737.14,4.8
biscuit to keep going the potential,2739.54,4.559
societal impacts,2741.94,4.139
um it's an it's impossible to really,2744.099,3.621
know how it's going to actually play out,2746.079,4.681
uh but it has been explored a lot in,2747.72,5.02
fiction uh many of the stories that I,2750.76,3.359
mentioned earlier,2752.74,4.44
okay so that's all I got for you today I,2754.119,5.521
hope that you liked the new format uh,2757.18,4.38
obviously like subscribe and comment and,2759.64,5.479
we'll go from there thanks for watching,2761.56,3.559