good evening everyone and welcome to another
edition of The Dog Walk! this is actually episode 15, I forget what I said yesterday...
technically, I wasn't wrong but unfortunately one of the recordings that I had made, I found
out when I was checking it out after uploading it to YouTube yesterday, that it's uh not really
usable... so let's just say that this is now going to be episode 15 and yesterday's very quick
little chat will be episode 14. so anyway, uh The Dog Walk is basically just m
y occasion
to offload, hopefully daily, my thoughts of what's been going on during the day, while I
am out walking Mystra and Daisy, my two pups that you see in front of you. and hopefully some
of this is thought-provoking or interesting to you. I never talk about or rarely talk about
the same topics, I go from gardening, to AI and everything in between and I hope that you
will enjoy the ride. so speaking of AI, I think that's actually going to be the topic for today,
because I was watchin
g a video from asmongold, again... yes I discovered him only a few months
ago, but I have to admit I have been binging a lot of his content. I find that he actually has
some pretty good takes on several topics. I agree with a lot of the stuff that he says. even if I
don't with others... a few topics. but one of the things that he was talking about today was about
the big controversy uh that's currently going on with this game called PalWorld, which is being
accused of all sorts of nasty thi
ngs, as it is supposedly --or I guess the allegations are that--
it's basically just a ripoff of Pokemon that was procedurally.. that was generated... that was made
through the use of generative AI, and as I'm sure you know AI is a big topic right now, especially
when it comes to artists. there was that big Screen Actors Guild and Writers Guild strike a
few months ago, and I just thought that I would talk a little bit about this topic because some of
the comments and opinions that were voic
ed by his readers mostly ---or sorry readers--- his viewers
mostly, were so completely wrong, they clearly showed that people don't understand what AI is or
how it works, which I guess makes sense... (Being an Engineer,)I don't really understand how all
the different artists, art movements, or how to correctly differentiate from one another this or
that, just because I haven't looked into it... but at least I don't go around voicing my opinions
about these movements and these art styles! bu
t that's beside the point... so I thought I
would clear the air a little bit, cuz some of the things that Zach said also were a little
bit well... not quite correct. he was close, but not quite there on a few things and I want
to talk about this because I actually have some background in this. as I've mentioned in the past,
I do have a degree in that stuff. I did a double major in computer science and space science, and I
was actually lucky enough to, in my fourth year of University, and we
are talking almost 20 years ago
now uh yeah 2024... yeah actually 20 years-- 19 years ago now. I was lucky enough to study under
one of the Pioneers in the field, who was doing a lot of cutting edge research at the time for the
military and the government on Neural networks. so I feel I have some --we're going this way-- I have
some credibility or maybe not credibility but at least some some of the prerequisite knowledge
that most of the people who are commenting on this topic seem to lack
. so yeah that was a little
bit about me. I obviously am not fully up to speed with all the latest and greatest capabilities
of AI, but I do know enough about the basis of it and the basics that I feel I can shed a little
bit of light on some of the topic. so the first thing that I want to discuss when talking about
AI is.. I guess I'll try to debunk a few myths or misconceptions. so starting at the basics, AI
is a very generic term that can mean a lot, a lot of different things. come on gi
rls let's keep
going... and we haven't been through this path for a little while so the girls are probably going to
be extra slow and stopping today because they have a lot of messages to catch up on, so sorry about
that. but anyway so starting with the basics, AI is not what most people think it is. I mean
this might be changing over the last year because of Chat GPT and those other tools that are now
becoming-- reaching the Zeitgeist, so people are starting maybe to change, but I still do
n't think
that people truly grasp the difference between a neural network, a natural intelligence system
and an expert programmed AI. so what do I mean by that is that most people when they think of AI
at the basic level they think of a system where some human came in and basically as an expert,
programmed all the possible scenarios and all the possible responses into the AI, and told --- when
this happens, you do this... when the the client in the chat window says that he has an issue with
products XYZ, then you pull up these following questions related to products XYZ. if the client
types in the word "agent", you automatically redirect them to an agent blah blah... you
know think of the, at the most basic level, kind of like the you know uh automatic telephone
menus right? to talk about your I don't know... your checking account press one, to talk about
credit cards press two, to talk about mortgages press three, ETC right? so that's what is
called an expert system or an e
xpert AI and that was ---oh there's something there, definitely
pretty sure that's a rabbit---Daisy! yeah see? I can see the rabbit right there I don't know if
you can? Mystra making me... okay well anyway if I can... hello rabbit how's it going buddy? cool!
anyway.... so so yeah that is- that was the first form of "artificial intelligence" (that's been
around for around 50 years) and that's what people still kind of think it is. even with generative
AI being much more advanced than that, I
feel that most people in their mind they're thinking
you know there's a human you know, there's... the Wizard of Oz is behind the curtain pulling
the strings and telling the AI how to react to prompts and basically training up the system uh
for responses. um that's not really true or at least that's not the final form of AI. the final
form of AI and this is... I say the final Form-- this is something that as I said I was studying
almost 20 years ago-- the final form of AI is actually neura
l networks, and the best person I'll
go I'll mention this right away before I forget to mention it because it's important, the best
person that you could possibly ever find in my opinion as a source for information on this topic
or to you know get the Layman to understand it is Ray Kurtzweil. you Pro-- might not know who he
is, Ray Kurtzweil is a very famous uh "futurist" although I don't really like that term... he was
one of the... for I don't know how many years he was the the chief tech
nology Officer of Google...
so if I just mention that out of his biography, excluding everything else, hopefully that at least
piques your interest to say "hey, maybe you know that Kurtzweil guy must know a little bit about
what he's talking about if he made it all the way to the title of Chief technology Officer of Google
right?" So he's written several books, including one of my all-time favorites which is called "how
to create a mind", where he explores technology, AI progress Etc and on
e of his main points, which
is a little bit off topic for AI but I'll still mention it since I'm talking about him at this
point, is Moore's law is not a new thing, and it's not something that we truly understand so I can't
really do justice to his entire argument that he elaborates in a 200 Page or whatever 200 plus page
book in 2 minutes, but he shows in his book that the principle of Moore's Law, which is you know
the density of transistors on a chip will double every 18 months or so I t
hink it was, that's just
the newest iteration of the overall progress of human technology as far as we can go back. and
he literally almost goes back to the invention of fire and he shows that throughout human history
technological progress has been progressing sorry using the same word tce and there's another
rabbit right over there, so Daisy and Mystra are probably going to go for it... yeah so basically,
since the dawn of mankind this has been the case, this has been a thing, and uh we a
re now
realizing-- my God that rabbit really wasn't scared-- sorry for the shaky cam as I mentioned
Mystra is pretty strong, so yeah basically Moore's law is just the computer version of it, where he
(Moore) noticed that transistors were doubling in density which means -twice as many processors
means- twice as much (computing) capacity right? that's just a latest iteration but if you go back
all the way to the first inventions that's always been true and his main point is--- goes back
to e
xponential growth, another topic that I've spoken of in one of my previous videos. maybe if
I become a good Creator I'll put a little card or something to link you to that one... but yeah
his point is because exponential growth is... exponential people always always underestimate
how fast things will change, or how fast things progress. it will just keep getting faster. I
think anyone who's been around for a decade or two or yeah more than a decade let's say will see
this in their daily liv
es, how fast is technology progressing now compared to when you were born...
and because of the exponential growth factor, you are still projecting a linear growth in
your mind when, like Asmongold was saying "oh in 200 years there'll be computers that are
--that have brains like humans" like sure, I agree with the second part of your sentence but
when you factor in exponential growth, think more like 10 20 Max 30 years (not 200), and that's just
based on what we know now not what's actuall
y, how do I say, available in big universities and
the public-- uh sorry-- unknown to the public right? the bigger --whether it's a CIA or research
University-- computers... all that stuff... we don't know how smart they are! but point is that
it is very likely in my mind that there currently are computer systems out there that have the same
processing power as a human brain. now does that mean they are as smart as a human? right now no
because all those systems are still limited by the dom
ains which they can access. we can operate our
body and we're not limited, we're not just applied in doing image analysis, and that's the only thing
that I do as an AI system, or you know optimizing production chains or whatever it is right? we're
more (...) can't say that word in English... polyvalent-- anyway, we're more versatile,
there you go! but that's just the situation now, and once we start to integrate multiple
different AIs from different expert Fields, you know when you integrat
e the vision AI that
Tesla is developing with the optimization AI that Amazon is producing, and when you start
to combine all those things with the neural Nets that Boston Dynamics is creating to control
their robots, you integrate all that together, and pretty soon we'll be obsolete. Now you might
not believe that, so let me explain a little bit about what I mean by processing power. it's very
simple... oh actually yeah no I'm not going to explain how our brain works right now, I'll start
with processing power. so basically, how we.. processing power is calculated by the quantity of
operations that can be completed within a certain period of time. so how that is done, how that
is calculated, is very easy: you take... let's say you've got 200 transistors and I don't know
why I said 200, let's pick round numbers... let's say you have 100 transistors and each transistor
can do two--- can do 10 operations per second, well that means that overall your system can do 10
* 100 righ
t so 1000 operations every second. Okay, but how do you scale that? you can scale it by
getting more transistors onto your system so that every second with each transistor that does
10 operations well now you have 200 so now you can do 2,000 operations per second, so your system
is twice as capable or twice as smart as the other one. that's one way you can do it... but the
other way you can do it is uh by taking those 100 processors or 100 transistors and speeding them
up so that each of th
em can now do 20 operations per second, so now if you do the math it all comes
as a wash, both of those systems now can do 2,000 operations per second, both of the systems
are twice as capable as the previous one... and those are two very valid ways of increasing
capability right? so human (brains) currently work in a way that we have a lot a lot a lot
of transistors or connectors, our neurons. but since our neurons are working at speed-- at
chemical speed, these neurons are very slow, wher
eas computers are building neural
networks that have less "neurons" right, like we don't have computers right now that have
as many transistors or processor sorry processors right now that have as many transistors as there
are (neurons) in the human brain, not even close, but these processors and transistors operate
at orders of magnitude faster than our chemical brain, at the speed of light or electricity
actually, compared to speed of chemicals... so that's how uh computers-- I realize I'
m missing a
whole bunch of context, I'm all over the place... but that's the first I guess part of the equation
on how it's possible to make a computer that's as smart, as capable as a human or how to increase
the capability of your system... you can go the nature route which is make billions and billions
and billions and billions of transistors that connect to each other slowly AKA the way humans
(and all other biological beings) are getting around, (or the other way) that is by having
le
ss transistors or connectors but that talk to each other much much much faster... and in
the end as with the first example I gave there, it all comes out as a wash. that was weird
now sorry something fell on me, as things are melting right now, it's close to freezing
or close to the freezing point... so anyway, now I'm sure you're saying "okay but what does
that have to do with humans or how we understand, or how does that make a computer smart like a
human?" well now let's go to the second
part... so now that you understand kind of how to measure
the speed or the capability of a system, let's talk a little bit about how human brains work. so
you might think that your ideas come from nowhere, but you're probably not a brain scientist... what
has been found out in that field is that the way our brain works is that we have a whole ton of
neurons I don't know the number I know it's not even billions it's like orders of magnitude more
than that but the point is that we have tons a
nd tons of neurons, and these neurons --Mystra!
these neurons, the way they work is they start off uh kind of independent but as we grow and we
learn neurons build connections to other neurons. hopefully you've done enough high school biology
that this is a refresher and not new information for you, but if it is you can look it up yourself
the way it works right? you've got the nucleus of each neural cell, each neuron creates a
bunch of, I believe the term is dendrites, which are basically
the links, and these
dendrites connect to dendrites from other neurons, and that's how you know information and electrical
signals get passed on in our brain. oh I need to get out of here... this is way too dirty...
so that's how it works. now how-- and sorry it's a lot of information to try to organize,
especially assuming no prior knowledge-- so and Mystra is stuck with her leash, let me fix that...
Mystra, lift up your... yeah there you go, okay, Freedom! okay so where was I? yeah so tha
t's
the basic component of the brain is a neuron, and the neuron has connectors that connect to
other neurons, and how our brain works is that there's... it uses oh my I don't want to explain
ion channels and all that... the best way to exp-- the simplest way not the best way, the simplest
way to explain it is that each connection to a neuron or between neurons, those dendrites, if
you use it that path becomes reinforced. imagine a path that you're making through a forest:
the first time a
round there's brush everywhere, it's very difficult and you need to do a lot
of effort to get from point A to point B, but once you've gotten from point A to point B
once now, the path is a little bit clearer right? so the next time you go from point A to point
B it'll be a little bit easier, and then you'll clear a little bit more of the resistance, and
then the next time it'll be a little bit easier, and the time after that a bit more, and a bit
more until eventually the path becomes like
this, a nice paved Road, and the connection is super
easy. now the opposite is also true right? if there is a connection that is not useful and
you don't use a path that you don't walk, well eventually the forest regrows and the path
becomes much more difficult to pass again up until a certain point where if this path, or in
this case this connection between neurons, is not used for long enough it becomes trimmed and pruned
as it's in excess and not necessary and your brain doesn't want to
waste resources (energy/nutrients)
on maintaining it. so that my friends is how we learn, and this has been demonstrated. again I
won't quote the researchers and this and that, if you don't believe me you can look it up
yourself, but it's been shown that that's how all animals or all biological beings that have
brains like ours, that's how they learn. your brain starts when you're a baby by an explosion of
neuron connections that it creates, and eventually you reinforce these connections,
you make new
connections as you learn more stuff, create new paths in your brain, and eventually a brain wants
to be efficient cuz we--- our species did not evolve in our in the Society- environment that
we currently live in, not in abundance, and we evolved in a context where efficiency was key (for
survival). so anyway, point is that, at the basis of the principle, when you learn something your
brain is making a connection easier by reinforcing the ion channels between the two neurons, an
d
decreasing the activation threshold that will make one neuron fire. so I guess I didn't explain that
part, so I'll speak a little bit more about it. what determines whether that path actually works
and you go through and you can figure out how to add 2+ 4. it's... yeah, that's a bad example...
sorry, I'm trying to gather my thoughts... again, no script is an issue yeah so the way it works is
that as you-- from--- okay, how the connection is established or not is dependent on... it's kind
of like a light switch uh where on one side of the dendrite you will have some... I'm simplifying it
but a certain amount of positive ions let's say, and on the other side on the other dendrite for
the other neuron you'll have a bunch of negative ions and if you accumulate enough electrical
energy, or if when you activate the first neuron that activates enough positive ions that you
can make the connection, that you exceed (the activation threshold)... sorry I'm imagining a
bridge that's k
ind of being built from both sides, and if you put enough energy into the first side
that will allow ---and there's a little Gap at the top of the bridge-- sorry, I'm building the image
(in my mind) as I'm talking about it. so yeah there's a bridge, but there's a gap in the middle
of the bridge between the two neurons, and you're driving your little car for your idea(thought) of
trying to figure out "how do I add two and three?" and your little car, if you give it enough speed,
that or if t
he Gap-- sorry I'm really struggling with trying to make a clear image, and I'm writing
the analogy as I'm speaking... so okay, you've got a bridge with a gap, and whenever a neuron
activates it launches a car at 60 Miles an hour towards that bridge, and if you've reinforced or
you know uh added enough connection, or enough if you've reduced the resistance enough, so if you've
kind of shortened the Gap in the bridge enough that your car can jump and land on the other
side of the bridge, the
n the neuron on the other side of that bridge will now activate itself and
launch a 60 mph car down its neurons to all the other neurons to which it connects, and then you
know those 60 MPH cars will reach that next Gap or those next gaps, and if 60 Miles an hour is fast
enough to jump the hole well then that'll activate the next round down the line. if it's not then it
just stops, the signal stops there, and that's it. so that's how our brain transmits information. I'm
sorry for using this
explanation, I'm sure there's neuroscientists who could explain it much more
--with a much better illustration (analogy) than I can, but hopefully you get what I'm trying to
say. so as you learn things you're making the gap between the neurons (in that specific neural
path) smaller, so that the car can make the jump, and as you don't use a path you're making that
Gap wider and the car will just jump into the hole and the the signal, the potential doesn't
reach the next neuron. so that is e
xactly how resistors and transistors work right? transistors,
as a lot of people point out they're binary, they're one and zeros right? but what was a
big innovation that came with neural networks is they realized that "hey, we can put a variable
resistor or resistance in between each transistor that will determine whether when you activate--
when you switch that first transistor from 0 to 1, does that carry over to the next one or not, based
on the resistance". same Principle as I just sai
d, right? If the resistor is to-- if the resistance
is too high, the Gap in between the two sides of the bridge is too large and the 60 Mph car can't
make the jump, so the next transistor doesn't get activated. so that again is a mechanical version
of the way-- or artificial version of-- exactly how our neurons work! it's not just the one and
zero, yes or no, binary as in the old traditional systems of a computer as what your computer-- the
way your computer works right? now that's not how
neural networks work. neural networks go and they
add that component that changes the resistance, and it's a variable resistance, and it basically
acts the same way as our brain does. so what you do when you "train a system" is you put an
input, and then you just wait for the output of the system, and if the output is what you were
looking for then you give it a pat on the back and say "hey that was great, good job, you gave
me the right answer" and if it isn't well then you also tell it an
d you say "hey that's wrong,
you need to adjust something, try again!" and you don't tell it necessarily like in some older-- in
some paradigms you can or you do, but in general you don't tell it "hey go to transistor 76 and
reduce the resistance by three and run it again", that's not how it works or not how it should
work... the way it works is you just say "hey wrong answer try again" and the computer will
try again it'll change something you don't even know what necessarily but it'll cha
nge it and
then it'll try and it'll give you a new answer and again you give it more feedback... is this the
right answer yes? no? okay try again! or good job! (you do this for as many iterations as required)
and you know what, that's exactly... oh Daisy really wants to go home, sorry, I guess we're
going this way. that's the exact same way that a teacher will teach you how to add right? what's
1+ 1? 3? oh no sorry Timmy try again. what's one plus one? two? oh great, Timmy you got it right!
let's move on, or let's do another example. so... and you know Timmy or your neural network might
get the the correct answer for one plus one but then you train it more by giving it a new problem
right? what's 2 + 2? 2+ 2 is 3! sorry Timmy, no you're wrong try again! and it'll adjust and
Timmy will adjust the way he thinks, try again... just like your neural network will adjust its
resistance it'll try again and will try and see... hey also when I did... you know now that I've
changed the
se values,(the variable resistances) if I go back to the 1+1 am I still getting the right
answer? yes? no! oh crap okay well clearly what I changed wasn't the right thing to change so
let me try again Etc. and you do this, you iterate this millions and millions of times, and
that's how humans learn, that's how ants learn, that's how computers or neural networks learn.
sorry, I have to get Daisy in. I think it's been long enough of a walk, I guess this will be just
really part one. I wanted
to go a lot further, but consider this the intro on the AI and how
---but more about how human and natural brains work and how you can emulate that in a computer,
which will be the basis for part two tomorrow, explaining a little bit more about how AI actually
works and why it can learn the same way we can. so thanks for listening, sorry this was a little
bit convoluted... it's a complex topic that has a lot of background knowledge required and I
struggled to summarize what I was trying to
say, but I hope you got something out of it, and
please look forward to part two tomorrow where I'll explain-- go a little bit more about actual
AI now that we've got the background out of the way. thanks for listening, please give me all
your comments suggestions complaints and insults, do all the Social Media stuff. I appreciate your
time and thanks for joining me on the dog walk!
Comments