Latest / Project 2028 / Artificial Intelligence- the Pros, the Cons, and the Politics

Transcript

Artificial Intelligence- the Pros, the Cons, and the Politics · Project 2028

0:09
All right, welcome to Project 2028. We have a
0:11
special episode today specifically focusing on
0:15
AI, AI's infrastructure demands, some of the
0:19
geopolitics, the Biden spectrum program, just
0:22
a lot of tech stuff and that crossing over into
0:26
politics today and how it affected the election
0:29
and specifically climate change and a lot of
0:32
the imperatives surrounding that and surrounding
0:35
the policy that the Trump administration... is
0:37
now pushing so let's get right into it So in
0:46
my opinion, one of the biggest technological
0:49
advancements in the past couple years has been
0:54
kind of AI coming front and center in a lot of
0:58
areas. And it's come with a lot of benefits,
1:02
obviously. It's helping people become more efficient,
1:06
but it's also come with a lot of risks to jobs
1:10
and the environment. So that's going to be what
1:13
we're talking about today. So some of the pros
1:18
of AI are that it can increase productivity and
1:22
it can help with research. It can... It can be
1:30
used in things like Netflix, Spotify to like
1:34
adapt and help give better recommendations for
1:37
what you need. It can help advance in health
1:40
care, safety and security. It's constantly available
1:44
and can help solve complex problems. So the pros
1:48
are definitely things that help make us more
1:50
efficient. But one of the major cons is how it's
1:54
affecting the environment and that it requires
1:58
energy. to use and tons of water to cool off
2:03
the um servers generators oh sorry what no sorry
2:08
uh servers um they basically they run massive
2:12
data massive server rooms um they take tons of
2:16
energy i mean you have data centers um where
2:19
they're building models obviously we're at gpt
2:21
4 .1 4 .5 There's a lot of different models out
2:26
there. But a GPT -3 scale model, one that has
2:29
a lot less parameters, is a lot less, parameters
2:32
are like the way they fine tune the model, a
2:34
lot less advanced. It took up 1 ,287 megawatt
2:39
hours, equivalent to what I think is around,
2:43
from my math, 100 and something annual use of
2:48
the U .S. home's energy. When you think about
2:51
it, it's not that much. It's a few blocks to
2:54
train a world -class AI model. But then again,
2:59
it's still a lot of energy. It's very expensive
3:02
to generate that energy. But yeah, sorry, LaChan.
3:07
No, those are all great points. I think with
3:10
AI, one of the main things it comes down to is
3:12
weighing the costs versus the benefits. I know
3:16
personally... I would say that I am a person
3:19
who feels really strongly about caring for the
3:21
environment, protecting the world we live in,
3:24
and the world that we're going to pass down to
3:27
future generations. But I will also say that
3:30
I am a person who has found AI very useful. I
3:33
wouldn't say that I'm using it every day. I wouldn't
3:36
say that I rely on it. But I would say that I
3:39
think, especially as a busy student, I think
3:42
some of the tools are helpful, not for writing
3:45
essays, but... It can help you make study guides
3:48
and flashcards. If you input all your notes,
3:51
it can give you quiz questions. So those are
3:54
the things that I've found most helpful. But
3:56
I also understand that if everyone relies on
4:00
AI, we will fall. I think it'll start to have
4:04
a bigger impact than we could realize. Yeah,
4:07
I mean, of course, this episode you're listening
4:11
to right now has been run through. Not intentionally,
4:13
but a lot of AI algorithm to compress it to,
4:17
you know, I just paused, right? It cuts all that
4:21
kind of stuff out. There's a lot of machine learning
4:23
that goes into it. And that's just listening
4:26
to a podcast episode. I mean, when you look at,
4:30
you know, more complicated things, if just the
4:32
CEO of Shopify the other day, I heard this in
4:35
a podcast I listened to. He's now saying that
4:37
they can't hire an employee. And if they do,
4:39
they have to prove why they're... the job couldn't
4:43
be done with AI and obviously Shopify is a web
4:45
company so a lot of their stuff has to do with
4:49
coding and with you know back -end engineering
4:54
of the websites that they host and that they
4:57
make so obviously you know it's kind of imperative
5:02
for them to have the efficiency and have you
5:06
know the frontier ai models instead of having
5:09
maybe a person because an ai model costs a lot
5:12
less than a person especially if they can do
5:13
roughly the same job um i've listened and i've
5:18
heard most coders say that you know ai is not
5:21
as creative it can't do the same things um it's
5:24
limited in the actions it can take it's not an
5:26
agent it can't go out there and it can't do stuff
5:29
it can't sign up for things because obviously
5:31
there's hosting stuff and all this um And it
5:34
can't do that. It's very limited. It can make
5:36
you the code, but then you got to plug it in
5:38
and you have to do all that stuff. But it just
5:41
kind of shows how far, you know, AIs came and
5:45
how much, you know, how much less the inference
5:48
costs than it used to. AI costs used to be so
5:51
large. But, you know, the data center draw right
5:55
now is higher than ever. It's 1 .5 % of the world's
5:59
electricity goes into data centers. Now, albeit.
6:02
Not all AI. All of your information is on the
6:07
cloud. It's hosted in data centers, which is
6:09
a lot of it. This podcast, for example, hosted
6:12
on the cloud. But a lot of everything is just
6:17
put into these data centers that also have AI
6:20
running in them. Amazon is consistently building
6:24
AI centers throughout the United States just
6:27
to keep up with demand for their web services
6:30
like AWS. I think that's, I think that's a great
6:35
point. I think it's just in, it's more present
6:38
than you would think, you know, it's not just
6:39
like, oh, I only use AI when I go on chat GPT.
6:43
It's like very involved in daily life. Even when
6:48
you Google something, like it used to be the
6:50
first thing that comes up as a website, but now
6:52
the first thing that comes up is an AI overview.
6:55
And, oh yeah, sorry. No, sorry. But if I'm Google,
6:58
like I. I have a lot of like weird queries where
7:02
I have to explain it. A lot of it's just about
7:04
like something I want to know. The other day
7:06
I wanted to find a specific Federalist paper.
7:08
You can't Google that in natural language. It
7:12
was a random query, but I needed to know it.
7:14
So I asked ChachiBT, but I've been going more
7:17
and more increasingly to Google because obviously
7:19
it now can process natural language, which...
7:24
There's a specific language you used to have
7:25
to put into Google. I only did it for a few years
7:28
until AI was created, but it's the keyword search.
7:31
But now you can type stuff in the way me and
7:35
you speak to each other, and it will come up
7:37
and populate 90 % of the time with the correct
7:41
answer. And obviously Google's had some problems.
7:43
It's told people to eat rocks and to put glue
7:47
on their pizzas. These are actual things that
7:48
happen. Concerning. But, you know, the 95 % hit
7:56
rate on just allowing it to happen is good. And
7:59
additionally, you know, it now declines for a
8:03
lot of questions, especially if you get into
8:05
political science and political matters. It shuts
8:09
off because Google's had a lot of problems, like
8:11
when they wouldn't generate accurate founding
8:13
fathers without making them diverse, which everybody
8:17
got mad about because Google was kind of just
8:21
a weird moment. But, you know, I think Google's
8:26
handling it better than they used to. I think
8:27
they're turning it off for some things. But on
8:31
the environmental point, you know, if you look
8:33
at cooling, obviously the water's not disappearing.
8:36
It's just going up into the atmosphere. Or it's
8:40
just hot water. But they consume billions of
8:43
liters of water annually. They burn billions
8:47
of kilowatt hours on chilling. And chillers,
8:50
which are just like cooling systems for these
8:52
AI plants. And obviously, it's, you know, if
8:57
you have a thriving ecosystem based around the
9:00
place, it can be mildly destructive occasionally.
9:04
But this water is not disappearing from the earth.
9:06
It's only going up kind of in an expedited water
9:10
cycle. But, you know, a lot of people have kind
9:14
of been, especially the water use particularly,
9:16
people have made many comments. on just how destructive
9:20
it has been in their communities, especially
9:22
near the data centers and stuff. And then just
9:27
a few things. Now, all these companies, AWS,
9:30
Amazon, which obviously builds almost all of
9:34
their data centers in the United States, obviously
9:35
they host stuff outside where it is cheaper.
9:39
But Apple as a Oracle, this is just basically
9:43
a data center place in that they're making more
9:47
renewable powered sites. waste heat reuse because
9:51
there are a lot of heat. Heat's very useful,
9:53
especially to create energy for use for energy
9:56
for district heating and, you know, model efficiency
9:59
research. This was the FCC study commission paper
10:04
that they did a year ago, year or two ago. And,
10:08
you know, the more efficient that those can get,
10:10
I believe, you know, the better it will, it'll
10:14
only get better for AI's. And as they get more
10:16
efficient, it'll obviously be, easier to train
10:18
models um but to kind of transit sorry lucia
10:22
oh no no sorry that was just me um but to kind
10:26
of transition to efficient models um to kind
10:29
of put a spotlight on arguably the the thing
10:32
that kind of took the political spotlight and
10:35
um lucia have you you've heard of deep sea yes
10:39
i mean it caused the stock market uh You know,
10:42
a bit of upheaval. NVIDIA and a few tech stocks
10:45
like on the QQQ board or whatever, but yeah.
10:48
I don't think I'm as knowledgeable about it as
10:53
you are, but I've definitely heard a bit. Yeah.
10:56
Well, a little quick overview. If you haven't
10:58
heard of it, you know, it was a Chinese AI model.
11:02
Oh, backtrack. A Chinese fund had kind of just
11:07
a side project building AI model. NVIDIA chips,
11:11
which are used to make these AMLs, are heavily
11:13
restricted coming out of the United States. Like,
11:15
heavily. Like, they can't sell them to China
11:18
for this reason. And DeepSeek used these crappy
11:22
chips from NVIDIA, which is the AI subsidiary
11:25
of this investment company, and they managed
11:27
to make a model that is no longer, because AI
11:30
is going so fast, but a few months ago was the
11:33
top model in the world. And it was open source,
11:36
so anybody could download it if you had... Not
11:38
a MacBook or something, but if you had multiple
11:40
servers and enough space and storage and RAM
11:44
to run it, you could, hypothetically, for free.
11:47
And you would run it yourself. Nobody would get
11:51
your data. Obviously, there's a lot of national
11:54
security implications. I asked it personally
11:56
many times, and it can think. It'll show you
11:59
what it's thinking about. AI does this now. This
12:01
is why it was so interesting. ChatGPT could do
12:04
that, but it costs you $200 a month and whatever.
12:07
And I asked it many times to tell me what happened
12:10
at Tiananmen Square. And it refused. And actually,
12:13
I saw, I took a screenshot of it. And it thought
12:15
to itself, it actually said, my people won't
12:20
let me think about this or won't let me comment
12:22
on this because I live in or was made in China.
12:27
Obviously, if you ask ChatGPT or Gemini. they'll
12:29
be more than happy to tell you the full history
12:31
of what happened in Tiananmen Square. But Luciana,
12:36
any thoughts on just kind of where that is? I
12:39
think that is actually so interesting. I don't
12:45
know. I think that's very interesting. I think
12:47
as AI gets more and more advanced, it'll be very
12:50
neat to see what it does and doesn't say in regards
12:56
to like... political things or how it's different
12:59
coming from different places in the world. But
13:02
I actually, I didn't know that it, I didn't know
13:05
that story that you told. I think that's very
13:07
interesting. Very. But also, as you said, the
13:13
political implications are quite massive. If
13:18
you look at like, just one of the biggest things,
13:21
TSMC, obviously not a US company, it's in the
13:24
name, Taiwan Semiconductor Manufacturing. um
13:27
corporation they in the united states and taiwan
13:32
well that's kind of how taiwan's made they i
13:34
mean tsmc makes 94 of the world's leading chips
13:37
um more nodes than anyone else it's it's incredible
13:42
it's a it's a company that single -handedly supports
13:44
uh much of the economy in Taiwan. And that's
13:48
strategically why we want to keep Taiwan independent
13:52
of the CCP and why the Biden administration did
13:56
that. And we'll talk about the Biden administration
13:57
a bit later with technology. But if you look
14:01
at TSMC, they're building new fabs, which are
14:04
places because Apple makes a chip and then they
14:07
give it to TSMC or whoever they want. TSMC is
14:10
usually it. TSMC actually makes the chip. Apple
14:13
doesn't make a single chip. And they're building
14:16
a new fab in Arizona that builds the chips, which
14:21
will deliver 30 % of TSMC's top chips or technically
14:26
top nodes, which are the way they make the chips.
14:29
But, you know, concentrated supply chains in
14:32
Taiwan have been kind of dispersing very quickly.
14:36
The U .S. Chips Act, obviously, of the Biden
14:39
administration. granted $39 billion in grants
14:43
to this kind of, you know, diversification. Obviously,
14:49
tariffs had a big implication before Trump took
14:52
off much of the semiconductor tariffs. But Luchana,
14:57
what's your thought on domesticating and the
15:01
security in Taiwan? It's kind of a multifaceted
15:05
question, but anything? I'm going to be completely
15:09
honest with you. I personally don't think I know
15:12
enough about security in Taiwan to give you a
15:15
good answer on that. No, that's fair. I mean,
15:20
Pete Higgs could barely answer it. It's a hard
15:22
question. U .S., obviously. My favorite guy,
15:27
Alex Wong, was not in Taiwan, but he was in North
15:31
Korea and South Korea variously throughout the
15:34
past few years, dealing with national security
15:35
there. But it's an interesting environment. China
15:39
has obviously, they run practice missions right
15:46
outside Taiwan that are just openly aggressive
15:51
and basically just showing that they are ready
15:55
to take Taiwan at a moment's notice. Biden was
15:58
very strong in saying that we will defend Taiwan.
16:01
Trump, a bit less so, but obviously it is a national
16:05
interest of ours to make sure that This wouldn't
16:08
follow it or it would it would change the paradigm
16:11
of AI. No, I think I think you've helped to inform
16:18
me on that. I still I think I should still do
16:24
research. I think that because it's a topic that.
16:27
I don't know, maybe some of our listeners won't
16:29
know about. I can link some resources in our
16:32
episode descriptions. But I do think that's very
16:35
interesting. Unless you have more to say about
16:38
that, I have a couple questions for you just
16:42
regarding the future of AI. So I feel like, and
16:47
I don't know, there's a chance I haven't really
16:49
grown up in a time where there have been a ton
16:50
of new advancements. When we were born, we pretty
16:55
much already had TVs, iPhones, iPads, computers.
17:01
But when AI came out and ChatGPT and all that
17:05
stuff, I feel like it was just a very new thing
17:09
and it advanced so quickly. It's really changed
17:13
a ton in the year, two years? I don't know. Two
17:18
years. I think it's shaped a lot of systems,
17:23
especially in, like, schools. It changed a lot,
17:26
but it also, with regards to plagiarism, but
17:30
it also really helped students like myself who
17:34
use it as a study tool. I also think it's making
17:37
a lot of changes in the job industry as well
17:41
as among artists. There's a rise of AI art. which
17:46
is hurting artists because you wouldn't commission
17:51
a drawing when ChatGPT could make you one where
17:55
maybe it's not quite as high quality, but...
17:59
less expensive, more convenient. So I think AI
18:02
has really made a ton of changes and the technology
18:05
itself has developed a lot recently. I'm just
18:08
curious to see how you think it's going to continue
18:11
to evolve and what the big changes that come
18:14
next will be, whether that's that the systems
18:16
itself get more intelligent or that it becomes
18:19
more and more prevalent or what. I think the
18:23
systems are, of course, getting intelligent.
18:27
three weeks um i would say check up on a thing
18:29
called uh lm arena it's the universal measurement
18:33
of ai and there's a different leader every five
18:36
days because these things come out so quickly
18:38
openai 4 .1 came out the other day meta's new
18:41
llama model um came out the other day um and
18:45
there was some controversy because the benchmarks
18:47
were correctly you know it was it was kind of
18:50
a weird thing um so it's now not in the lead
18:53
because they took it off the board um because
18:55
of some fraudulent behavior to say the least
18:57
um but you know if you just i i think a lot of
19:02
it is going to depend on some court cases right
19:04
now the new york times uh vox media um and a
19:08
few other companies are are um suing open ai
19:11
and others because you can just ask ChatGPT to
19:15
go through a paywall for you because it scrapes
19:17
all the data. And if you want a painting generated
19:21
by a specific artist, it'll give it for you.
19:24
Now, did that artist ever give OpenAI any money?
19:29
Or did OpenAI give them any money? No. So I think
19:33
the implications of a few of these court cases,
19:35
especially when it comes down to trademarks and
19:37
intellectual property, In kind of just the whole
19:41
sphere of IP, because when you look to Google
19:43
20 years ago, and this is based off what I've
19:49
read, it was a lot like this in which there was
19:52
a lot that was just being put out there without
19:56
the explicit permission of the author or person
20:01
who disseminated the information originally.
20:03
So I think those cases in the upcoming years,
20:07
and these are going to take... Years to litigate.
20:09
I mean, there's a good chance I'll be done college
20:12
before some of these cases will be finished.
20:17
Will, you know, take time, take a lot of time
20:20
to litigate, to say the least. Yeah. What, kind
20:26
of jumping off of that, what jobs or professions
20:29
or industries do you think will be impacted the
20:32
most, if you had to, like, name a few specifically?
20:38
I mean, obviously, just to look at the Shopify
20:40
example, it's mainly the ones that built it.
20:43
The coders, the people who are building it, actually,
20:51
because there's no physical work involved. They
20:54
type keys on a keyboard, which AIs are getting
20:56
increasingly good at. They're getting increasingly
20:58
good at going around the web and doing stuff
21:00
for you. They're not foolproof, but they're getting
21:04
much better at it by the day. And as they reach,
21:08
you know, a sort of super intelligence, which
21:11
is when AI will finally be able to be indistinguishable,
21:14
if not smarter than human brains, which is the
21:17
goal of open AI. And that's when you'll see,
21:21
I think, a shift. I think, look, AI, there's
21:26
some jobs that AI can't do and won't be able
21:30
to do. And I think those jobs revolve more around
21:34
manual labor than anything else. But I think
21:38
if you look at it in 34 years, if there's more
21:41
AI automation, it's just going to be more prosperous
21:44
for society, hopefully, because labor will be
21:49
much – or high -demand labor. But if you look
21:53
at politics or if you look at law, these are
21:56
things that are subjective, that require not
22:00
just a computational view but a subjective view
22:03
of the world. I don't think AI will ever be able
22:06
to place, especially politics, because it doesn't
22:09
make any sense. You might want an AI model being
22:13
present if it was smart enough, but it would
22:17
never happen and it wouldn't be legal in any
22:20
regard. But there are some things that only humans
22:24
can do. I think if you look at that kind of thing,
22:27
I think that those are very foolproof against
22:29
future threats. Yeah. I agree. Well, yeah, and
22:36
the last thing I kind of want to talk about,
22:38
and then Luchana, you can kind of finish up,
22:42
is, you know, another thing that had a lot to
22:46
do with AI, not necessarily fully AI, but technology,
22:50
kind of coming after the CHIPS Act, kind of part
22:53
of Biden's signature, you know, AI and tech bills,
22:57
which... one of them was the CHIPS Act, was Biden's
23:02
Spectrum program. Now, if you know about Ezra
23:04
Klein, you know he has basically been, the past
23:06
few weeks, going around and telling everybody
23:08
how horrible this program was. And it was really
23:12
bad. Three years of study. The strategy has done
23:20
nothing, like zero. There is no internet that
23:23
this thing has done. And it's basically... it
23:27
was to make spectrum which is internet connection
23:30
go throughout the rural united states and just
23:32
through the united states um you know the fcc's
23:37
authority lapsed on march 2023 um stalling all
23:43
the 5g like releases um and then there was the
23:48
auctions that just weren't working anyways it's
23:50
a long story but basically there was a lot of
23:53
money 15, 16 billion dollars and stuff that has
23:57
just not been used for anything. And it was all
24:03
an internet plan and it's, you know, it's debatable
24:05
whether that was intentional or kind of a subproduct
24:12
of the way that it was written because, you know,
24:17
it kind of gave deference to the spectrum providers
24:20
already, not, you know, a government subsidized
24:22
effort. but it's kind of boring, so I'm going
24:25
to finish up. But, you know, it's been heavily
24:28
criticized, even by people on the left who really
24:30
supported it and thought the Biden administration
24:32
did make, they made a big misstep on this effort.
24:40
Is there anything else you want to kind of talk
24:42
about? I think that's all I've got. We have covered
24:45
some of the topics that I was really interested
24:48
in, and I think the main takeaway is that The
24:51
entire time we've had AI, it's been constantly
24:54
developing. Like you said, there's a new top
24:56
AI model every few days. It's going to keep developing,
25:01
which means that the things it can do, the jobs
25:05
it could replace, the environmental impacts it
25:07
could have are also going to keep changing. So
25:11
it's important to be consistently weighing the
25:14
risks versus benefits for us as a society of
25:17
using it. Yeah, I think that's why we take away.
25:24
And also, if you're looking at the transcript
25:26
at all this time, if you're on Apple Podcasts
25:29
and Spotify, that was generated by AI. We do
25:34
that personally through our plan, and Apple does
25:36
it personally too. So, you know, another pervasive
25:41
AI, just something so simple as an RSS -based
25:46
podcast. It just kind of shows how wide it's
25:50
spread. And obviously, Apple's trying to incorporate
25:51
AI into everything. I think when you look at
25:54
the future, I think it's hard not to see AI.
25:58
It's hard not to see AI in politics, for media
26:02
and marketing. When you look at the big 2016
26:08
revelation with Facebook, with Britannica. Oh
26:14
my God, how did I just forget that? What's the
26:15
name? Not Britannica. Encyclopedia? No, no, it's
26:20
not an encyclopedia. Sorry, it was a data company.
26:22
I read the book. No, sorry, I forgot what it
26:26
was. But anyways, it was this huge revelation.
26:29
And, you know, it was big for, and it was kind
26:34
of a revelation in social media for politics.
26:36
And I think that is where AI is kind of going
26:41
to. So we'll wrap it up. Sending you the outro.
26:50
All right. Thank you for listening. It was Cambridge
26:52
Analytica. But we hope you had a bit of enlightenment
26:56
about AI. It was quite a short episode, but we
26:59
thought we would just give a bit of a brief on
27:01
the implications in politics and, you know, where
27:05
we're going from here. But I'll send it over
27:06
to Luciana for the contact information. We hope
27:11
you loved this episode. Please tell your friends
27:13
about us. We are Project 2028. You can listen
27:16
to us wherever you get your podcasts. Please
27:19
also follow us on Instagram. We are at project2028
27:23
.podcast. There we will post important political
27:27
news in real time along with updates about current...
27:33
episode release dates, and future episodes. And
27:36
if you have any questions, comments, or concerns,
27:39
you can reach out to us via email. We are projects2028
27:43
.pod at gmail .com. Thank you so much and have
27:47
a great day.