Latest / Journey to the Fringe / Fringey Mini: Thirst Trap Tech

Transcript

Fringey Mini: Thirst Trap Tech · Journey to the Fringe

• 0:00
  Okay, Chelsea, I am going back to giving you
• 0:02
  a pick with the vaguest idea of what we're going
• 0:05
  to be talking about in each of the articles.
• 0:07
  I love that for us. One is Epstein adjacent.
• 0:12
  One is fascism adjacent. And one is AI adjacent.
• 0:20
  All of them seem pretty depressing. I'm going
• 0:24
  to go with AI and hope that it is about robots.
• 0:29
  Okay. Well, I will tell you that the Epstein
• 0:32
  one was probably the least depressing, but...
• 0:36
  Really? The... Yeah, that one's on the New York
• 0:38
  Post, so I don't really want to support it either.
• 0:41
  Okay, well, fuck them. Okay, and the depressing
• 0:45
  news coming out of the AI world right now, coming
• 0:47
  from elpais, E -L -P -A -I -S dot com, written
• 0:53
  by Manuel Pascual, on June 3rd, 2026, article
• 0:58
  title, AI will consume as much water in 2030
• 1:01
  as 1 .3 billion people. That's insane. By 2030,
• 1:06
  water consumption linked to the use of artificial
• 1:09
  intelligence will be equivalent to that of 1
• 1:11
  .3 billion people in sub -Saharan Africa, while
• 1:14
  it will require nearly three times the annual
• 1:16
  energy consumption of Pakistan, Bangladesh, and
• 1:19
  Nigeria, countries with a combined population
• 1:21
  of 650 million people. And water shortages, no?
• 1:25
  Yeah. I think these are just like comparisons
• 1:28
  of these people live on such little amounts that
• 1:31
  it's easy to compare how far things can go. I
• 1:34
  don't think they actually mean that's where it's
• 1:36
  going. Just the comparison. In terms of carbon
• 1:39
  emissions, these could reach 400 million tons
• 1:41
  of CO2 equivalent compared with the United Kingdom's
• 1:44
  total emissions. The operation of AI will require
• 1:46
  14 ,500 kilometers of land, including infrastructure
• 1:50
  and supply chains, twice the size of the Jakarta
• 1:53
  metropolitan area. A megacity with more than
• 1:56
  32 million inhabitants, or 10 times that of Mexico
• 1:59
  City. These are some of the figures cited by
• 2:01
  the author of a report published this Wednesday
• 2:03
  by the United Nations University Institute for
• 2:05
  Water, Environment, and health in addition to
• 2:08
  these projections based on conservation growth
• 2:10
  estimates the report also contains striking data
• 2:13
  about the current situation if the data centers
• 2:16
  powering ai were a country their present electrical
• 2:19
  consumption 448 terawatt hours would be on par
• 2:23
  with that of france So if AI was its own country,
• 2:26
  it would have the energy equivalent of France.
• 2:28
  That's fucked. The institution had previously
• 2:30
  published reports warning about the carbon emissions
• 2:32
  associated with the growing use of AI. On this
• 2:35
  occasion, researchers have also taken into account
• 2:38
  the energy and water consumed by the data centers
• 2:40
  that power AI. In the case of water, this includes
• 2:43
  both cooling systems and electricity generation.
• 2:45
  Quote, this report is not a case against artificial
• 2:48
  intelligence. What? End quote. Said Professor
• 2:52
  Kevin Madani. the director of UNUINWEH, which
• 2:57
  is a terrible acronym. That's the Institute of
• 2:59
  Water Environmental Health. Quote, it is a call
• 3:02
  for using it responsibly and addressing its unintended
• 3:04
  impact proactively to make it sustainable and
• 3:08
  equitable. We have a narrow window to ensure
• 3:11
  that the backbone of the technological revolution
• 3:13
  of our era develops within planetary limits.
• 3:28
  End quote. And I should have said there was a
• 3:33
  break in the quote because this is a person who
• 3:34
  said that. I did not expect that. Said Xiaolei
• 3:37
  Ren, professor of computational engineering at
• 3:40
  the University of California, Riverside, and
• 3:42
  an AI sustainability specialist who did not participate
• 3:44
  in the study. An AI sustainability specialist.
• 3:48
  Yeah. I did not know that was a thing, and I
• 3:51
  don't think it was a thing five years ago. Nope.
• 4:09
  Quote. Quote. every kilowatt hour of electricity
• 4:17
  used to train or run an ai model carries environmental
• 4:20
  footprints including carbon footprint from the
• 4:22
  generation mix a water footprint from electricity
• 4:24
  production and cooling and a land footprint for
• 4:26
  energy infrastructure reservoirs and fuel extraction
• 4:30
  The carbon footprint can vary up to 70 % if,
• 4:33
  for example, coal is replaced by bioenergy as
• 4:36
  the source of electricity powering AI. However,
• 4:39
  this would in turn increase the water footprint
• 4:42
  30 -fold and the land footprint 100 -fold. The
• 4:45
  complexity of manned AI's environmental impact
• 4:47
  is therefore extremely high. Low emissions do
• 4:50
  not equate to lower water consumption or low
• 4:52
  land use. Assessing AI's environmental impact
• 4:55
  using a single metric can obscure its harmful
• 4:57
  effects and shift them to other regions. Quote,
• 5:00
  if we keep judging AI sustainability by carbon
• 5:03
  alone, we might think that renewables make AI
• 5:05
  infrastructure clean, but that is solving only
• 5:08
  one problem while creating other problems often
• 5:10
  in places that didn't ask for it. End quote,
• 5:13
  explained Miriam Axel, the study's lead author.
• 5:16
  The report also draws another interesting conclusion.
• 5:19
  Until recently, the prevailing consensus was
• 5:21
  that most of the energy consumption associated
• 5:23
  with the AI model occurred during the training.
• 5:26
  Did I not just read that? That's okay. I'll keep
• 5:30
  going. I don't know. Associated with. associated
• 5:35
  with an AI model occurred during the training
• 5:38
  phase. That is before it's used by the public.
• 5:40
  However, Axel's team's data challenged this view.
• 5:43
  Inference, the computations carried out every
• 5:46
  time a user submits a query so the model can
• 5:49
  respond, account for the dominant share between
• 5:51
  80 and 90 % of total consumption. The success
• 5:54
  of these tools used by hundreds of millions of
• 5:56
  people every day has reversed the balance. Researchers
• 6:00
  also assess the energy consumption linked to
• 6:02
  different uses of AI. A standard conversation
• 6:05
  with a chatbot such as ChatGBT or Gemini uses
• 6:08
  200 times more energy than a basic AI function,
• 6:11
  like classifying suspicious emails into spam.
• 6:14
  Interesting. Yeah. using as a baseline generating
• 6:17
  synthetic images consumes 1400 times more but
• 6:21
  a short video can require 200 000 times more
• 6:25
  energy this is one of the most comprehensive
• 6:27
  technical reports on the environmental impact
• 6:30
  of current ai systems but the conclusion focus
• 6:34
  on the impact of gbt4 which is a model from more
• 6:37
  than three years ago And three years in the AI
• 6:40
  sector is an eternity, said Alex Hernandez, a
• 6:44
  researcher at the Quebec AI Institute, led by
• 6:46
  Joshua Bengio at the University of Montreal,
• 6:49
  who did not take part in the study. That the
• 6:51
  report's conclusions are based on data from older
• 6:55
  models, Hernandez says, speaks to the sector's
• 6:57
  lack of transparency. Quote, the main limitation
• 7:00
  of the study is the difficulty of obtaining concrete
• 7:03
  data on the consumption of current systems, end
• 7:05
  quote, he added. That's probably far enough.
• 7:08
  There is a bit more to the article. We'll link
• 7:10
  it in the description so you can read it. But
• 7:13
  that is crazy. And one of those things too, like
• 7:16
  it's not just water they're talking about, but
• 7:17
  like when they're using the water, it is water
• 7:20
  that we had to treat to make usable by humans
• 7:23
  like us. The first world people who enjoy very
• 7:26
  clean water. No, I'm not happy with that article
• 7:29
  at all. It is infuriating with something that
• 7:32
  we don't even need that is ruining future generations,
• 7:36
  in my opinion, because so many people are relying
• 7:39
  on it and using it instead of thinking for themselves.
• 7:42
  And I think a part of humanity and its creativity
• 7:45
  and our ability to think is something that is
• 7:48
  so beautiful about humanity. And we're just like...
• 7:52
  doing this yeah and like the AI just sucks up
• 7:55
  to you it's just sucking up to you like look
• 7:58
  at that yeah and yes disclaimer all of our images
• 8:02
  are generated by AI that is just something we're
• 8:05
  not good enough at art nor rich enough to hire
• 8:08
  someone to do the art we want made but it is
• 8:11
  what it is if that were to go away I would be
• 8:13
  perfectly okay with just not putting art yeah
• 8:16
  but like I don't like it. And the best case scenario
• 8:21
  for AI, like the thing that they're pushing for
• 8:24
  is it takes everybody's jobs. Yeah. And there
• 8:27
  doesn't seem to be any planet what to do with
• 8:29
  it at that point. And it takes away our ability
• 8:31
  to think. And when like new generations coming
• 8:35
  have this as a tool, you don't even know how
• 8:37
  that's going to impact society in the future.
• 8:39
  No, younger generations aren't using it as much
• 8:41
  as us. Really? Sorry, I shouldn't say us as the
• 8:44
  older generations. Yeah. Oh, you should see the
• 8:46
  backlashes at universities right now. Well, I
• 8:49
  do know. They are pissed with the job prospects
• 8:52
  that AI have left them with. A lot of Facebook,
• 8:55
  the people still using Facebook are boomers.
• 8:58
  Yeah. And a lot of like our generation, younger
• 9:01
  generations are moving away from Facebook. So
• 9:03
  actually, now that you say that, I believe that.
• 9:05
  But I think it's taking so much away from humanity,
• 9:09
  from the earth. And I think that it's not something
• 9:14
  that people necessarily know the implications
• 9:16
  for because it's not right in front of their
• 9:18
  face, which can be said for a lot of what's happening
• 9:21
  with climate change. Yeah. With that, though,
• 9:24
  you guys got 48 hours. Is this, like... Go make
• 9:28
  a picture. I mean, it's fun. Yeah, it's terrible
• 9:30
  for the environment. It's fun to make those pictures
• 9:32
  sometimes. I will say that. Go make the Bigfoot
• 9:35
  Orca do some fun stuff. Although, man, it took
• 9:38
  me so many prompts to get it to make that genital
• 9:40
  picture. Okay, last time I'll mention that for
• 9:43
  a while. What genital picture? I haven't even
• 9:46
  seen it yet. The Bigfoot one that has come out.
• 9:48
  Not at this point in us recording it, but it'll
• 9:51
  be out on a previous episode. Oh, it will. I
• 9:53
  have a fun fact, though, to help us recover from
• 9:56
  this. The website that you so beautifully said
• 9:59
  the name of. El Pais. Yeah. El Pais means the
• 10:03
  country in Spanish. Oh. And now I've just converted
• 10:06
  some beautiful Spanish into Everybody's Wednesday.
• 10:10
  Look at that. So I thought I'd end with that.
• 10:14
  Bye. Bye.