1
00:00:00,160 --> 00:00:03,600
Artificial intelligence tools 
are transforming the way that 

2
00:00:03,600 --> 00:00:07,400
people find information on the 
Internet, and this change is 

3
00:00:07,400 --> 00:00:10,120
happening faster than publishers
can adapt. 

4
00:00:10,400 --> 00:00:13,600
When users question chat bots 
rather than using search 

5
00:00:13,600 --> 00:00:17,160
engines, they're given answers 
rather than links to follow. 

6
00:00:17,400 --> 00:00:20,680
And this is changing the 
economics of the Internet, and 

7
00:00:20,680 --> 00:00:24,960
in particular, the economics of 
news providers who spend money 

8
00:00:24,960 --> 00:00:28,600
gathering information that they 
can no longer monetize. 

9
00:00:28,960 --> 00:00:32,479
Over the last few years, 
millions of users have switched 

10
00:00:32,479 --> 00:00:35,920
from search engines to AI chat 
tools for research, 

11
00:00:35,920 --> 00:00:38,600
recommendations and real time 
answers. 

12
00:00:38,920 --> 00:00:43,280
Tools like ChatGPT, Clawed, and 
Perplexity are now directly 

13
00:00:43,280 --> 00:00:46,880
answering questions that 
wouldn't send readers to primary

14
00:00:46,880 --> 00:00:50,080
online sources that they felt 
they could trust. 

15
00:00:50,280 --> 00:00:54,520
As users drift from trusted news
sources, they place growing 

16
00:00:54,520 --> 00:00:58,960
faith in AI systems trained to 
mimic authority tools that 

17
00:00:58,960 --> 00:01:02,480
scrape the web for answers but 
offer no accountability. 

18
00:01:03,160 --> 00:01:07,920
About a year ago, Google rolled 
out AI Overviews and then more 

19
00:01:07,920 --> 00:01:12,840
recently, AI Mode features that 
answer questions directly on the

20
00:01:12,840 --> 00:01:16,480
Google search page, often 
without crediting the original 

21
00:01:16,480 --> 00:01:19,720
sources. 
For users, this feels seamless. 

22
00:01:19,920 --> 00:01:22,920
They don't have to constantly 
refine their search terms to 

23
00:01:22,920 --> 00:01:26,160
find the information they were 
looking for, and they don't have

24
00:01:26,160 --> 00:01:29,840
to read 10 different articles to
work out an answer to their 

25
00:01:29,840 --> 00:01:32,520
question. 
But for publishers, this has 

26
00:01:32,520 --> 00:01:37,120
been designed Asterisk, a report
by Ender's analysis based on 

27
00:01:37,120 --> 00:01:41,520
Systrix data, was released a few
weeks ago showing that news, 

28
00:01:41,520 --> 00:01:44,600
visibility and search results 
has collapsed. 

29
00:01:44,840 --> 00:01:49,640
The Mirrors presence on Google 
is down 80% since 2019. 

30
00:01:49,880 --> 00:01:52,080
The Mail has lost more than 
half. 

31
00:01:52,320 --> 00:01:55,760
Even the Financial Times, a 
specialist publication with a 

32
00:01:55,760 --> 00:02:00,920
loyal subscriber base, saw a 21%
drop in traffic this spring. 

33
00:02:01,360 --> 00:02:05,160
The report attributes this 
decline primarily to changes in 

34
00:02:05,160 --> 00:02:10,479
Google systems, in particular 
the roll out of AI Overviews and

35
00:02:10,639 --> 00:02:13,760
AI mode. 
Rather than publisher strategy. 

36
00:02:14,560 --> 00:02:19,000
This shift is structural. 
AI tools are intercepting the 

37
00:02:19,000 --> 00:02:22,240
audience before they reach the 
source of the information that 

38
00:02:22,240 --> 00:02:25,720
they're seeking. 
Google's AI Overviews is doing 

39
00:02:25,720 --> 00:02:29,680
this even for users who weren't 
looking for an AI answer. 

40
00:02:30,000 --> 00:02:33,560
Publishers can't opt out from 
this without disappearing from 

41
00:02:33,560 --> 00:02:36,880
search entirely, meaning that 
the economic model that 

42
00:02:36,880 --> 00:02:41,480
sustained journalism, clicks, 
subscriptions and advertising is

43
00:02:41,480 --> 00:02:45,640
being rapidly eroded by systems 
that extract value from news 

44
00:02:45,640 --> 00:02:50,440
providers without returning any.
This chart from The Economist 

45
00:02:50,440 --> 00:02:54,960
shows that news and media are by
no means the most affected by 

46
00:02:54,960 --> 00:02:58,720
this change. 
Science and education sites and 

47
00:02:58,720 --> 00:03:02,720
reference sites like Wikipedia 
are doing even worse, with 

48
00:03:02,720 --> 00:03:05,680
health information being the 
most impacted. 

49
00:03:06,200 --> 00:03:09,760
Search engines were, and still 
are to a certain extent, the 

50
00:03:09,760 --> 00:03:13,840
gateway to the Internet, but 
that gate is rapidly closing. 

51
00:03:14,120 --> 00:03:17,480
The shift in user behaviour is 
already measurable. 

52
00:03:17,720 --> 00:03:22,080
In April, Apple reported the 
first ever decline in Safari 

53
00:03:22,080 --> 00:03:26,120
search volume, a change they 
attributed to users turning to 

54
00:03:26,160 --> 00:03:29,400
AI chat tools instead of 
traditional search engines. 

55
00:03:29,760 --> 00:03:35,120
A Tech Radar survey conducted in
December found that 27% of US 

56
00:03:35,120 --> 00:03:41,720
users and 13% of UK users now 
use AI tools like ChatGPT 

57
00:03:41,880 --> 00:03:44,920
instead of search engines to 
begin their information 

58
00:03:44,920 --> 00:03:48,120
gathering. 
The reason cited includes speed,

59
00:03:48,200 --> 00:03:53,840
specificity, and ease of use. 
Google referrals to news sites 

60
00:03:53,840 --> 00:04:00,280
have dropped from around 65% in 
2019 to 30% today, but this 

61
00:04:00,280 --> 00:04:05,240
decline began long before AI 
overviews a few years ago. 

62
00:04:05,280 --> 00:04:08,480
People used to say that nothing 
ever disappears from the 

63
00:04:08,480 --> 00:04:12,320
Internet, usually as a warning 
against posting anything that 

64
00:04:12,320 --> 00:04:15,920
you might later be ashamed of. 
But we've since learned that 

65
00:04:15,920 --> 00:04:20,720
things can still disappear, even
if they're never erased, because

66
00:04:20,720 --> 00:04:23,160
they just become impossible to 
find. 

67
00:04:23,680 --> 00:04:27,400
Over the last few years, Google 
became less and less efficient 

68
00:04:27,440 --> 00:04:30,920
at finding the information you 
were searching for, both because

69
00:04:30,920 --> 00:04:35,200
the web became saturated with 
SEO optimized content and 

70
00:04:35,200 --> 00:04:38,560
because Google started returning
sponsored search results. 

71
00:04:38,800 --> 00:04:42,120
Users found that they had to 
sift through clutter to find 

72
00:04:42,120 --> 00:04:45,640
credible sources. 
AI tools turned up on the scene 

73
00:04:45,840 --> 00:04:49,240
right as Internet search was 
faltering, and according to The 

74
00:04:49,240 --> 00:04:53,560
Atlantic, AI chat bots have 
already replaced search for more

75
00:04:53,560 --> 00:04:58,600
than 25% of Americans. 
Google's own AI features have 

76
00:04:58,600 --> 00:05:02,600
made this switch even more 
extreme, as they provide AI 

77
00:05:02,600 --> 00:05:05,840
answers to people who were 
planning on using traditional 

78
00:05:05,840 --> 00:05:08,480
search. 
The problem is that if things 

79
00:05:08,480 --> 00:05:12,200
continue in this direction, AI 
will eat itself. 

80
00:05:12,400 --> 00:05:16,160
If news organization and 
reporters can no longer earn a 

81
00:05:16,160 --> 00:05:20,120
living by doing the hard work of
researching an important story, 

82
00:05:20,360 --> 00:05:24,200
that work just won't be done, 
and chat bots will just have to 

83
00:05:24,200 --> 00:05:28,640
resort to making up answers, 
relying on press releases or on 

84
00:05:28,640 --> 00:05:32,200
propaganda posted deliberately 
to mislead readers. 

85
00:05:32,520 --> 00:05:37,040
Content producers are obviously 
working to find a new structure 

86
00:05:37,160 --> 00:05:40,800
where AI companies pay them for 
valuable information. 

87
00:05:41,080 --> 00:05:44,720
But if that doesn't work, the 
open web may evolve into 

88
00:05:44,720 --> 00:05:48,480
something very different. 
So let's look at how AI is 

89
00:05:48,480 --> 00:05:52,560
accelerating the collapse of the
webs information economy, what 

90
00:05:52,560 --> 00:05:55,920
it means for the future of 
journalism and online content, 

91
00:05:56,080 --> 00:05:59,520
and why the stakes are higher 
than most people realize. 

92
00:06:00,080 --> 00:06:04,320
The problem Google has with AI 
companies is not the AI 

93
00:06:04,320 --> 00:06:08,320
technology itself, but what that
technology does to their 

94
00:06:08,320 --> 00:06:11,760
existing business model. 
Google is not really in the 

95
00:06:11,760 --> 00:06:15,320
business of giving you the best 
answer to your search questions.

96
00:06:15,480 --> 00:06:18,560
It's in the business of giving 
you a good enough answer so that

97
00:06:18,560 --> 00:06:21,880
you stick with Google search 
while encouraging you to click 

98
00:06:21,880 --> 00:06:24,520
on an ad. 
If you ask what's the best 

99
00:06:24,520 --> 00:06:28,560
webcam, Google has to ensure 
that you'll see ad placements 

100
00:06:28,560 --> 00:06:31,680
paid for by its customers at the
top of the list. 

101
00:06:32,080 --> 00:06:36,080
Google is now pitching that 
their new AI driven approach to 

102
00:06:36,080 --> 00:06:40,080
search allows you to let Google 
do the Googling for you. 

103
00:06:40,760 --> 00:06:45,080
In theory, with the AI mode you 
can still see the links and use 

104
00:06:45,080 --> 00:06:48,960
your own judgement, but not many
users are clicking on the links 

105
00:06:48,960 --> 00:06:52,680
anymore and because they're 
advertiser driven business model

106
00:06:52,840 --> 00:06:56,760
needs to continue working. 
The answers you get might be 

107
00:06:56,760 --> 00:06:59,680
colored by the need to sell 
certain products. 

108
00:07:00,160 --> 00:07:03,960
If you ask for health advice, as
it appears that many users are, 

109
00:07:04,160 --> 00:07:08,360
will you be served information 
from reputable sources or from 

110
00:07:08,360 --> 00:07:11,680
questionable supplement sellers 
who are paying for clicks? 

111
00:07:12,160 --> 00:07:15,440
I'm not sure how much you can 
trust this AI stuff. 

112
00:07:15,760 --> 00:07:19,200
Recently I've been getting all 
these adverts on YouTube from a 

113
00:07:19,200 --> 00:07:23,760
guy who says he's 68. 
He's also AI and he's been doing

114
00:07:23,760 --> 00:07:28,440
Tai chi and is in great shape. 
And I figure that I'd like to be

115
00:07:28,440 --> 00:07:32,080
AI and in great shape too. 
But when I click on the link, it

116
00:07:32,080 --> 00:07:34,880
tries to get me to do those 
exercises that you see old 

117
00:07:34,880 --> 00:07:38,520
people doing at the park. 
And I don't want to do that as I

118
00:07:38,520 --> 00:07:41,920
worry that they'll scramble my 
brain like they clearly did for 

119
00:07:41,920 --> 00:07:44,280
this guy. 
I mean, look at what he wrote on

120
00:07:44,280 --> 00:07:48,880
the white board. 
I started Tai chi chi 50. 

121
00:07:49,080 --> 00:07:54,240
Now I'm now I'm 68. 
Doctors ask me for advice. 

122
00:07:54,760 --> 00:07:58,640
They're probably not ACA asking 
him for advice on how to write. 

123
00:07:59,200 --> 00:08:03,520
Anyhow, Good journalism is 
expensive to produce. 

124
00:08:03,760 --> 00:08:07,560
It requires reporters on the 
ground, editors with judgement, 

125
00:08:07,680 --> 00:08:11,320
and teams of fact checkers. 
Traditionally, this work was 

126
00:08:11,320 --> 00:08:14,840
funded through either 
subscriptions or advertising. 

127
00:08:15,120 --> 00:08:19,200
If users stop visiting websites,
the incentive to produce 

128
00:08:19,200 --> 00:08:22,040
original content disappears 
entirely. 

129
00:08:22,320 --> 00:08:25,960
Why would anyone write a news 
report just for it to be scraped

130
00:08:25,960 --> 00:08:30,200
by a dozen AI bots, scrambled up
with other news, and delivered 

131
00:08:30,200 --> 00:08:33,960
to an audience who have no 
interest in how it was created? 

132
00:08:34,559 --> 00:08:38,360
The result would be fewer 
investigations, fewer foreign 

133
00:08:38,360 --> 00:08:42,159
correspondents, fewer deep dives
into complex issues. 

134
00:08:42,360 --> 00:08:46,240
The web turns into a hall of 
mirrors, reflecting summaries of

135
00:08:46,240 --> 00:08:51,120
summaries, AI hallucinations and
press releases with no original 

136
00:08:51,120 --> 00:08:54,200
source insight. 
This effects publishers, 

137
00:08:54,280 --> 00:08:57,480
advertising advertisers, and AI 
companies alike. 

138
00:08:58,000 --> 00:09:02,600
AI tools depend on fresh, high 
quality content so that they can

139
00:09:02,600 --> 00:09:05,560
function. 
They need journalists to gather 

140
00:09:05,560 --> 00:09:09,680
facts, analysts to interpret 
data, and reviewers to test 

141
00:09:09,680 --> 00:09:12,440
products. 
If those people can't earn a 

142
00:09:12,440 --> 00:09:15,200
living, the source material 
dries up. 

143
00:09:15,440 --> 00:09:19,160
The bots will still generate 
answers, but those answers will 

144
00:09:19,160 --> 00:09:22,560
be stitched together from 
outdated articles, press 

145
00:09:22,560 --> 00:09:27,480
releases, and propaganda. 
AI ends up eating itself. 

146
00:09:28,080 --> 00:09:31,720
Publishers are trying to fight 
back through licensing deals, 

147
00:09:31,720 --> 00:09:36,000
lawsuits, and technical blocks, 
but the economics are brutal. 

148
00:09:36,280 --> 00:09:40,400
If the ad money flows elsewhere 
and the traffic never arrives, 

149
00:09:40,560 --> 00:09:45,320
the open web may not survive. 
In its current form, AI won't 

150
00:09:45,320 --> 00:09:48,960
just disrupt journalism, it'll 
hollow out the information 

151
00:09:49,040 --> 00:09:51,320
ecosystem that it was trained 
on. 

152
00:09:52,080 --> 00:09:55,120
Journalism isn't the only thing 
under threat. 

153
00:09:55,360 --> 00:09:59,960
The entire ecosystem of online 
reviews, once a vital part of 

154
00:09:59,960 --> 00:10:03,280
consumer decision making, is 
breaking down, too. 

155
00:10:03,760 --> 00:10:07,760
Before the Internet, reviews 
were published in magazines, but

156
00:10:07,760 --> 00:10:09,840
those weren't always 
trustworthy. 

157
00:10:10,160 --> 00:10:14,760
Big advertisers could influence 
editorial decisions, and readers

158
00:10:14,760 --> 00:10:17,280
had limited ways to verify 
claims. 

159
00:10:17,800 --> 00:10:21,160
The Internet offered an 
alternative independent 

160
00:10:21,160 --> 00:10:24,520
reviewers who built trust 
through transparency and 

161
00:10:24,520 --> 00:10:27,840
consistency. 
Tech reviewers, car reviewers, 

162
00:10:27,840 --> 00:10:32,320
and niche experts earned loyal 
audiences by being honest and 

163
00:10:32,320 --> 00:10:35,760
monetize that trust through 
views and subscriptions. 

164
00:10:36,120 --> 00:10:39,560
Online reviews were one of the 
best things about the early 

165
00:10:39,560 --> 00:10:42,680
Internet. 
Platforms like Amazon offered a 

166
00:10:42,680 --> 00:10:47,080
new kind of transparency, real 
feedback from real users. 

167
00:10:47,280 --> 00:10:49,480
For shoppers, this was a 
breakthrough. 

168
00:10:49,640 --> 00:10:52,480
You didn't have to trust an 
advert or a sales pitch. 

169
00:10:52,600 --> 00:10:55,160
You could read what other 
consumers thought. 

170
00:10:55,520 --> 00:10:59,960
That trust didn't last. 
Sellers began gaming the system.

171
00:11:00,200 --> 00:11:03,600
Some offered free gifts or 
discounts in exchange for five 

172
00:11:03,600 --> 00:11:06,800
star reviews. 
Others paid bot farms to flood 

173
00:11:06,800 --> 00:11:10,800
platforms with fake praise or to
sabotage competitors. 

174
00:11:11,160 --> 00:11:14,360
Studies show that in categories 
like electronics and 

175
00:11:14,360 --> 00:11:17,760
supplements, the majority of 
reviews may be fake. 

176
00:11:18,640 --> 00:11:22,280
Now AI tools are summarizing 
reviews, but they don't 

177
00:11:22,280 --> 00:11:25,480
distinguish between honest 
feedback and manipulated 

178
00:11:25,480 --> 00:11:28,240
content. 
If trusted reviewers lose 

179
00:11:28,240 --> 00:11:31,640
traffic because their work is 
scraped and repackaged without 

180
00:11:31,640 --> 00:11:35,760
attribution or compensation, 
their incentive to produce high 

181
00:11:35,760 --> 00:11:38,720
quality, unbiased reviews 
disappears. 

182
00:11:39,200 --> 00:11:42,440
The review ecosystem isn't just 
about choosing the right 

183
00:11:42,440 --> 00:11:44,520
headphones or booking a good 
hotel. 

184
00:11:44,760 --> 00:11:47,080
It's part of how we evaluate 
truth. 

185
00:11:47,360 --> 00:11:51,920
If AI tools can't be trusted on 
this front because their source 

186
00:11:51,920 --> 00:11:55,520
material is compromised, they 
lose a key pillar of 

187
00:11:55,520 --> 00:11:59,440
credibility. 
As AI tools scrape and summarize

188
00:11:59,440 --> 00:12:03,280
content, often without 
permission or compensation, some

189
00:12:03,280 --> 00:12:07,600
creators are pushing back. 
The musician Ben Jordan, who was

190
00:12:07,600 --> 00:12:10,960
one of the most interesting 
channels on YouTube, developed a

191
00:12:10,960 --> 00:12:16,040
tool called Poisonify to protect
artists from unauthorized AI 

192
00:12:16,040 --> 00:12:18,760
training. 
It's designed to prevent AI 

193
00:12:18,760 --> 00:12:22,000
companies from using music 
without the creator's 

194
00:12:22,000 --> 00:12:25,320
permission. 
His video on this project, along

195
00:12:25,320 --> 00:12:28,760
with his most recent one on AI 
companies that are logging and 

196
00:12:28,760 --> 00:12:32,520
sharing your vehicle's location 
without your consent, are well 

197
00:12:32,520 --> 00:12:34,920
worth watching. 
I'll include links in the 

198
00:12:34,920 --> 00:12:38,800
description. 
Poisonify works by adding 

199
00:12:38,800 --> 00:12:43,240
imperceptible adversarial noise 
to a music track so that the 

200
00:12:43,240 --> 00:12:47,600
audio sounds unchanged to human 
listeners, but sounds entirely 

201
00:12:47,600 --> 00:12:52,120
different to AI scrapers. 
His software not only prevents a

202
00:12:52,280 --> 00:12:56,080
IS from replicating an artist to
music without permission, but it

203
00:12:56,080 --> 00:13:00,160
may even punish them for taking 
the music without permission by 

204
00:13:00,160 --> 00:13:04,360
corrupting their training data. 
If scraped, the Poison tracks 

205
00:13:04,560 --> 00:13:08,560
can possibly degrade an AI 
model's performance, making it 

206
00:13:08,560 --> 00:13:12,360
harder for AI companies to 
profit from stolen content. 

207
00:13:12,880 --> 00:13:15,760
Publishers are deploying 
countermeasures, too. 

208
00:13:16,080 --> 00:13:19,240
Cloudfare and other 
infrastructure providers now 

209
00:13:19,240 --> 00:13:21,840
offer tools to block AI 
crawlers. 

210
00:13:22,040 --> 00:13:25,560
Some media groups are 
negotiating licensing deals with

211
00:13:25,680 --> 00:13:30,400
AI firms, and others are suing. 
The New York Times, for example,

212
00:13:30,480 --> 00:13:35,080
has filed lawsuits against Open 
AI and Microsoft, arguing that 

213
00:13:35,080 --> 00:13:38,720
their models were trained on 
copyrighted journalism without 

214
00:13:38,720 --> 00:13:42,360
permission. 
The most effective strategy may 

215
00:13:42,360 --> 00:13:45,440
be branding. 
In a world where AI can mimic 

216
00:13:45,440 --> 00:13:49,760
tone and summarize content, 
personality becomes a mode. 

217
00:13:50,080 --> 00:13:54,200
Publishers are promoting 
individual voices, columnists, 

218
00:13:54,200 --> 00:13:58,280
Youtubers, sub stack writers as 
a way to build loyalty and 

219
00:13:58,280 --> 00:14:01,400
retain traffic. 
The Wall Street Journal recently

220
00:14:01,400 --> 00:14:05,880
advertised for a talent coach to
help journalists build personal 

221
00:14:05,880 --> 00:14:08,880
brands. 
This is based on the idea that 

222
00:14:08,880 --> 00:14:11,800
readers follow people rather 
than platforms. 

223
00:14:12,280 --> 00:14:16,240
This shift mirrors the rise of 
the creator economy, where 

224
00:14:16,240 --> 00:14:19,720
independent journalists, 
analysts and reviewers build 

225
00:14:19,720 --> 00:14:23,600
direct relationships with their 
audiences through newsletters, 

226
00:14:23,600 --> 00:14:28,560
podcasts and paid subscriptions.
But even these models depend on 

227
00:14:28,560 --> 00:14:31,560
visibility. 
If AI tools intercept the 

228
00:14:31,600 --> 00:14:35,360
audience before they reach the 
creator, the economics still 

229
00:14:35,360 --> 00:14:39,480
collapse. 
The irony is that this strategy,

230
00:14:39,480 --> 00:14:44,120
building personality as a mode, 
only works if the personality is

231
00:14:44,120 --> 00:14:47,280
real. 
AI generated influencers are 

232
00:14:47,320 --> 00:14:51,080
already gaining traction, 
synthetic voices, synthetic 

233
00:14:51,080 --> 00:14:55,600
faces, and synthetic opinions. 
If the last defensible asset is 

234
00:14:55,600 --> 00:15:00,200
authenticity, then the next arms
race is over what it means to be

235
00:15:00,200 --> 00:15:03,320
real. 
Some startups are trying to 

236
00:15:03,320 --> 00:15:06,520
build new models around AI 
access. 

237
00:15:06,760 --> 00:15:11,400
Toll bid, which describes itself
as a paywall for bots, allows 

238
00:15:11,400 --> 00:15:15,640
content sites to charge AI 
crawlers variable rates, 

239
00:15:15,840 --> 00:15:18,560
charging more for news stories 
than old ones. 

240
00:15:18,840 --> 00:15:22,920
The firm argues that charging 
for access incentivizes 

241
00:15:22,920 --> 00:15:27,160
uniqueness, unlike traditional 
search, which rewards generic 

242
00:15:27,160 --> 00:15:30,480
content. 
Another approach comes from Pro 

243
00:15:30,480 --> 00:15:34,760
Rata, which proposes that ad 
revenue from AI generated 

244
00:15:34,760 --> 00:15:38,720
answers should be redistributed 
to the sites whose content 

245
00:15:38,720 --> 00:15:42,960
contributed to those answers. 
It's engine already shares 

246
00:15:42,960 --> 00:15:46,680
revenue with over 500 partner 
publications. 

247
00:15:47,320 --> 00:15:51,000
A growing concern with users 
turning to AI chat bots for 

248
00:15:51,000 --> 00:15:54,160
information is that the 
companies behind them have been 

249
00:15:54,160 --> 00:15:56,920
aggressively overselling their 
capabilities. 

250
00:15:57,360 --> 00:16:00,760
These systems are pitched as 
super intelligent beings, 

251
00:16:00,880 --> 00:16:05,280
smarter than any expert, 
unbiased by design and capable 

252
00:16:05,280 --> 00:16:09,120
of answering anything. 
That framing gives users a false

253
00:16:09,120 --> 00:16:11,240
sense of confidence in the 
output. 

254
00:16:11,600 --> 00:16:16,560
Elon Musk has claimed that his 
chatbot Grog is more intelligent

255
00:16:16,560 --> 00:16:20,600
than PhD holders in every 
discipline, that it can discover

256
00:16:20,600 --> 00:16:25,080
new physics and outperform 
humans on humanity's last exam, 

257
00:16:25,360 --> 00:16:29,000
a benchmark designed to test 
reasoning across 100 

258
00:16:29,000 --> 00:16:31,760
disciplines. 
Of course, he has also been 

259
00:16:31,760 --> 00:16:35,520
claiming that his cars can drive
themselves for nearly a decade 

260
00:16:35,520 --> 00:16:39,240
now, so we should probably take 
those claims with a pinch of 

261
00:16:39,240 --> 00:16:42,720
salt. 
Angela Collier made an excellent

262
00:16:42,720 --> 00:16:47,200
and funny video a few weeks ago 
called Vibe Physics where she 

263
00:16:47,200 --> 00:16:50,640
explains how tech entrepreneurs 
like the guys on the All In 

264
00:16:50,640 --> 00:16:55,720
podcast mistake chatbot fluency 
for intelligence, especially 

265
00:16:55,720 --> 00:16:58,960
when the bot starts talking 
about subjects outside their 

266
00:16:58,960 --> 00:17:01,920
expertise, like her field 
Physics. 

267
00:17:02,320 --> 00:17:07,760
I'll go down this thread with 
GBT or Grok and I'll start to 

268
00:17:07,760 --> 00:17:11,240
get to the edge of what's known 
in quantum physics. 

269
00:17:11,319 --> 00:17:14,119
They're quick to spot errors 
when the chatbot discusses 

270
00:17:14,119 --> 00:17:17,640
something that they know, but 
once it shifts to unfamiliar 

271
00:17:17,640 --> 00:17:21,240
territory, they're suddenly 
amazed by its brilliance. 

272
00:17:21,480 --> 00:17:25,119
The illusion of super 
intelligence kicks in precisely 

273
00:17:25,119 --> 00:17:28,440
when the user lacks the tools to
evaluate the answer. 

274
00:17:28,720 --> 00:17:32,280
Meta, which really ought to go 
back to calling itself Facebook,

275
00:17:32,280 --> 00:17:36,800
soon says that it's AI powered. 
Nerd Glasses will give users a 

276
00:17:36,920 --> 00:17:40,840
cognitive edge, and who knows, 
they might even be right. 

277
00:17:41,080 --> 00:17:44,400
Maybe trying to ignore the 
adverts being served to you all 

278
00:17:44,400 --> 00:17:47,360
day long while you're trying to 
focus on more important things 

279
00:17:47,600 --> 00:17:49,760
will boost your cognitive 
abilities. 

280
00:17:50,440 --> 00:17:54,640
When users stop reading news 
outlets where they understand 

281
00:17:54,640 --> 00:17:58,840
the editorial standards, 
reputations and biases of both 

282
00:17:58,840 --> 00:18:02,920
the journalist and the publisher
and start asking chat bots for 

283
00:18:02,920 --> 00:18:06,680
answers, assuming neutrality 
where bias may exist. 

284
00:18:06,920 --> 00:18:10,560
They're not just changing 
platforms, they're changing who 

285
00:18:10,560 --> 00:18:15,400
they trust to explain the world.
And that trust may be misplaced.

286
00:18:16,200 --> 00:18:20,160
AI systems are not neutral. 
They reflect the data they're 

287
00:18:20,160 --> 00:18:24,160
trained on, the assumptions of 
their coders, and the incentives

288
00:18:24,240 --> 00:18:26,040
of the companies that deploy 
them. 

289
00:18:26,480 --> 00:18:30,480
Bias can creep in through 
training data, algorithmic 

290
00:18:30,480 --> 00:18:35,520
design, and moderation choices. 
Sometimes it's accidental, but 

291
00:18:35,520 --> 00:18:39,560
sometimes it's deliberate. 
Grok is a case study in what 

292
00:18:39,560 --> 00:18:44,280
happens when those filters fail.
After a July update designed to 

293
00:18:44,280 --> 00:18:48,400
make it sound more raw, Grok 
began referring to itself as 

294
00:18:48,400 --> 00:18:52,400
Mecca Hitler. 
Turkey banned the chat bot not 

295
00:18:52,400 --> 00:18:55,640
because of the Nazi stuff, but 
because it also insulted 

296
00:18:55,640 --> 00:18:59,480
President Erdogan. 
Other bots have had breakdowns, 

297
00:18:59,480 --> 00:19:03,240
too, fabricating sources, 
hallucinating facts, and 

298
00:19:03,240 --> 00:19:07,480
misattributing quotes. 
The myth of super intelligence 

299
00:19:07,680 --> 00:19:12,040
encourages users to place their 
trust in machines that sound 

300
00:19:12,040 --> 00:19:15,600
authoritative, even when they 
lack transparency or 

301
00:19:15,600 --> 00:19:18,680
accountability. 
As chat bots become the default 

302
00:19:18,680 --> 00:19:23,280
source of information, users may
begin to rely on them more than 

303
00:19:23,280 --> 00:19:28,120
qualified experts, especially in
areas where expertise is hard 

304
00:19:28,120 --> 00:19:31,960
earned and context matters. 
This shift is happening at a 

305
00:19:31,960 --> 00:19:36,000
time when public confidence in 
expert communities is already 

306
00:19:36,000 --> 00:19:39,520
under pressure. 
According to Pew Research, trust

307
00:19:39,520 --> 00:19:43,160
in scientists has declined 
steadily over the last five 

308
00:19:43,160 --> 00:19:47,160
years, with fewer than half of 
Americans now expressing strong 

309
00:19:47,160 --> 00:19:49,600
confidence in scientific 
leaders. 

310
00:19:50,080 --> 00:19:54,320
As AI systems take on the role 
of explainer in chief, the 

311
00:19:54,320 --> 00:19:58,600
institutions that once anchored 
public understanding risk being 

312
00:19:58,600 --> 00:20:01,560
sidelined. 
It's quite worried that in that 

313
00:20:01,560 --> 00:20:05,840
economist chart I showed earlier
that health information is most 

314
00:20:05,840 --> 00:20:09,720
heavily impacted by the move to 
AI tools for information. 

315
00:20:10,200 --> 00:20:13,720
The open web was built on a 
simple exchange. 

316
00:20:13,960 --> 00:20:18,440
Publishers create content, users
visit websites, and their 

317
00:20:18,440 --> 00:20:22,280
attention is then monetized, 
funding the next round of 

318
00:20:22,280 --> 00:20:25,760
reporting. 
AI tools are already breaking 

319
00:20:25,760 --> 00:20:29,240
that loop. 
If no one pays for the news, the

320
00:20:29,240 --> 00:20:32,880
news stops being reported. 
Investigations don't happen. 

321
00:20:33,080 --> 00:20:37,200
Deep dives into complex issues 
are replaced by AI summaries of 

322
00:20:37,200 --> 00:20:39,760
press releases and 
hallucinations. 

323
00:20:40,240 --> 00:20:44,120
Well, some argued that citizen 
journalism can fill the gap. 

324
00:20:44,320 --> 00:20:48,600
Professional journalism can't be
replaced by individuals posting 

325
00:20:48,600 --> 00:20:51,880
on social media. 
Eyewitness accounts and grass 

326
00:20:51,880 --> 00:20:55,920
roots reporting can be powerful,
but they don't come with 

327
00:20:55,920 --> 00:20:59,880
editors, legal teams, or the 
resources to spend months 

328
00:20:59,880 --> 00:21:02,720
verifying leads and protecting 
sources. 

329
00:21:03,240 --> 00:21:08,000
The BB CS investigation into the
Wagner Group in Libya, NB CS 

330
00:21:08,000 --> 00:21:12,200
expose on forced adoptions in 
Christian boarding homes, and 

331
00:21:12,200 --> 00:21:16,960
Pro Publica's reporting on US 
Supreme Court ethic scandals all

332
00:21:16,960 --> 00:21:21,600
required institutional backing 
and sustained access to data and

333
00:21:21,600 --> 00:21:24,200
sources. 
These aren't stories that could 

334
00:21:24,200 --> 00:21:27,240
have been broken by someone with
a smartphone and a Twitter 

335
00:21:27,240 --> 00:21:30,200
account. 
Journalism isn't just about 

336
00:21:30,200 --> 00:21:32,920
being there. 
It's about knowing what to do 

337
00:21:33,000 --> 00:21:36,400
with what you find and having 
the infrastructure to do it 

338
00:21:36,400 --> 00:21:40,520
responsibly. 
The disruption unfolding reaches

339
00:21:40,520 --> 00:21:44,360
beyond journalism and could 
reshape the entire information 

340
00:21:44,440 --> 00:21:49,480
ecosystem if the source material
dries up due to lack of funding.

341
00:21:49,680 --> 00:21:53,440
The bots will still generate 
answers, but those answers will 

342
00:21:53,440 --> 00:21:57,160
be stitched together from 
outdated articles, corporate 

343
00:21:57,160 --> 00:22:01,880
peor, and propaganda. 
The web could begin to resemble 

344
00:22:01,920 --> 00:22:06,800
a television camera pointed at a
television recycling recycled 

345
00:22:06,800 --> 00:22:09,560
content. 
I've talked in the past on this 

346
00:22:09,560 --> 00:22:13,480
channel about the strategy of 
blitz scaling, where well funded

347
00:22:13,480 --> 00:22:17,480
tech disruptors prioritize 
growth over profitability. 

348
00:22:17,800 --> 00:22:22,680
The idea is that rapid expansion
attracts high market valuations,

349
00:22:22,840 --> 00:22:26,920
allowing venture capitalists to 
flip unprofitable or barely 

350
00:22:26,920 --> 00:22:30,360
profitable companies to the 
public before before they ever 

351
00:22:30,360 --> 00:22:32,680
find a sustainable source of 
profit. 

352
00:22:33,120 --> 00:22:37,400
Blitzschaled firms often lose 
money, deliberately undercutting

353
00:22:37,400 --> 00:22:41,000
incumbents with the goal of 
eliminating competition and 

354
00:22:41,000 --> 00:22:44,520
raising prices later. 
Uber didn't need to be 

355
00:22:44,520 --> 00:22:47,160
profitable to reshape urban 
transport. 

356
00:22:47,320 --> 00:22:50,800
It just needed to be fast, cheap
and relentless. 

357
00:22:51,400 --> 00:22:54,160
AI is now following the same 
playbook. 

358
00:22:54,480 --> 00:22:58,040
Vast sums are being spent to 
build tools that have yet to 

359
00:22:58,040 --> 00:23:02,440
find a viable business model. 
But even without a path to 

360
00:23:02,440 --> 00:23:06,560
profitability, they could still 
destroy the economic scaffolding

361
00:23:06,720 --> 00:23:10,360
that supports journalism, 
education, and public knowledge.

362
00:23:11,200 --> 00:23:14,080
Technology eventually comes for 
everything. 

363
00:23:14,240 --> 00:23:17,120
But when it comes for the 
institutions that help us 

364
00:23:17,160 --> 00:23:20,720
understand the world, the stakes
are higher than most people 

365
00:23:20,720 --> 00:23:24,160
realize. 
Now it has always been as long 

366
00:23:24,160 --> 00:23:26,840
as man has been around, 
everything that can be replaced 

367
00:23:26,840 --> 00:23:28,480
will be replaced. 
But there are things that are 

368
00:23:28,480 --> 00:23:31,760
irreplaceable. 
Business models evolve. 

369
00:23:32,000 --> 00:23:34,400
When the Internet first 
appeared, it looked like 

370
00:23:34,400 --> 00:23:37,680
newspapers were doomed. 
But the industry adapted, 

371
00:23:37,680 --> 00:23:40,240
reinventing itself for the 
digital age. 

372
00:23:40,640 --> 00:23:44,080
Napster made it seem like music 
would be free forever, 

373
00:23:44,200 --> 00:23:47,880
threatening that industry. 
Yet musicians found new ways to 

374
00:23:47,920 --> 00:23:51,800
earn through touring, streaming 
and direct fan engagement. 

375
00:23:52,320 --> 00:23:56,080
The information economy is 
facing a similar moment, but 

376
00:23:56,080 --> 00:23:59,920
having access to high quality 
information is essential to the 

377
00:23:59,920 --> 00:24:03,320
functioning of the modern world.
And while today's disruption 

378
00:24:03,320 --> 00:24:07,120
feels existential, it's more 
likely the journalism and 

379
00:24:07,120 --> 00:24:10,800
knowledge institutions will 
adapt to the new environment 

380
00:24:11,000 --> 00:24:15,200
rather than vanishing from it. 
The business models may change, 

381
00:24:15,360 --> 00:24:19,840
the platforms may shift, but the
demand for truth, context, and 

382
00:24:19,840 --> 00:24:24,160
accountability isn't going away.
Thanks for tuning into this 

383
00:24:24,160 --> 00:24:27,600
week's podcast, with particular 
thanks to my supporters on 

384
00:24:27,600 --> 00:24:30,840
Patreon whose contributions make
this work. 

385
00:24:31,080 --> 00:24:34,400
If you want to support the 
podcast, I'll leave a link in 

386
00:24:34,400 --> 00:24:36,880
the show notes. 
Have a great week and talk to 

387
00:24:36,880 --> 00:24:38,200
you again soon. 
Bye.