1
00:00:01,040 --> 00:00:04,680
From Indias largest newsroom, 
I'm Arun George and this is the 

2
00:00:04,680 --> 00:00:16,520
Times of India podcast. 
We're just two phases into the 

3
00:00:16,520 --> 00:00:19,760
long national elections in India
and two Bollywood actors have 

4
00:00:19,760 --> 00:00:22,760
already filed complaints over 
fake videos featuring them being

5
00:00:22,760 --> 00:00:25,720
shared on social media. 
Aamir Khan and Ranveer Singh 

6
00:00:25,760 --> 00:00:28,360
file Police Complaints over 
videos that was circulated on 

7
00:00:28,360 --> 00:00:30,560
social media. 
They claim they were campaigning

8
00:00:30,560 --> 00:00:40,320
for a political party. 
Celebrate Kare Hamari. 

9
00:00:40,320 --> 00:00:47,000
Dukhi Hui Jeevan. 
We done this episode at the 

10
00:00:47,000 --> 00:00:49,600
beginning of this year about 
deep fake videos, and we're 

11
00:00:49,600 --> 00:00:51,080
bringing back that episode 
today. 

12
00:00:51,440 --> 00:00:54,360
We've spoken with Karen Ribello,
who's a deputy editor with fact 

13
00:00:54,360 --> 00:00:57,840
checking organization Boom Live,
about why it's so hard to tackle

14
00:00:58,000 --> 00:01:01,160
AI generated fake videos and 
other forms of disinformation on

15
00:01:01,160 --> 00:01:03,680
the Internet. 
We'd ask Karen what kind of 

16
00:01:03,680 --> 00:01:06,560
misinformation is most commonly 
found in India. 

17
00:01:08,480 --> 00:01:12,720
India is a country where you see
a lot of misinformation purely 

18
00:01:12,720 --> 00:01:18,320
in terms of the volumes of 
misinformation, it is a lot like

19
00:01:18,320 --> 00:01:22,440
on an average we do about four 
to six fact checks every day. 

20
00:01:22,440 --> 00:01:26,120
And right now I think in the 
country there may be 15 to 20 

21
00:01:26,120 --> 00:01:29,520
fact checkers as well and they 
do similar volumes as well. 

22
00:01:29,520 --> 00:01:34,080
So purely in terms of like, the 
amount of misinformation is 

23
00:01:34,080 --> 00:01:36,600
enormous. 
What we've generally seen over 

24
00:01:36,600 --> 00:01:40,880
the past five to six years, we 
had a big problem and I would 

25
00:01:40,880 --> 00:01:43,720
say we still have a big problem 
with something called as cheap 

26
00:01:43,720 --> 00:01:46,520
fakes. 
So cheap fakes are something 

27
00:01:46,520 --> 00:01:50,640
where you know, it's 
manipulation which doesn't 

28
00:01:50,640 --> 00:01:57,520
involve very high technology As 
such I crop a video or I speed 

29
00:01:57,520 --> 00:02:02,120
up the playback speed of a video
or I, you know, morph a photo 

30
00:02:02,120 --> 00:02:04,320
using Photoshop or something 
like that. 

31
00:02:05,280 --> 00:02:07,760
Those sort of things would be 
called as cheap fakes. 

32
00:02:07,760 --> 00:02:11,960
Now that doesn't involve AI. 
What we have seen over the past 

33
00:02:11,960 --> 00:02:16,840
one year I'd say is that we are 
seeing an increasing amount of 

34
00:02:17,240 --> 00:02:21,880
AI based misinformation and this
trend sort of started I think 

35
00:02:21,880 --> 00:02:25,520
around mid last year. 
So in May last year, you had 

36
00:02:25,520 --> 00:02:29,440
that image of wrestlers 
protesting and they were 

37
00:02:29,440 --> 00:02:32,680
detained inside a police van 
during a protest. 

38
00:02:33,200 --> 00:02:36,720
And there was an image which 
went viral, which showed them 

39
00:02:37,280 --> 00:02:40,160
smiling while they were sitting 
inside that van. 

40
00:02:40,160 --> 00:02:44,200
And the narrative around that 
photo was, look, these guys are 

41
00:02:44,200 --> 00:02:47,360
doing it just for a few laughs 
or that they're not serious 

42
00:02:47,360 --> 00:02:50,920
about this protest. 
We can see that here through 

43
00:02:51,120 --> 00:02:55,360
social media posts from netizens
like this one, which reads that 

44
00:02:55,360 --> 00:02:58,960
the photo is part of a a drama, 
part of a toolkit in order to 

45
00:02:58,960 --> 00:03:02,920
break our country. 
And of course later on we found 

46
00:03:02,920 --> 00:03:08,040
out it was edited or doctored 
with an app called Face App, 

47
00:03:08,040 --> 00:03:12,440
which is an AI based editing app
that happened in in May last 

48
00:03:12,440 --> 00:03:15,440
year. 
And since then there has been an

49
00:03:15,440 --> 00:03:22,920
increase in the sort of AI based
misinformation, especially that 

50
00:03:22,920 --> 00:03:26,680
has been catered to a local 
audience in India. 

51
00:03:27,120 --> 00:03:31,160
The deepfakes that we used to 
see were mostly Joe Biden or 

52
00:03:31,400 --> 00:03:36,000
Zelensky, you know, something 
related to the Russia, Ukraine 

53
00:03:36,000 --> 00:03:38,440
war. 
What we started seeing last year

54
00:03:38,440 --> 00:03:42,880
is like deepfakes in the Indian 
context and this sort of 

55
00:03:42,880 --> 00:03:46,680
increase. 
We saw a lot of deepfake 

56
00:03:47,200 --> 00:03:50,800
pornographic videos involving 
Indian actresses. 

57
00:03:50,800 --> 00:03:55,120
There were always pictures that 
were circulated on platforms 

58
00:03:55,120 --> 00:03:58,440
like X. 
And then of course, the Rashmika

59
00:03:58,440 --> 00:04:03,160
Mandana incident blew up. 
I did a story this month that 

60
00:04:03,160 --> 00:04:05,680
during the Madhya Pradesh 
elections. 

61
00:04:05,680 --> 00:04:08,840
So the Madhya Pradesh elections 
happened in October, November. 

62
00:04:09,800 --> 00:04:14,760
We found that a lot of videos 
were doctored with AI voice 

63
00:04:14,760 --> 00:04:18,480
cloning. 
So you had people make voice 

64
00:04:18,480 --> 00:04:24,280
clones of Shivraj Singh Chauhan 
and Kamal Nath and overlay them 

65
00:04:24,280 --> 00:04:29,280
onto real videos of these 
politicians where you know, you 

66
00:04:29,280 --> 00:04:33,000
have them saying completely 
fabricated statements, things 

67
00:04:33,000 --> 00:04:35,920
that they've never said. 
And those videos went viral. 

68
00:04:36,400 --> 00:04:40,320
Just to see how, how effective 
and how convincing it is, I went

69
00:04:40,320 --> 00:04:44,840
and created 2 audio clones of of
those two leaders and it was 

70
00:04:44,840 --> 00:04:47,440
terrifying because it was so 
real. 

71
00:04:47,440 --> 00:04:52,200
It sounded just like them. 
Me barbar Karahu Kamal Nath ko 

72
00:04:52,200 --> 00:04:56,040
roko Congress agai to Sabko 
Pandora surupe mahine or path 

73
00:04:56,080 --> 00:04:59,240
surupe. 
Me gas Denne lagegi, Chodo ugly 

74
00:04:59,240 --> 00:05:01,160
Barbie hamara jetna namkin ho 
jayega. 

75
00:05:02,280 --> 00:05:05,360
And you could never tell that a 
machine had created that or any 

76
00:05:05,360 --> 00:05:09,840
sort of AI. 
So in terms of the use of, you 

77
00:05:09,840 --> 00:05:13,280
know, AI to spread 
misinformation, India has 

78
00:05:13,280 --> 00:05:17,880
already seen it last year itself
in the context of an election. 

79
00:05:18,360 --> 00:05:23,280
So we already have precedence in
terms of the general election. 

80
00:05:23,280 --> 00:05:26,280
We can expect to see AI based 
misinformation. 

81
00:05:27,560 --> 00:05:31,640
What changes when it's 
Artificial intelligence based? 

82
00:05:32,720 --> 00:05:37,000
You know, these AI voice clones 
are also being used to scam a 

83
00:05:37,000 --> 00:05:41,400
lot of people and to create 
videos where they're 

84
00:05:41,400 --> 00:05:44,440
impersonating the voices of 
actors. 

85
00:05:44,440 --> 00:05:47,960
And they're either, you know, 
promoting some betting apps or 

86
00:05:47,960 --> 00:05:51,760
some fraudulent investment 
scheme or like diabetes 

87
00:05:51,760 --> 00:05:54,960
medicines. 
And all of these videos are like

88
00:05:54,960 --> 00:05:57,560
sort of rampant. 
They are almost everywhere on 

89
00:05:57,560 --> 00:06:00,280
Facebook, on Instagram and stuff
like that. 

90
00:06:00,280 --> 00:06:04,360
So now I think what you're 
seeing is now the first 

91
00:06:04,360 --> 00:06:07,200
iteration of how these things 
are playing out. 

92
00:06:07,880 --> 00:06:12,200
And the reason why I guess a lot
of people and a lot of fact 

93
00:06:12,200 --> 00:06:15,120
checkers and researchers and 
technologists are worried is 

94
00:06:15,120 --> 00:06:19,840
because of the level of 
sophistication and and the scale

95
00:06:19,840 --> 00:06:23,840
that this sort of technology 
offers to bad actors. 

96
00:06:24,120 --> 00:06:28,000
So it's very easy to do 
something and you know just 

97
00:06:28,280 --> 00:06:32,480
amplify it on a very large scale
and it's and it's a lot more 

98
00:06:32,480 --> 00:06:36,880
difficult to catch. 
I would say the the reason why 

99
00:06:36,880 --> 00:06:41,680
this thing here is problematic 
is also because the tools that 

100
00:06:41,680 --> 00:06:46,360
we currently have to catch and 
detect these things are not very

101
00:06:46,360 --> 00:06:50,240
reliable, especially with AI, 
audio, voice clones. 

102
00:06:50,480 --> 00:06:54,320
Just by hearing it the average 
person or even a fact checker 

103
00:06:54,320 --> 00:06:57,280
trained fact checker, you would 
not be able to spot the 

104
00:06:57,280 --> 00:07:01,280
difference. 
And plus I guess the human brain

105
00:07:01,280 --> 00:07:06,320
has evolved in such a way that 
it overcompensates for, you 

106
00:07:06,320 --> 00:07:09,320
know, discrepancies in audio. 
So if I tell you that I'm 

107
00:07:09,320 --> 00:07:13,760
playing a voice of so and so 
person, and even if it has a few

108
00:07:13,760 --> 00:07:19,120
glitches, your brain will still 
make that leap to try to make it

109
00:07:19,120 --> 00:07:21,840
sound like the person I've just 
told you so. 

110
00:07:21,840 --> 00:07:27,200
In that sense, AI voice clones 
are far more dangerous than AI 

111
00:07:27,200 --> 00:07:31,400
based images or videos because 
it is harder to catch. 

112
00:07:32,440 --> 00:07:36,160
Karen, could you talk a bit 
about the misinformation sort of

113
00:07:36,560 --> 00:07:41,080
industry in India, especially 
given India is both a massive 

114
00:07:41,080 --> 00:07:44,720
creator as well as a massive 
consumer of misinformation? 

115
00:07:45,680 --> 00:07:49,320
The kind of misinformation that 
we see in India is, is a bit 

116
00:07:49,320 --> 00:07:53,000
unique in the sense that the two
big baskets if I have to 

117
00:07:53,000 --> 00:07:58,040
categorize misinformation in the
country, it would be religion 

118
00:07:58,040 --> 00:08:01,520
and politics. 
And now increasingly the the 

119
00:08:01,520 --> 00:08:03,920
separation between the two is 
blurring. 

120
00:08:04,440 --> 00:08:07,880
So that is something we don't 
see in a lot of other countries 

121
00:08:07,880 --> 00:08:12,480
where you have so many claims 
around religion that is like a 

122
00:08:12,480 --> 00:08:14,120
unique feature that we see in 
India. 

123
00:08:14,120 --> 00:08:17,680
But generally I would say, you 
know the the sort of 

124
00:08:17,680 --> 00:08:21,240
misinformation that we see tends
to come from the right wing 

125
00:08:21,880 --> 00:08:26,760
purely on the basis of the fact 
that there is a nationalistic 

126
00:08:26,760 --> 00:08:32,600
government in power and there is
a tide of majoritarianism that 

127
00:08:33,440 --> 00:08:37,320
is sweeping the country. 
So the sort of narrative that 

128
00:08:37,320 --> 00:08:40,520
accompanies it will also see a 
lot of misinformation. 

129
00:08:41,000 --> 00:08:44,600
That is not to say that there's 
no misinformation coming from 

130
00:08:44,600 --> 00:08:49,920
the left, but it's the scale and
the reach are not comparable at 

131
00:08:49,920 --> 00:08:52,640
all. 
It's almost like negligible 

132
00:08:52,920 --> 00:08:57,080
majority of it is coming from 
the right wing ecosystem and 

133
00:08:57,120 --> 00:09:01,280
that ecosystem has a lot of 
different motivations for 

134
00:09:01,640 --> 00:09:05,280
creating misinformation. 
So one is that it's very 

135
00:09:05,280 --> 00:09:08,080
lucrative in India to create 
misinformation. 

136
00:09:08,080 --> 00:09:13,080
You know, if you are a YouTube 
or a content creator, then then 

137
00:09:13,080 --> 00:09:16,720
sort of jumping onto this 
bandwagon, you can get a lot of 

138
00:09:16,720 --> 00:09:18,960
traction and make a lot of 
money. 

139
00:09:18,960 --> 00:09:22,320
And that is what we tend to see.
Right earlier it was Facebook 

140
00:09:22,320 --> 00:09:26,920
pages and accounts. 
Now it's shifted to YouTube 

141
00:09:26,920 --> 00:09:31,400
videos and influencers and and 
that sort of ecosystem is 

142
00:09:31,400 --> 00:09:35,920
financially very rewarding. 
Then you also have a whole set 

143
00:09:35,920 --> 00:09:40,800
of people who ideologically are 
aligned to this Causeway. 

144
00:09:41,120 --> 00:09:47,160
So they may not be doing it for 
any financial gain as such. 

145
00:09:48,000 --> 00:09:52,640
They're doing it just because 
they 100% believe that what 

146
00:09:52,640 --> 00:09:55,320
they're saying is right and so 
they do it. 

147
00:09:55,320 --> 00:09:58,880
So these are the sort of people 
that we see as well. 

148
00:09:58,960 --> 00:10:03,920
And another unique thing I 
suppose in India and why we do 

149
00:10:03,920 --> 00:10:09,040
have a lot of misinformation is 
that around, you know 2015, 2016

150
00:10:09,040 --> 00:10:13,040
you had a lot of people come 
online for the first time with 

151
00:10:13,040 --> 00:10:17,080
mobile Internet rates falling 
through the floor. 

152
00:10:17,360 --> 00:10:22,840
So then you have a whole section
which you know is not that I 

153
00:10:22,840 --> 00:10:28,200
would say savvy when it comes to
on just in terms of dealing with

154
00:10:28,200 --> 00:10:32,240
online misinformation. 
So there are number of factors 

155
00:10:32,240 --> 00:10:38,480
that have sort of created this 
very unique situation that we 

156
00:10:38,480 --> 00:10:43,680
find ourselves in right now. 
In terms of India's handling of 

157
00:10:43,920 --> 00:10:48,560
misinformation, for one, how has
it been so far? 

158
00:10:48,880 --> 00:10:51,800
And you know, we now have two 
laws coming in. 

159
00:10:51,800 --> 00:10:56,320
You've got these sort of public 
announcements against deepfakes 

160
00:10:56,320 --> 00:10:59,640
and AI generated imagery and 
voice cloning. 

161
00:11:00,000 --> 00:11:02,040
How much do you see these 
things? 

162
00:11:02,040 --> 00:11:04,920
Solving it in the short term at 
least? 

163
00:11:06,120 --> 00:11:09,600
Yeah, I'm not too optimistic 
that the laws are going to be of

164
00:11:09,600 --> 00:11:11,400
much help. 
I think they're going to 

165
00:11:11,400 --> 00:11:16,000
complicate things further, 
especially if they are not, if 

166
00:11:16,000 --> 00:11:19,880
they don't take all stakeholders
on board and their opinions. 

167
00:11:20,200 --> 00:11:25,000
So in my opinion, generally laws
have not been very effective 

168
00:11:25,000 --> 00:11:28,600
when it comes to curbing 
misinformation. 

169
00:11:28,600 --> 00:11:31,360
You know, a few countries around
the world have have sort of 

170
00:11:31,360 --> 00:11:36,560
tried it, but it tends to lean 
towards, you know, either 

171
00:11:36,560 --> 00:11:39,880
censoring content or crushing 
dissent. 

172
00:11:40,080 --> 00:11:44,840
We have to be very careful plus 
the nature of this this problem 

173
00:11:44,840 --> 00:11:50,840
is so complex that you can't 
like have A1 size fits all 

174
00:11:50,840 --> 00:11:54,200
solution. 
So to be honest I'm not very 

175
00:11:54,200 --> 00:11:56,600
optimistic. 
There are some low hanging fruit

176
00:11:56,600 --> 00:11:59,400
which is easy to do. 
One is the election Commission 

177
00:11:59,400 --> 00:12:04,080
should should come out with some
sort of guidelines that prevent 

178
00:12:05,320 --> 00:12:09,880
the use of generative AI and 
deepfakes, especially in in any 

179
00:12:09,880 --> 00:12:14,400
sort of campaign, purely because
we don't know the extent to how 

180
00:12:14,400 --> 00:12:19,960
this tech can be abused. 
We have examples most recently 

181
00:12:20,560 --> 00:12:25,600
in Bangladesh where you had deep
fakes of candidates and their 

182
00:12:25,600 --> 00:12:30,160
voices were cloned and it was 
overlaid onto a video S it 

183
00:12:30,160 --> 00:12:33,280
appeared like the candidates 
saying that they are dropping 

184
00:12:33,280 --> 00:12:37,600
out of the elections and asking 
people not to vote for them. 

185
00:12:37,600 --> 00:12:41,960
So that kind of thing is 
terrifying when it comes to deep

186
00:12:41,960 --> 00:12:47,560
fake pornography and CSAM, which
is child sexual abuse material. 

187
00:12:48,080 --> 00:12:52,960
There is no sort of difference 
of opinion from anyone that this

188
00:12:52,960 --> 00:12:56,880
is bad for for everyone, right? 
Everybody's on the same page 

189
00:12:56,880 --> 00:12:59,600
that this stuff is harmful and 
it should be taken down. 

190
00:13:00,160 --> 00:13:03,280
That should be addressed on war 
footing. 

191
00:13:03,680 --> 00:13:08,560
From my own experience, I'll 
tell you that I have reported 

192
00:13:08,560 --> 00:13:15,120
several handles on Elon Musk's X
platform for tweeting and 

193
00:13:15,120 --> 00:13:19,920
posting explicit deepfake 
pornographic material, and the 

194
00:13:19,920 --> 00:13:23,840
reply that I've gotten from X 
has been that this does not 

195
00:13:23,840 --> 00:13:28,160
violate our community standards 
and it has happened on more than

196
00:13:28,160 --> 00:13:33,280
one occasion. 
Platforms also have wake up to 

197
00:13:33,280 --> 00:13:37,160
this and and do more and not 
just pay lip service to this 

198
00:13:37,160 --> 00:13:39,600
problem. 
The path ahead looks very 

199
00:13:39,600 --> 00:13:44,280
difficult. 
How effective are these platform

200
00:13:44,280 --> 00:13:48,240
based initiatives so far in 
terms of curbing the spread of 

201
00:13:48,240 --> 00:13:52,960
misinformation? 
So right now you have a very 

202
00:13:52,960 --> 00:13:57,640
unique situation where these 
platforms are almost imploding 

203
00:13:57,640 --> 00:14:01,000
like X. 
It's become this hub of 

204
00:14:01,120 --> 00:14:03,640
conspiracy theories and 
misinformation. 

205
00:14:03,640 --> 00:14:08,200
And we saw this during the 
Israel Hamas war where you had 

206
00:14:08,200 --> 00:14:13,440
misinformation on steroids. 
And it was like a dam breaking 

207
00:14:13,440 --> 00:14:16,760
in terms just in terms of the 
volume of and the amount of 

208
00:14:16,760 --> 00:14:21,440
misinformation and all of it was
playing out on X. 

209
00:14:21,840 --> 00:14:25,800
So X went and changed its blue 
tick verification system where 

210
00:14:25,800 --> 00:14:30,920
it allowed basically anyone to 
pay $8 and buy that blue tick 

211
00:14:30,920 --> 00:14:34,240
mark. 
And at the same time it 

212
00:14:34,680 --> 00:14:37,920
completely decimated its trust 
and safety teams. 

213
00:14:38,480 --> 00:14:42,520
Facebook as well, you know, laid
off a lot of people in in those 

214
00:14:42,520 --> 00:14:45,320
teams. 
So now what you're seeing are 

215
00:14:45,320 --> 00:14:49,120
like the consequences of those 
actions. 

216
00:14:49,520 --> 00:14:54,840
It brings me to the point that, 
you know, there is a structural 

217
00:14:55,440 --> 00:14:59,440
and fundamental problem with the
way these platforms are built. 

218
00:14:59,880 --> 00:15:04,920
They are built in a way that it 
fosters polarization, that the 

219
00:15:04,920 --> 00:15:11,160
most divisive content and 
conversation and material out 

220
00:15:11,160 --> 00:15:15,520
there gets traction. 
I also see there's a lot of 

221
00:15:16,920 --> 00:15:21,240
conflict of interest in the 
sense that a lot of these and 

222
00:15:21,240 --> 00:15:25,120
it's not just the platforms, I 
guess it's all big tech is that 

223
00:15:25,120 --> 00:15:29,400
on one hand they are saying that
you know they're worried about 

224
00:15:29,720 --> 00:15:32,120
AI based misinformation and all 
of that. 

225
00:15:32,520 --> 00:15:36,520
But on the other hand they are 
coming out with products and 

226
00:15:36,520 --> 00:15:41,080
technology and features that 
enable you to to use AI. 

227
00:15:41,440 --> 00:15:45,840
So let's say a Facebook comes 
out with a feature that enables 

228
00:15:46,320 --> 00:15:49,840
AI editing where you can change 
the background of a photo, you 

229
00:15:49,840 --> 00:15:52,040
can insert something that was 
never there. 

230
00:15:52,040 --> 00:15:55,080
You can in fact make a whole 
video with things that were 

231
00:15:55,080 --> 00:15:58,280
never in there. 
Now they use polar bears and 

232
00:15:58,280 --> 00:16:02,560
puppy dogs in in examples for, 
you know, these products. 

233
00:16:02,840 --> 00:16:05,400
But that is not what it's going 
to be used for, right? 

234
00:16:05,560 --> 00:16:09,760
People are going to use it with 
real life instances of people 

235
00:16:10,120 --> 00:16:13,440
and do all sorts of harm. 
So this conflict of interest 

236
00:16:13,440 --> 00:16:16,000
where you know on one hand 
you're saying oh we are doing 

237
00:16:16,000 --> 00:16:18,720
everything we can and we're 
using our tech and we're 

238
00:16:18,720 --> 00:16:21,600
partnering with fact checkers 
and and all of that. 

239
00:16:21,600 --> 00:16:25,000
And then on the other hand, you 
are building these products 

240
00:16:25,000 --> 00:16:28,960
which again fuel the same 
problem that we are fighting. 

241
00:16:29,680 --> 00:16:32,240
Karen Ribello says that these 
tools that allow the cloning of 

242
00:16:32,240 --> 00:16:35,040
face and voice should be more 
restricted than they are. 

243
00:16:35,280 --> 00:16:37,240
She points out that while they 
may be created for 

244
00:16:37,240 --> 00:16:39,600
entertainment, it doesn't take 
very long for them to be 

245
00:16:39,600 --> 00:16:42,800
misused. 
First of all, I don't think this

246
00:16:42,800 --> 00:16:45,400
technology should be widely 
available. 

247
00:16:45,960 --> 00:16:50,440
Like what is the point of 
allowing someone to clone 

248
00:16:50,440 --> 00:16:53,640
somebody else's voice? 
And if you are doing it, then 

249
00:16:53,640 --> 00:16:57,880
you need to build detectors 
within your product that are 

250
00:16:57,880 --> 00:17:02,040
free and not subscription based 
where anybody can use it. 

251
00:17:02,200 --> 00:17:07,200
These image generation websites,
they say they have safeguards, 

252
00:17:07,640 --> 00:17:12,319
but if you are determined to 
sort of bypass those rules, you 

253
00:17:12,319 --> 00:17:14,839
can find a way to beat the 
system. 

254
00:17:15,200 --> 00:17:19,520
And we've seen that, right. 
You have examples of AI based 

255
00:17:19,520 --> 00:17:22,040
images used in the Israel Hamas 
war. 

256
00:17:22,400 --> 00:17:26,440
I think they need to really 
clarify what their sort of 

257
00:17:26,440 --> 00:17:32,440
approach is to tackling this 
problem because as of now, I 

258
00:17:32,440 --> 00:17:36,160
think all their efforts are like
coming to naughty. 

259
00:17:36,880 --> 00:17:40,640
But with a lot of these AI 
tools, The thing is also they're

260
00:17:40,640 --> 00:17:44,400
seen as tools to foster 
creativity, a sort of leap 

261
00:17:44,400 --> 00:17:47,440
forward in terms of the human 
imagination, right? 

262
00:17:47,720 --> 00:17:51,400
Would say markers like we had, 
say, a reverse image search 

263
00:17:51,400 --> 00:17:55,080
earlier or something like that. 
Do markers like that actually 

264
00:17:55,080 --> 00:17:59,200
help detect a fake? 
Do people actually invest their 

265
00:17:59,200 --> 00:18:04,760
time in finding out it's fake? 
So from our experience, I mean, 

266
00:18:05,560 --> 00:18:09,400
I I don't think people spend 
that much time and this argument

267
00:18:09,400 --> 00:18:13,040
of these tools have been built 
for enhancing creativity or 

268
00:18:13,040 --> 00:18:16,680
productivity. 
It sort of reminds me of the 

269
00:18:16,680 --> 00:18:21,800
conversation that happened when 
social media platforms were new 

270
00:18:21,800 --> 00:18:26,600
around 15 or 16 years ago. 
Social media platforms came 

271
00:18:26,600 --> 00:18:32,240
into, you know, existence with 
people promising us the idea of 

272
00:18:32,240 --> 00:18:37,120
building communities. 
Now how much of of community 

273
00:18:37,120 --> 00:18:40,680
building has happened over 
social media than actual 

274
00:18:40,680 --> 00:18:43,280
destruction is a conversation 
for another time. 

275
00:18:43,760 --> 00:18:48,600
You know, we are making the same
mistakes with generative AI 

276
00:18:48,600 --> 00:18:53,360
because the problems that all 
these social media platforms 

277
00:18:53,800 --> 00:18:57,880
have faced. 
I see all of the same problems 

278
00:18:57,880 --> 00:19:02,280
with generative AI as well. 
When it's in its first few 

279
00:19:02,360 --> 00:19:05,280
iterations, right now is the 
time to talk about it and fix 

280
00:19:05,280 --> 00:19:07,360
these problems before it gets 
worse. 

281
00:19:08,240 --> 00:19:10,880
A lot of the debate around the 
curbing of misinformation has 

282
00:19:10,880 --> 00:19:14,160
come to center around artificial
intelligence based tools that 

283
00:19:14,160 --> 00:19:15,760
are used to create 
misinformation. 

284
00:19:16,040 --> 00:19:19,240
What's spoken of less now is 
that platforms like X, YouTube, 

285
00:19:19,240 --> 00:19:22,320
and WhatsApp are still very 
active sources to spread 

286
00:19:22,320 --> 00:19:25,200
misinformation in spite of all 
the measures that have been 

287
00:19:25,200 --> 00:19:28,120
taken over the years. 
Karen Rubello explains the 

288
00:19:28,120 --> 00:19:31,560
things that platforms can do as 
a sort of quick fix to curb the 

289
00:19:31,560 --> 00:19:35,320
spread of misinformation. 
There have been studies which 

290
00:19:35,320 --> 00:19:40,960
show that, you know, when people
come across information that has

291
00:19:40,960 --> 00:19:45,360
been flagged by the platform as 
misleading, whether it's on X or

292
00:19:45,360 --> 00:19:50,520
on Facebook or anything, there 
is a sort of slowing down effect

293
00:19:50,600 --> 00:19:53,120
that happens in terms of its 
velocity. 

294
00:19:53,480 --> 00:19:57,040
We can't just say, you know, if 
whoever wants to forward it will

295
00:19:57,040 --> 00:20:00,680
still forward it. 
But yeah, it makes a difference 

296
00:20:00,680 --> 00:20:03,480
in terms of slowing down its 
distribution for sure. 

297
00:20:03,960 --> 00:20:09,800
In addition, steps like. 
Enforcing some sort of monetary 

298
00:20:10,360 --> 00:20:14,320
penalty or something. 
So, you know, making sure that 

299
00:20:14,320 --> 00:20:18,160
pages that or accounts that 
share this information can't 

300
00:20:18,160 --> 00:20:21,080
earn a lot by sharing 
misinformation, those steps help

301
00:20:21,080 --> 00:20:25,200
as well because what we saw 
during the Israel Hamas war is 

302
00:20:25,200 --> 00:20:29,720
that a lot of the misinformation
came from verified blue tick 

303
00:20:29,720 --> 00:20:33,600
accounts. 
Now a policy that says, you 

304
00:20:33,600 --> 00:20:36,800
know, they will not be able to 
monetize if their content has 

305
00:20:36,800 --> 00:20:40,520
been marked or flagged, that 
certainly helps because it 

306
00:20:40,600 --> 00:20:43,480
incentivizes the creator of 
misinformation. 

307
00:20:43,720 --> 00:20:46,840
So small steps like that 
definitely play a huge role. 

308
00:20:47,760 --> 00:20:52,200
There's been a lot written about
the work of a fact checker and 

309
00:20:52,200 --> 00:20:54,480
the sort of thanklessness that 
it is. 

310
00:20:55,200 --> 00:21:00,600
What has it been like for you to
work on curbing the spread of 

311
00:21:00,600 --> 00:21:04,560
misinformation, especially 
given, I mean, even AWF report 

312
00:21:04,560 --> 00:21:07,440
kind of points out that you're 
literally sitting in the hub of 

313
00:21:07,440 --> 00:21:10,600
it in some ways. 
Yeah. 

314
00:21:11,920 --> 00:21:15,080
So I was thinking about it the 
other day, and I figured that 

315
00:21:15,080 --> 00:21:19,920
I've spent almost like half my 
career in this space, which is a

316
00:21:19,920 --> 00:21:23,840
long time, so to say. 
Misinformation has definitely 

317
00:21:23,840 --> 00:21:28,320
evolved in the sense that it's 
become more pervasive, it's 

318
00:21:28,320 --> 00:21:33,320
become more sophisticated and 
it's definitely more influential

319
00:21:33,520 --> 00:21:36,360
in terms of its effects that we 
see. 

320
00:21:37,160 --> 00:21:41,640
Initially that I used to think 
that only negative news events 

321
00:21:41,640 --> 00:21:45,080
leads to to a lot of 
misinformation and and that 

322
00:21:45,080 --> 00:21:47,080
still holds true to a large 
extent. 

323
00:21:47,480 --> 00:21:51,800
But what we're seeing right now 
is that, you know, almost every 

324
00:21:51,840 --> 00:21:55,640
newsworthy event is sort of 
accompanied by misinformation in

325
00:21:55,640 --> 00:21:58,000
India. 
So it's almost become par for 

326
00:21:58,000 --> 00:22:01,720
the course. 
And yeah, it's an uphill battle 

327
00:22:01,720 --> 00:22:08,200
and and personally for me it's 
been hard but completely worth 

328
00:22:08,200 --> 00:22:10,240
it. 
I wouldn't change it any other 

329
00:22:10,240 --> 00:22:15,640
way. 
Today's episode was produced by 

330
00:22:15,640 --> 00:22:20,160
Jayaraj Singh and Sahil Gupta. 
For a daily spotlight on people,

331
00:22:20,240 --> 00:22:23,000
ideas and stories that matter, 
subscribe to us. 

332
00:22:23,360 --> 00:22:27,800
We're available on TOI plus 
Spotify, Apple, Google Podcasts 

333
00:22:28,000 --> 00:22:29,760
and all other platforms of your 
choice. 

334
00:22:30,240 --> 00:22:35,240
For any new steps, e-mail us at 
toipodcast at Timesinternet in.