1
00:00:00,000 --> 00:00:05,000
Okay, so have you ever wondered if AI could actually predict

2
00:00:06,360 --> 00:00:07,560
who will win an election?

3
00:00:07,560 --> 00:00:09,760
Oh yeah, especially during an election year,

4
00:00:09,760 --> 00:00:12,000
it's definitely something you think about.

5
00:00:12,000 --> 00:00:14,200
Right, exactly, and that's what this research paper

6
00:00:14,200 --> 00:00:15,360
we're looking at is all about.

7
00:00:15,360 --> 00:00:16,800
Yeah, and they use these things

8
00:00:16,800 --> 00:00:18,600
called large language models.

9
00:00:18,600 --> 00:00:19,600
LLMs.

10
00:00:19,600 --> 00:00:21,200
Yeah, you got it.

11
00:00:21,200 --> 00:00:22,440
And those are super interesting

12
00:00:22,440 --> 00:00:24,720
because they don't just crunch numbers.

13
00:00:24,720 --> 00:00:26,960
Oh, so it's not just statistics?

14
00:00:26,960 --> 00:00:30,120
Nope, they actually can understand and generate

15
00:00:30,120 --> 00:00:32,880
like human-like text, you know?

16
00:00:32,880 --> 00:00:33,720
Wow.

17
00:00:33,720 --> 00:00:35,960
So think about them like sifting through all these speeches

18
00:00:35,960 --> 00:00:39,440
and social media posts, even news articles,

19
00:00:39,440 --> 00:00:41,440
just to kind of get a sense of how people feel

20
00:00:41,440 --> 00:00:42,840
about the candidates.

21
00:00:42,840 --> 00:00:43,960
That's really wild.

22
00:00:43,960 --> 00:00:45,520
I like, I never thought about it that way.

23
00:00:45,520 --> 00:00:46,360
Yeah, it's pretty cool.

24
00:00:46,360 --> 00:00:48,440
But I can already see how predicting elections

25
00:00:48,440 --> 00:00:52,160
is like way more complicated than say predicting the weather.

26
00:00:52,160 --> 00:00:53,000
Oh, absolutely.

27
00:00:53,000 --> 00:00:54,600
Because it's not just data points, right?

28
00:00:54,600 --> 00:00:56,840
Right, you're dealing with people like their opinions,

29
00:00:56,840 --> 00:00:58,600
all these factors that come together.

30
00:00:58,600 --> 00:01:00,080
And actually one of the biggest challenges

31
00:01:00,080 --> 00:01:01,480
for the researchers was like getting

32
00:01:01,480 --> 00:01:04,640
enough detailed voter information.

33
00:01:04,640 --> 00:01:05,920
Oh yeah, I can imagine,

34
00:01:05,920 --> 00:01:07,360
because that's probably really hard to come by.

35
00:01:07,360 --> 00:01:08,200
Yeah, exactly.

36
00:01:08,200 --> 00:01:11,000
It's expensive and there's all these privacy concerns.

37
00:01:11,000 --> 00:01:12,560
So they had to get a little creative.

38
00:01:12,560 --> 00:01:13,720
Okay, so what did they do?

39
00:01:13,720 --> 00:01:16,560
So they used this thing called synthetic data.

40
00:01:16,560 --> 00:01:17,600
Synthetic data.

41
00:01:17,600 --> 00:01:20,400
Yeah, basically they created like virtual voters.

42
00:01:20,400 --> 00:01:22,160
Wait, so they just like made up voters?

43
00:01:22,160 --> 00:01:23,240
Well, not exactly.

44
00:01:23,240 --> 00:01:26,160
Think of it more like building a simulated world,

45
00:01:26,160 --> 00:01:28,200
you know, and they populate this world

46
00:01:28,200 --> 00:01:29,440
with these virtual voters.

47
00:01:29,440 --> 00:01:32,320
And each voter has like a realistic background

48
00:01:32,320 --> 00:01:34,440
and a demographic profile.

49
00:01:34,440 --> 00:01:37,680
And all that is based on publicly available information.

50
00:01:37,680 --> 00:01:39,560
So they're not just making stuff up.

51
00:01:39,560 --> 00:01:41,040
Oh, that's pretty clever.

52
00:01:41,040 --> 00:01:42,920
Yeah, this way they could test their AI

53
00:01:42,920 --> 00:01:44,560
on like a massive scale.

54
00:01:44,560 --> 00:01:45,520
Okay, that makes sense.

55
00:01:45,520 --> 00:01:48,520
Yeah, without running into those real world data limitations.

56
00:01:48,520 --> 00:01:50,160
But wouldn't just looking at demographics

57
00:01:50,160 --> 00:01:52,320
be kind of limiting?

58
00:01:52,320 --> 00:01:53,240
That's a good point.

59
00:01:53,240 --> 00:01:55,960
Because elections are about way more than just age

60
00:01:55,960 --> 00:01:56,920
to where someone lives.

61
00:01:56,920 --> 00:01:58,520
Right, yeah, you're absolutely right.

62
00:01:58,520 --> 00:01:59,840
The researchers knew that.

63
00:01:59,840 --> 00:02:00,680
Okay.

64
00:02:00,680 --> 00:02:02,840
So to capture that human element,

65
00:02:02,840 --> 00:02:04,720
you know, they took it a step further

66
00:02:04,720 --> 00:02:07,160
and factored in the political climate.

67
00:02:07,160 --> 00:02:08,200
The political climate,

68
00:02:08,200 --> 00:02:10,160
so like what was going on at the time?

69
00:02:10,160 --> 00:02:12,840
Exactly, like each candidate's policy

70
00:02:12,840 --> 00:02:14,880
positions their past experiences

71
00:02:14,880 --> 00:02:16,680
and even how they were perceived

72
00:02:16,680 --> 00:02:19,760
on that whole liberal conservative spectrum.

73
00:02:19,760 --> 00:02:22,000
So they weren't just throwing a bunch of random data

74
00:02:22,000 --> 00:02:22,880
at the AI?

75
00:02:22,880 --> 00:02:25,920
No, no, they were really trying to make it think

76
00:02:25,920 --> 00:02:27,240
like a person.

77
00:02:27,240 --> 00:02:28,080
No.

78
00:02:28,080 --> 00:02:30,640
Trying to simulate how someone might actually weigh

79
00:02:30,640 --> 00:02:31,920
all these different factors

80
00:02:31,920 --> 00:02:33,520
before deciding who to vote for.

81
00:02:33,520 --> 00:02:34,480
That's really interesting.

82
00:02:34,480 --> 00:02:35,960
Yeah, and to do that,

83
00:02:35,960 --> 00:02:38,960
they used a method called multi-step reasoning.

84
00:02:38,960 --> 00:02:40,320
Multi-step reasoning, okay.

85
00:02:40,320 --> 00:02:43,080
So instead of just giving the AI all the info

86
00:02:43,080 --> 00:02:44,960
at once they broke the process down,

87
00:02:44,960 --> 00:02:47,000
like imagine you wouldn't just look at someone's age

88
00:02:47,000 --> 00:02:48,400
and be like, oh, they'll vote this way.

89
00:02:48,400 --> 00:02:49,240
Right.

90
00:02:49,240 --> 00:02:50,760
You'd consider their background,

91
00:02:50,760 --> 00:02:53,240
what they're saying, what's happening politically, you know.

92
00:02:53,240 --> 00:02:55,200
Yeah, it makes it a lot more nuanced for sure.

93
00:02:55,200 --> 00:02:57,840
Exactly, and that's what the AI was doing.

94
00:02:57,840 --> 00:03:00,280
So first it looked at where a synthetic voter

95
00:03:00,280 --> 00:03:01,960
would fall on the political spectrum

96
00:03:01,960 --> 00:03:03,680
based on their virtual profile

97
00:03:03,680 --> 00:03:05,600
and the current political landscape.

98
00:03:05,600 --> 00:03:07,560
And then in a separate step,

99
00:03:07,560 --> 00:03:10,560
it would predict how that voter would cast their ballot.

100
00:03:10,560 --> 00:03:12,720
Okay, so they've trained this AI,

101
00:03:12,720 --> 00:03:16,400
given it all this data in a way to reason through it.

102
00:03:16,400 --> 00:03:18,200
But did it actually work?

103
00:03:18,200 --> 00:03:19,760
Yeah, that's the big question, right?

104
00:03:19,760 --> 00:03:22,600
Right, like did it predict the outcome of real elections?

105
00:03:22,600 --> 00:03:26,280
Well, you know, no AI is a perfect crystal ball.

106
00:03:26,280 --> 00:03:27,120
Of course not.

107
00:03:27,120 --> 00:03:28,520
Especially when it comes to something

108
00:03:28,520 --> 00:03:30,080
as complex as an election.

109
00:03:30,080 --> 00:03:32,280
But they did put their AI to the test

110
00:03:32,280 --> 00:03:37,160
using data from the 2016 and 2020 US presidential elections.

111
00:03:37,160 --> 00:03:38,000
Okay.

112
00:03:38,000 --> 00:03:39,840
And the results were surprisingly good,

113
00:03:39,840 --> 00:03:41,960
especially with that multi-step reasoning.

114
00:03:41,960 --> 00:03:42,800
Really?

115
00:03:42,800 --> 00:03:44,200
Yeah, and you know what's really interesting?

116
00:03:44,200 --> 00:03:45,720
They actually took this AI,

117
00:03:45,720 --> 00:03:47,720
the one they trained on past elections

118
00:03:47,720 --> 00:03:50,600
and used it to predict the 2024 election.

119
00:03:50,600 --> 00:03:51,440
Wait, hold on.

120
00:03:51,440 --> 00:03:52,920
They're predicting the future with this thing.

121
00:03:52,920 --> 00:03:53,760
Well, you have to remember,

122
00:03:53,760 --> 00:03:56,560
this prediction was made like back in November 2023.

123
00:03:56,560 --> 00:03:57,400
Okay.

124
00:03:57,400 --> 00:03:58,760
So it's really based on a snapshot

125
00:03:58,760 --> 00:04:01,000
of the political scene at that time, you know?

126
00:04:01,000 --> 00:04:02,120
Yeah, that makes more sense.

127
00:04:02,120 --> 00:04:04,080
It's not like a sure thing or anything.

128
00:04:04,080 --> 00:04:05,640
It's more like a what if scenario.

129
00:04:05,640 --> 00:04:06,640
Right, okay, I get it.

130
00:04:06,640 --> 00:04:08,840
Just based on the information they had at the time.

131
00:04:08,840 --> 00:04:11,840
So did the AI come up with anything surprising?

132
00:04:11,840 --> 00:04:13,280
Well, one of the most interesting things

133
00:04:13,280 --> 00:04:15,440
was a potential shift in Wisconsin.

134
00:04:15,440 --> 00:04:16,280
Wisconsin?

135
00:04:16,280 --> 00:04:17,120
Oh, that's interesting.

136
00:04:17,120 --> 00:04:20,320
Yeah, actually predicted a Trump win in 2024.

137
00:04:20,320 --> 00:04:21,160
Really?

138
00:04:21,160 --> 00:04:22,760
Yeah, which would be a pretty big deal.

139
00:04:22,760 --> 00:04:23,800
Yeah, for sure.

140
00:04:23,800 --> 00:04:26,680
Especially considering how close it was in 2020.

141
00:04:26,680 --> 00:04:27,640
Exactly.

142
00:04:27,640 --> 00:04:29,800
Hmm, so what's behind that shift?

143
00:04:29,800 --> 00:04:32,480
I wonder like, is it demographics,

144
00:04:32,480 --> 00:04:35,360
the political climate or something else entirely?

145
00:04:35,360 --> 00:04:37,160
You know, the paper doesn't really go into

146
00:04:37,160 --> 00:04:38,720
the specifics for each state.

147
00:04:38,720 --> 00:04:39,560
Oh, okay.

148
00:04:39,560 --> 00:04:41,440
But remember, the AI was looking

149
00:04:41,440 --> 00:04:43,160
at a whole bunch of factors,

150
00:04:43,160 --> 00:04:45,200
you know, everything from policy positions

151
00:04:45,200 --> 00:04:48,680
to how each candidate resonated with different demographics.

152
00:04:48,680 --> 00:04:49,520
Right, and we don't even know

153
00:04:49,520 --> 00:04:51,160
the Democratic nominee will be yet.

154
00:04:51,160 --> 00:04:52,000
Yeah, that's true.

155
00:04:52,000 --> 00:04:54,560
The AI was just assuming it would be Kamala Harris.

156
00:04:54,560 --> 00:04:57,800
Huh, it's crazy to think how much things can change

157
00:04:57,800 --> 00:05:00,440
between now and the actual election.

158
00:05:00,440 --> 00:05:01,280
Oh, for sure.

159
00:05:01,280 --> 00:05:03,080
Like the whole picture could be totally different.

160
00:05:03,080 --> 00:05:04,600
Yeah, exactly.

161
00:05:04,600 --> 00:05:07,640
These AI predictions are tied to the data

162
00:05:07,640 --> 00:05:09,920
that was available at a specific point in time.

163
00:05:09,920 --> 00:05:10,920
It's like a moving target.

164
00:05:10,920 --> 00:05:11,760
You got it.

165
00:05:11,760 --> 00:05:15,480
So even though it's tempting to see these 2024 predictions

166
00:05:15,480 --> 00:05:17,120
as like some kind of fortune telling.

167
00:05:17,120 --> 00:05:17,960
Right.

168
00:05:17,960 --> 00:05:19,120
We really have to remember,

169
00:05:19,120 --> 00:05:22,160
they're just one piece of a much larger puzzle.

170
00:05:22,160 --> 00:05:24,200
And a constantly changing puzzle at that.

171
00:05:24,200 --> 00:05:25,040
Exactly.

172
00:05:25,040 --> 00:05:27,000
You know, the real value of this research

173
00:05:27,000 --> 00:05:30,000
isn't about getting the future exactly right.

174
00:05:30,000 --> 00:05:32,480
It's more about what the AI can tell us

175
00:05:32,480 --> 00:05:33,720
about human behavior.

176
00:05:33,720 --> 00:05:34,560
Oh, I see.

177
00:05:34,560 --> 00:05:35,600
Especially when it comes to something

178
00:05:35,600 --> 00:05:37,360
as complex as politics.

179
00:05:37,360 --> 00:05:39,160
So it's like getting a fresh perspective

180
00:05:39,160 --> 00:05:41,000
on how elections really work.

181
00:05:41,000 --> 00:05:43,200
Yeah, like a new set of eyes, you know.

182
00:05:43,200 --> 00:05:45,800
Because the AI can see those patterns and trends

183
00:05:45,800 --> 00:05:46,680
that we might miss.

184
00:05:46,680 --> 00:05:47,520
Exactly.

185
00:05:47,520 --> 00:05:48,680
And help us understand how voters

186
00:05:48,680 --> 00:05:50,120
actually make their decisions.

187
00:05:50,120 --> 00:05:50,960
Right.

188
00:05:50,960 --> 00:05:53,080
And that could lead to some really interesting discussions.

189
00:05:53,080 --> 00:05:53,920
You know?

190
00:05:53,920 --> 00:05:54,920
Yeah, I could see that.

191
00:05:54,920 --> 00:05:58,360
And help us move beyond like those simple assumptions

192
00:05:58,360 --> 00:05:59,200
we all have.

193
00:05:59,200 --> 00:06:01,800
To like really dig into the issues that matter.

194
00:06:01,800 --> 00:06:04,680
So this AI model, it's not just about like

195
00:06:04,680 --> 00:06:07,400
throwing data into a black box and getting a prediction.

196
00:06:07,400 --> 00:06:08,240
Nope, not at all.

197
00:06:08,240 --> 00:06:10,800
They really put a lot of thought into how they built it.

198
00:06:10,800 --> 00:06:11,640
Yeah.

199
00:06:11,640 --> 00:06:13,040
To make sure it could actually simulate

200
00:06:13,040 --> 00:06:14,360
human decision making.

201
00:06:14,360 --> 00:06:15,200
They did.

202
00:06:15,200 --> 00:06:16,960
And that's why they used that combination

203
00:06:16,960 --> 00:06:19,680
of real world data and synthetic data.

204
00:06:19,680 --> 00:06:21,160
Oh, right, the synthetic voters.

205
00:06:21,160 --> 00:06:22,000
Exactly.

206
00:06:22,000 --> 00:06:24,520
So they could run all kinds of simulations, you know,

207
00:06:24,520 --> 00:06:27,280
and test different scenarios without needing access

208
00:06:27,280 --> 00:06:28,880
to real people's private info.

209
00:06:28,880 --> 00:06:30,280
Like a giant virtual lab.

210
00:06:30,280 --> 00:06:31,320
Yeah, it's a great way to put it.

211
00:06:31,320 --> 00:06:33,080
Where you can play with all these different variables

212
00:06:33,080 --> 00:06:34,640
and see how they affect the outcome.

213
00:06:34,640 --> 00:06:35,480
Exactly.

214
00:06:35,480 --> 00:06:37,680
They could change demographics, the political climate,

215
00:06:37,680 --> 00:06:40,600
really explore how those changes might play out.

216
00:06:40,600 --> 00:06:41,440
That's so cool.

217
00:06:41,440 --> 00:06:43,320
And then you have that multi-step reasoning

218
00:06:43,320 --> 00:06:44,160
we talked about.

219
00:06:44,160 --> 00:06:45,000
Right.

220
00:06:45,000 --> 00:06:47,880
And that's why it tries to mimic how a real person

221
00:06:47,880 --> 00:06:49,280
thinks about voting.

222
00:06:49,280 --> 00:06:51,400
Yeah, so they broke the process down

223
00:06:51,400 --> 00:06:54,120
into those smaller, more nuanced steps.

224
00:06:54,120 --> 00:06:57,200
So first they figure out where a synthetic voter falls

225
00:06:57,200 --> 00:06:58,440
on the political spectrum.

226
00:06:58,440 --> 00:06:58,960
Right.

227
00:06:58,960 --> 00:07:01,720
By giving the AI info about the voter's background

228
00:07:01,720 --> 00:07:04,240
and the policy positions of both parties.

229
00:07:04,240 --> 00:07:04,840
OK.

230
00:07:04,840 --> 00:07:07,440
And then they could categorize the voter, you know,

231
00:07:07,440 --> 00:07:10,920
as like very liberal, somewhat conservative, moderate,

232
00:07:10,920 --> 00:07:11,600
and so on.

233
00:07:11,600 --> 00:07:13,000
So it's not just about demographics.

234
00:07:13,000 --> 00:07:15,120
It's about understanding their political leanings too.

235
00:07:15,120 --> 00:07:16,280
Exactly.

236
00:07:16,280 --> 00:07:18,120
And that leads us to the second step,

237
00:07:18,120 --> 00:07:23,000
where they use all that information, the voter's profile,

238
00:07:23,000 --> 00:07:25,480
their estimated political leaning,

239
00:07:25,480 --> 00:07:27,440
the details about the candidates,

240
00:07:27,440 --> 00:07:29,320
and then they predict their vote.

241
00:07:29,320 --> 00:07:31,160
It's like a two-part process.

242
00:07:31,160 --> 00:07:34,560
Understand the voter, then simulate their decision making.

243
00:07:34,560 --> 00:07:36,360
And did that approach actually work?

244
00:07:36,360 --> 00:07:38,800
Did that multi-step reasoning hold up?

245
00:07:38,800 --> 00:07:41,280
Yeah, when they tested it against real election results.

246
00:07:41,280 --> 00:07:42,400
Well, no model is perfect.

247
00:07:42,400 --> 00:07:45,360
But when they tested it on state level results,

248
00:07:45,360 --> 00:07:46,920
the outcomes were pretty interesting.

249
00:07:46,920 --> 00:07:47,440
Oh, really?

250
00:07:47,440 --> 00:07:48,360
Like what?

251
00:07:48,360 --> 00:07:50,000
Well, it seemed to be especially good

252
00:07:50,000 --> 00:07:52,640
at predicting the outcomes in swing states, you know?

253
00:07:52,640 --> 00:07:53,320
Oh, wow.

254
00:07:53,320 --> 00:07:54,920
The ones that are always super close.

255
00:07:54,920 --> 00:07:55,420
Exactly.

256
00:07:55,420 --> 00:07:58,240
Those nail biters that often come down to just a handful

257
00:07:58,240 --> 00:07:58,720
of votes.

258
00:07:58,720 --> 00:07:59,920
Like it was picking up on something

259
00:07:59,920 --> 00:08:01,320
that regular polls were missing.

260
00:08:01,320 --> 00:08:02,760
That's what it looks like.

261
00:08:02,760 --> 00:08:06,040
So this AI could actually help us understand those key states

262
00:08:06,040 --> 00:08:07,040
even better.

263
00:08:07,040 --> 00:08:09,920
It definitely seems like it's pointing in that direction.

264
00:08:09,920 --> 00:08:12,720
Because those are the states that can really decide the whole

265
00:08:12,720 --> 00:08:13,240
election.

266
00:08:13,240 --> 00:08:13,880
Absolutely.

267
00:08:13,880 --> 00:08:17,600
And you know, this also highlights the potential of AI

268
00:08:17,600 --> 00:08:20,320
to complement our understanding of human behavior.

269
00:08:20,320 --> 00:08:22,920
Not to replace it, but to work alongside it.

270
00:08:22,920 --> 00:08:23,640
Exactly.

271
00:08:23,640 --> 00:08:25,520
Especially in a field like politics

272
00:08:25,520 --> 00:08:27,800
where the stakes are so high.

273
00:08:27,800 --> 00:08:29,280
It's like opening up a whole new chapter

274
00:08:29,280 --> 00:08:30,760
in how we think about elections.

275
00:08:30,760 --> 00:08:32,120
That's a great way to put it.

276
00:08:32,120 --> 00:08:34,320
This is really making me think, like,

277
00:08:34,320 --> 00:08:36,800
what else could we use this kind of AI for?

278
00:08:36,800 --> 00:08:37,400
Oh, yeah.

279
00:08:37,400 --> 00:08:39,080
Like beyond just predicting elections.

280
00:08:39,080 --> 00:08:40,720
I know it's pretty exciting to think about.

281
00:08:40,720 --> 00:08:43,000
Like if they can understand all those factors that

282
00:08:43,000 --> 00:08:44,680
go into voting behavior.

283
00:08:44,680 --> 00:08:45,040
Right.

284
00:08:45,040 --> 00:08:47,000
Imagine applying it to other stuff, too.

285
00:08:47,000 --> 00:08:47,680
Like what?

286
00:08:47,680 --> 00:08:49,920
Like how people respond to new policies

287
00:08:49,920 --> 00:08:51,600
or make health care choices.

288
00:08:51,600 --> 00:08:52,120
Oh, wow.

289
00:08:52,120 --> 00:08:55,840
Even predict social trends, like what ideas will catch on

290
00:08:55,840 --> 00:08:57,800
or how public opinion might change.

291
00:08:57,800 --> 00:09:00,000
Yeah, like it could tell us what the next big thing is.

292
00:09:00,000 --> 00:09:00,480
Right.

293
00:09:00,480 --> 00:09:00,920
Exactly.

294
00:09:00,920 --> 00:09:01,520
That's great.

295
00:09:01,520 --> 00:09:04,680
It's like we're unlocking this whole new level of understanding

296
00:09:04,680 --> 00:09:05,960
about human decisions.

297
00:09:05,960 --> 00:09:06,320
I know.

298
00:09:06,320 --> 00:09:07,440
It's kind of mind blowing.

299
00:09:07,440 --> 00:09:08,160
Yeah.

300
00:09:08,160 --> 00:09:12,000
But it also reminds us that AI is still just a tool.

301
00:09:12,000 --> 00:09:15,360
Like any tool, it can be used for good or bad.

302
00:09:15,360 --> 00:09:16,440
Yeah, that's true.

303
00:09:16,440 --> 00:09:20,080
So it's up to us to make sure we use it responsibly and ethically.

304
00:09:20,080 --> 00:09:20,840
For sure.

305
00:09:20,840 --> 00:09:21,400
For sure.

306
00:09:21,400 --> 00:09:24,680
But OK, let's go back to that 2024 prediction for a second.

307
00:09:24,680 --> 00:09:25,320
OK.

308
00:09:25,320 --> 00:09:27,440
I'm still curious about what the AI found,

309
00:09:27,440 --> 00:09:30,840
especially those potential shifts in those key states.

310
00:09:30,840 --> 00:09:31,120
Right.

311
00:09:31,120 --> 00:09:34,240
So just to reiterate, this was all based on data

312
00:09:34,240 --> 00:09:36,000
from November 2023.

313
00:09:36,000 --> 00:09:36,480
Right.

314
00:09:36,480 --> 00:09:39,200
And we all know how fast things can change in politics.

315
00:09:39,200 --> 00:09:40,240
Oh, yeah, for sure.

316
00:09:40,240 --> 00:09:44,000
But even as a snapshot in time, it's pretty interesting.

317
00:09:44,000 --> 00:09:46,560
Yeah, it's like a glimpse into one possible future.

318
00:09:46,560 --> 00:09:48,360
Exactly, based on what they knew at the time.

319
00:09:48,360 --> 00:09:50,200
So about those swing states you were talking about.

320
00:09:50,200 --> 00:09:52,160
Yeah, remember how the AI seemed to be really good

321
00:09:52,160 --> 00:09:53,040
at predicting those?

322
00:09:53,040 --> 00:09:53,320
Yeah.

323
00:09:53,320 --> 00:09:55,120
Well, Wisconsin was a perfect example.

324
00:09:55,120 --> 00:09:55,480
OK.

325
00:09:55,480 --> 00:09:57,720
It predicted a Trump win there in 2024.

326
00:09:57,720 --> 00:10:00,880
Wow, that would be a huge flip from 2020.

327
00:10:00,880 --> 00:10:02,480
Yeah, it would be pretty significant.

328
00:10:02,480 --> 00:10:04,240
Do we know why it predicted that?

329
00:10:04,240 --> 00:10:06,920
Like, was it demographics or something else?

330
00:10:06,920 --> 00:10:08,600
Unfortunately, the paper doesn't really

331
00:10:08,600 --> 00:10:10,560
give us all the details for each state.

332
00:10:10,560 --> 00:10:11,160
Oh, OK.

333
00:10:11,160 --> 00:10:13,480
But remember, they were looking at so many factors.

334
00:10:13,480 --> 00:10:17,240
Policy positions, demographics, the whole political climate

335
00:10:17,240 --> 00:10:18,640
is probably a mix of everything.

336
00:10:18,640 --> 00:10:19,840
Yeah, that makes sense.

337
00:10:19,840 --> 00:10:22,600
Oh, and don't forget, the AI was working with the assumption

338
00:10:22,600 --> 00:10:24,560
that Kamala Harris would be the nominee.

339
00:10:24,560 --> 00:10:25,040
Right.

340
00:10:25,040 --> 00:10:27,200
So you can see how these predictions can change,

341
00:10:27,200 --> 00:10:29,360
depending on even the smallest things.

342
00:10:29,360 --> 00:10:31,920
Yeah, it's a pretty fluid situation.

343
00:10:31,920 --> 00:10:33,400
New info could come out any day.

344
00:10:33,400 --> 00:10:35,120
Exactly.

345
00:10:35,120 --> 00:10:36,640
So what does all this mean for us,

346
00:10:36,640 --> 00:10:38,600
like, as we're going into this election?

347
00:10:38,600 --> 00:10:40,760
Yeah, should we believe these AI predictions

348
00:10:40,760 --> 00:10:41,840
or just ignore them?

349
00:10:41,840 --> 00:10:42,440
Right.

350
00:10:42,440 --> 00:10:44,280
I think the big takeaway here is

351
00:10:44,280 --> 00:10:47,720
that AI is becoming a really powerful tool in politics.

352
00:10:47,720 --> 00:10:48,560
Yeah.

353
00:10:48,560 --> 00:10:51,720
But it's not about replacing us, you know?

354
00:10:51,720 --> 00:10:52,320
Our judgment.

355
00:10:52,320 --> 00:10:54,600
Or ignoring what's actually happening in the world.

356
00:10:54,600 --> 00:10:57,320
So it's more about adding another layer of understanding.

357
00:10:57,320 --> 00:11:00,080
Exactly, like, seeing things from a different angle.

358
00:11:00,080 --> 00:11:01,280
This research is really cool.

359
00:11:01,280 --> 00:11:05,440
It shows how AI can help us understand human behavior.

360
00:11:05,440 --> 00:11:07,920
And that's something worth paying attention to, for sure.

361
00:11:07,920 --> 00:11:10,160
Well, thanks for taking us on this deep dive.

362
00:11:10,160 --> 00:11:11,360
I feel like I learned a lot.

363
00:11:11,360 --> 00:11:12,400
It was my pleasure.

364
00:11:12,400 --> 00:11:14,120
And to all our listeners out there.

365
00:11:14,120 --> 00:11:14,600
Yes.

366
00:11:14,600 --> 00:11:16,520
As we go into this election year,

367
00:11:16,520 --> 00:11:20,200
remember, AI can help us understand the trends and patterns.

368
00:11:20,200 --> 00:11:22,280
But it's up to each of us to stay informed

369
00:11:22,280 --> 00:11:23,560
and make our own decisions.

370
00:11:23,560 --> 00:11:24,560
Exactly.

371
00:11:24,560 --> 00:11:25,240
Well said.

372
00:11:25,240 --> 00:11:26,440
Thanks again for joining us.

373
00:11:26,440 --> 00:11:31,440
Thank you here.