1
00:00:00,000 --> 00:00:01,720
All right, strap in everyone.

2
00:00:01,720 --> 00:00:05,120
We are going deep on the EU's AI Act today.

3
00:00:05,120 --> 00:00:06,520
Excited to break this down.

4
00:00:06,520 --> 00:00:08,880
You really went all out with the prep materials, huh?

5
00:00:08,880 --> 00:00:11,400
Articles, research, forum discussions.

6
00:00:11,400 --> 00:00:13,280
Well, yeah, it's important.

7
00:00:13,280 --> 00:00:14,080
Absolutely.

8
00:00:14,080 --> 00:00:15,280
It's a beast of a document.

9
00:00:15,280 --> 00:00:19,520
113 articles, all the annexes, the whole shebang.

10
00:00:19,520 --> 00:00:20,920
Definitely not a light read.

11
00:00:20,920 --> 00:00:23,000
But this is about way more than just rules.

12
00:00:23,000 --> 00:00:27,480
It's like the EU's trying to be the world leader

13
00:00:27,480 --> 00:00:29,160
in responsible AI.

14
00:00:29,160 --> 00:00:30,240
That's a great way to put it.

15
00:00:30,240 --> 00:00:32,240
It's not just about limiting the bad stuff.

16
00:00:32,240 --> 00:00:34,880
They want to shape the whole development landscape.

17
00:00:34,880 --> 00:00:37,320
OK, so how are they actually doing that?

18
00:00:37,320 --> 00:00:41,320
This trustworthy AI idea seems to be the core, right?

19
00:00:41,320 --> 00:00:42,880
What does that even mean, practically?

20
00:00:42,880 --> 00:00:45,920
Well, imagine every AI system before it can even be sold

21
00:00:45,920 --> 00:00:47,720
has to go through rigorous testing,

22
00:00:47,720 --> 00:00:50,640
like prove it meets these really strict safety standards.

23
00:00:50,640 --> 00:00:51,680
OK, so that makes sense.

24
00:00:51,680 --> 00:00:54,080
Kind of like new medications or airplanes, right?

25
00:00:54,080 --> 00:00:56,280
They get tested like crazy before anyone uses them.

26
00:00:56,280 --> 00:00:56,920
Exactly.

27
00:00:56,920 --> 00:00:58,120
Makes sense, right?

28
00:00:58,120 --> 00:01:00,400
If AI is going to be everywhere, from health care

29
00:01:00,400 --> 00:01:01,160
to transportation.

30
00:01:01,160 --> 00:01:01,600
Oh, yeah.

31
00:01:01,600 --> 00:01:02,480
We got to know what's safe.

32
00:01:02,480 --> 00:01:02,840
Absolutely.

33
00:01:02,840 --> 00:01:06,600
But AI is so different, depending on what it's doing.

34
00:01:06,600 --> 00:01:08,680
They can't be treating a spam filter

35
00:01:08,680 --> 00:01:12,640
with the same seriousness as a surgical robot.

36
00:01:12,640 --> 00:01:13,080
Yeah, right.

37
00:01:13,080 --> 00:01:13,760
They aren't.

38
00:01:13,760 --> 00:01:16,880
And that's where their risk-based approach comes in.

39
00:01:16,880 --> 00:01:20,680
They've got this pyramid of risk, basically.

40
00:01:20,680 --> 00:01:21,360
I've seen that.

41
00:01:21,360 --> 00:01:25,440
At the very bottom, the stuff with almost no risk,

42
00:01:25,440 --> 00:01:28,000
your spam filters, those recommendation

43
00:01:28,000 --> 00:01:30,520
algorithms, things that haven't caused a global meltdown yet.

44
00:01:30,520 --> 00:01:31,360
Exactly.

45
00:01:31,360 --> 00:01:33,040
Basically unregulated.

46
00:01:33,040 --> 00:01:33,840
Then you move up.

47
00:01:33,840 --> 00:01:34,640
OK, what's next?

48
00:01:34,640 --> 00:01:35,920
Limited risk stuff.

49
00:01:35,920 --> 00:01:40,040
So your chatbots, deepfakes.transparency, is key there.

50
00:01:40,040 --> 00:01:42,760
You have to know if you're talking to AI, not a person.

51
00:01:42,760 --> 00:01:45,520
And any AI-generated stuff got to be labeled, informed,

52
00:01:45,520 --> 00:01:46,440
consent, all that.

53
00:01:46,440 --> 00:01:46,960
Right.

54
00:01:46,960 --> 00:01:49,240
Then we get to the high-risk zone.

55
00:01:49,240 --> 00:01:50,600
This is where it gets interesting.

56
00:01:50,600 --> 00:01:51,200
Lay it on me.

57
00:01:51,200 --> 00:01:54,760
So AI in critical infrastructure, law enforcement, health care.

58
00:01:54,760 --> 00:01:55,640
Whoa, OK, yeah.

59
00:01:55,640 --> 00:01:56,280
High stakes.

60
00:01:56,280 --> 00:01:59,280
Even job recruitment, credit scoring, anything

61
00:01:59,280 --> 00:02:01,200
that can really impact people's lives.

62
00:02:01,200 --> 00:02:03,160
So for that stuff, the EU saying?

63
00:02:03,160 --> 00:02:04,120
Strict.

64
00:02:04,120 --> 00:02:04,600
Requirements?

65
00:02:04,600 --> 00:02:06,280
What kind of requirements are we talking about?

66
00:02:06,280 --> 00:02:09,000
Rigorous testing.validation.

67
00:02:09,000 --> 00:02:12,800
The data used has to be top-notch human oversight

68
00:02:12,800 --> 00:02:14,040
to prevent bias.

69
00:02:14,040 --> 00:02:17,680
So no more move fast and break things with AI, huh?

70
00:02:17,680 --> 00:02:19,160
Not for this stuff.

71
00:02:19,160 --> 00:02:21,440
And then at the very top of this pyramid,

72
00:02:21,440 --> 00:02:22,160
there's more.

73
00:02:22,160 --> 00:02:25,520
Unacceptable risk, banned, outright.

74
00:02:25,520 --> 00:02:26,920
Whoa, what falls under that?

75
00:02:26,920 --> 00:02:31,200
Things like social scoring systems, ranking citizens

76
00:02:31,200 --> 00:02:32,200
based on behavior.

77
00:02:32,200 --> 00:02:32,720
Oh, wow.

78
00:02:32,720 --> 00:02:36,440
Real-time biometric ID in public for mass surveillance,

79
00:02:36,440 --> 00:02:39,320
anything that exploits vulnerable groups like kids

80
00:02:39,320 --> 00:02:40,960
or people with disabilities.

81
00:02:40,960 --> 00:02:42,640
Hardline, totally banned.

82
00:02:42,640 --> 00:02:43,160
OK, good.

83
00:02:43,160 --> 00:02:43,920
That's reassuring.

84
00:02:43,920 --> 00:02:46,680
But you mentioned earlier, systems like ChatGPT.

85
00:02:46,680 --> 00:02:47,680
Where do those fit in?

86
00:02:47,680 --> 00:02:48,960
That is the question, isn't it?

87
00:02:48,960 --> 00:02:50,040
General-purpose AI.

88
00:02:50,040 --> 00:02:51,760
It can be used for anything.

89
00:02:51,760 --> 00:02:52,320
Exactly.

90
00:02:52,320 --> 00:02:54,120
So the risk depends entirely on how it's used.

91
00:02:54,120 --> 00:02:54,520
Right.

92
00:02:54,520 --> 00:02:56,400
It's like that one minute it's writing code,

93
00:02:56,400 --> 00:02:58,280
the next it's generating fake news.

94
00:02:58,280 --> 00:03:00,320
So how do you even regulate that?

95
00:03:00,320 --> 00:03:01,200
It's a real puzzle.

96
00:03:01,200 --> 00:03:03,960
The EU knows these systems have huge EE potential,

97
00:03:03,960 --> 00:03:05,040
good and bad.

98
00:03:05,040 --> 00:03:06,400
So they're walking a tightrope.

99
00:03:06,400 --> 00:03:07,320
Trying to.

100
00:03:07,320 --> 00:03:11,960
They're still under the AI Act, but the new European AI

101
00:03:11,960 --> 00:03:15,400
office, they'll be keeping a close eye on this stuff.

102
00:03:15,400 --> 00:03:17,880
This AI office, it sounds like a pretty big deal.

103
00:03:17,880 --> 00:03:19,120
What's their role, exactly?

104
00:03:19,120 --> 00:03:23,760
They're like the AI watchdogs for all of Europe, centralized.

105
00:03:23,760 --> 00:03:25,280
And they've got power.

106
00:03:25,280 --> 00:03:26,560
Power to do what?

107
00:03:26,560 --> 00:03:30,480
Request info from companies, run their own evaluations

108
00:03:30,480 --> 00:03:32,480
of AI systems.

109
00:03:32,480 --> 00:03:34,480
If companies don't follow the rules.

110
00:03:34,480 --> 00:03:35,720
They can get penalized.

111
00:03:35,720 --> 00:03:37,800
Big fines, potentially.

112
00:03:37,800 --> 00:03:39,440
The EU's making a statement here.

113
00:03:39,440 --> 00:03:41,280
Yeah, messing around with AI development in Europe

114
00:03:41,280 --> 00:03:42,240
is not going to fly.

115
00:03:42,240 --> 00:03:42,880
Not anymore.

116
00:03:42,880 --> 00:03:45,840
OK, so we've got the core ideas, the risk pyramid,

117
00:03:45,840 --> 00:03:47,760
the things that are straight up banned,

118
00:03:47,760 --> 00:03:49,680
and this new AI office.

119
00:03:49,680 --> 00:03:51,560
It's clear the EU is all in on this.

120
00:03:51,560 --> 00:03:53,080
This isn't just a dip in the pool.

121
00:03:53,080 --> 00:03:56,040
They're diving headfirst into regulating AI.

122
00:03:56,040 --> 00:03:57,520
Absolutely, it's really impressive.

123
00:03:57,520 --> 00:03:58,520
The ripple effects of this.

124
00:03:58,520 --> 00:04:00,560
I don't think they're going to stay within Europe's borders.

125
00:04:00,560 --> 00:04:01,240
Definitely not.

126
00:04:01,240 --> 00:04:02,760
OK, so we've got the big picture,

127
00:04:02,760 --> 00:04:04,320
but let's get a little more specific.

128
00:04:04,320 --> 00:04:05,280
What are some of the ways they're

129
00:04:05,280 --> 00:04:07,760
planning to make this work practically?

130
00:04:07,760 --> 00:04:11,200
I've seen these regulatory sandboxes mentioned.

131
00:04:11,200 --> 00:04:13,040
Yes, the sandboxes.

132
00:04:13,040 --> 00:04:14,200
What are those all about?

133
00:04:14,200 --> 00:04:16,760
So they're basically these controlled environments

134
00:04:16,760 --> 00:04:21,360
where companies can test out their AI systems,

135
00:04:21,360 --> 00:04:23,560
but the regulators are watching.

136
00:04:23,560 --> 00:04:24,320
Interesting.

137
00:04:24,320 --> 00:04:26,480
Especially for potentially high risk stuff,

138
00:04:26,480 --> 00:04:30,120
like a beta test, but with the EU looking over your shoulder.

139
00:04:30,120 --> 00:04:33,640
So companies can get feedback, make changes

140
00:04:33,640 --> 00:04:35,600
before they break the rules without knowing.

141
00:04:35,600 --> 00:04:38,040
Exactly, and it lets the EU see what's

142
00:04:38,040 --> 00:04:40,520
being developed, spot issues early on.

143
00:04:40,520 --> 00:04:42,520
They're especially focused on using these

144
00:04:42,520 --> 00:04:43,880
to help smaller startups.

145
00:04:43,880 --> 00:04:44,560
Oh, that makes sense.

146
00:04:44,560 --> 00:04:46,160
Startups maybe don't have the resources

147
00:04:46,160 --> 00:04:47,840
to deal with all this complexity on their own.

148
00:04:47,840 --> 00:04:49,960
Exactly, leveling the playing field a bit.

149
00:04:49,960 --> 00:04:52,840
So they're trying to make sure innovation doesn't get choked

150
00:04:52,840 --> 00:04:53,920
out by all these rules.

151
00:04:53,920 --> 00:04:54,720
That's the goal.

152
00:04:54,720 --> 00:04:56,640
It's about collaboration, really.

153
00:04:56,640 --> 00:04:59,200
Startups, big companies, regulators,

154
00:04:59,200 --> 00:05:02,240
all working together to make AI happen responsibly.

155
00:05:02,240 --> 00:05:03,920
OK, I like that idea.

156
00:05:03,920 --> 00:05:05,680
But let's be realistic.

157
00:05:05,680 --> 00:05:08,520
The AI Act, it's going to upset some people.

158
00:05:08,520 --> 00:05:11,360
What happens when someone doesn't play by the rules?

159
00:05:11,360 --> 00:05:12,600
What are the consequences?

160
00:05:12,600 --> 00:05:15,760
Well, the European AI Office has real teeth.

161
00:05:15,760 --> 00:05:17,480
It's not just for show.

162
00:05:17,480 --> 00:05:21,680
They can issue warnings, demand changes to an AI system,

163
00:05:21,680 --> 00:05:25,760
and for serious violations, the fines can be huge.

164
00:05:25,760 --> 00:05:26,400
Wow.

165
00:05:26,400 --> 00:05:30,400
And in the worst cases, they can just ban a system outright.

166
00:05:30,400 --> 00:05:31,720
No more sales in Europe.

167
00:05:31,720 --> 00:05:32,920
So they're not messing around.

168
00:05:32,920 --> 00:05:33,560
Nope.

169
00:05:33,560 --> 00:05:36,600
And remember, the AI Office is keeping tabs on everyone,

170
00:05:36,600 --> 00:05:39,440
but technology, it moves fast.

171
00:05:39,440 --> 00:05:40,000
Oh, yeah.

172
00:05:40,000 --> 00:05:42,720
How are they going to keep this AI Act from being obsolete

173
00:05:42,720 --> 00:05:44,440
in a couple years?

174
00:05:44,440 --> 00:05:47,320
The AI world could look totally different by then.

175
00:05:47,320 --> 00:05:48,560
That's a good question.

176
00:05:48,560 --> 00:05:49,800
And the EU's thought about it.

177
00:05:49,800 --> 00:05:53,560
The Acts got these mechanisms built in to keep it adaptable.

178
00:05:53,560 --> 00:05:55,800
Oh, tell me more.

179
00:05:55,800 --> 00:05:58,440
They're going to review and update the list of high-risk systems

180
00:05:58,440 --> 00:06:00,640
regularly, so anything new that pops up.

181
00:06:00,640 --> 00:06:01,840
Gets added to the list.

182
00:06:01,840 --> 00:06:03,280
Exactly.

183
00:06:03,280 --> 00:06:05,840
They're also relying on industry standards and codes

184
00:06:05,840 --> 00:06:08,720
of practice, which can change faster than the law itself.

185
00:06:08,720 --> 00:06:10,520
So they're designing it to be upgradable.

186
00:06:10,520 --> 00:06:11,240
Exactly.

187
00:06:11,240 --> 00:06:14,200
And get this, they've even given the European Commission

188
00:06:14,200 --> 00:06:16,520
the power to make small tweaks through these things

189
00:06:16,520 --> 00:06:17,920
called delegated acts.

190
00:06:17,920 --> 00:06:18,760
Delegated acts.

191
00:06:18,760 --> 00:06:20,640
Yeah, without having to go through the whole lawmaking

192
00:06:20,640 --> 00:06:21,320
process again.

193
00:06:21,320 --> 00:06:22,640
So they can update it quicker.

194
00:06:22,640 --> 00:06:23,800
Yep, keep it current.

195
00:06:23,800 --> 00:06:27,080
OK, so they're building in flexibility.

196
00:06:27,080 --> 00:06:29,920
But what about the human element?

197
00:06:29,920 --> 00:06:33,720
We've been talking tech, but this AI Act,

198
00:06:33,720 --> 00:06:36,560
it depends on people getting it, accepting it right.

199
00:06:36,560 --> 00:06:37,480
You're absolutely right.

200
00:06:37,480 --> 00:06:39,520
Technology is only part of the story.

201
00:06:39,520 --> 00:06:41,320
The EU knows that, which is why they're

202
00:06:41,320 --> 00:06:42,960
emphasizing AI literacy.

203
00:06:42,960 --> 00:06:44,440
Call on AI literacy.

204
00:06:44,440 --> 00:06:47,640
Making sure people understand at a basic level

205
00:06:47,640 --> 00:06:50,800
how AI works, the good and the bad, the ethical stuff.

206
00:06:50,800 --> 00:06:51,560
But that's interesting.

207
00:06:51,560 --> 00:06:53,640
The Act even says companies have to train

208
00:06:53,640 --> 00:06:55,120
their employees on all this.

209
00:06:55,120 --> 00:06:55,840
Really?

210
00:06:55,840 --> 00:06:56,240
Yep.

211
00:06:56,240 --> 00:06:57,720
So it's not just regulating the tech,

212
00:06:57,720 --> 00:06:59,520
it's educating the people using it.

213
00:06:59,520 --> 00:07:03,640
Smart, because AI, it's already changing our world,

214
00:07:03,640 --> 00:07:05,240
and it's only going to get more intense.

215
00:07:05,240 --> 00:07:06,120
It is.

216
00:07:06,120 --> 00:07:08,240
And the more people understand, the better.

217
00:07:08,240 --> 00:07:09,880
It really is going to become more intense.

218
00:07:09,880 --> 00:07:11,720
And yeah, the more people get it,

219
00:07:11,720 --> 00:07:13,120
the better prepared we'll all be.

220
00:07:13,120 --> 00:07:15,080
OK, I'm getting this image of the EU

221
00:07:15,080 --> 00:07:19,880
trying to build this whole ecosystem of trust around AI.

222
00:07:19,880 --> 00:07:22,840
But sometimes it feels like it's only the big guys.

223
00:07:22,840 --> 00:07:24,160
You know, Google, Microsoft?

224
00:07:24,160 --> 00:07:26,200
The big tech companies.

225
00:07:26,200 --> 00:07:29,840
Does the AI Act do anything for the smaller players?

226
00:07:29,840 --> 00:07:30,760
The startups.

227
00:07:30,760 --> 00:07:32,080
That's a great point.

228
00:07:32,080 --> 00:07:34,320
And a lot of something that EU has been thinking about

229
00:07:34,320 --> 00:07:36,560
carefully, a lot of people actually

230
00:07:36,560 --> 00:07:40,800
think the AI Act could help smaller companies in the long run.

231
00:07:40,800 --> 00:07:41,360
Really?

232
00:07:41,360 --> 00:07:42,120
How so?

233
00:07:42,120 --> 00:07:42,840
Well, think about it.

234
00:07:42,840 --> 00:07:45,120
One of the hardest things for a startup

235
00:07:45,120 --> 00:07:48,840
is just getting people to trust them, customers, investors.

236
00:07:48,840 --> 00:07:49,960
They need to prove themselves.

237
00:07:49,960 --> 00:07:51,080
Exactly.

238
00:07:51,080 --> 00:07:55,760
And by following these tough AI Act rules, smaller companies,

239
00:07:55,760 --> 00:07:57,880
they can kind of show off their commitment

240
00:07:57,880 --> 00:08:01,560
to doing AI the right way, ethically, responsibly.

241
00:08:01,560 --> 00:08:02,640
It's like a badge of honor.

242
00:08:02,640 --> 00:08:03,720
We're doing this right.

243
00:08:03,720 --> 00:08:04,440
Yeah.

244
00:08:04,440 --> 00:08:05,640
And that can be a big deal when you're

245
00:08:05,640 --> 00:08:08,240
competing with those huge tech companies with all their money.

246
00:08:08,240 --> 00:08:10,360
So the AI Act, it's not just a burden.

247
00:08:10,360 --> 00:08:12,800
It could actually be a way for startups to stand out.

248
00:08:12,800 --> 00:08:13,680
Absolutely.

249
00:08:13,680 --> 00:08:15,840
And don't forget those regulatory sandboxes

250
00:08:15,840 --> 00:08:17,000
we were talking about.

251
00:08:17,000 --> 00:08:20,640
Those are there to specifically help smaller companies get

252
00:08:20,640 --> 00:08:22,920
their AI systems tested and refined.

253
00:08:22,920 --> 00:08:23,800
Oh, right.

254
00:08:23,800 --> 00:08:25,920
With the regulators help, it's like a safety net.

255
00:08:25,920 --> 00:08:26,480
Exactly.

256
00:08:26,480 --> 00:08:29,560
It levels the playing field so everyone has a fair shot.

257
00:08:29,560 --> 00:08:31,720
OK, this is making a lot more sense.

258
00:08:31,720 --> 00:08:33,080
But let's be practical for a minute.

259
00:08:33,080 --> 00:08:37,840
If a company screws up, violates the AI Act, what happens?

260
00:08:37,840 --> 00:08:41,560
Well, the EU has options, and they're not afraid to use them.

261
00:08:41,560 --> 00:08:42,200
Like what?

262
00:08:42,200 --> 00:08:44,080
It depends on how bad it is, right?

263
00:08:44,080 --> 00:08:46,280
For small stuff, maybe a warning,

264
00:08:46,280 --> 00:08:48,320
or they have to change something about the AI.

265
00:08:48,320 --> 00:08:50,920
But if it's serious, things get real.

266
00:08:50,920 --> 00:08:52,680
The fines can be huge.

267
00:08:52,680 --> 00:08:56,240
And in extreme cases, that AI system, it's banned.

268
00:08:56,240 --> 00:08:57,600
Can't sell it in Europe anymore.

269
00:08:57,600 --> 00:08:58,000
Wow.

270
00:08:58,000 --> 00:08:58,880
No more messing around.

271
00:08:58,880 --> 00:08:59,480
Nope.

272
00:08:59,480 --> 00:09:02,280
The European AI office is watching closely.

273
00:09:02,280 --> 00:09:03,240
They're making a point.

274
00:09:03,240 --> 00:09:04,560
They are.

275
00:09:04,560 --> 00:09:05,960
OK, so we've talked about benefit.

276
00:09:05,960 --> 00:09:09,120
The risks, if you break the rules, how it's all enforced.

277
00:09:09,120 --> 00:09:11,240
But here's the big question.

278
00:09:11,240 --> 00:09:15,680
How does this impact Europe in the global AI race?

279
00:09:15,680 --> 00:09:17,880
Ah, the million dollar question.

280
00:09:17,880 --> 00:09:21,320
Does regulation help, or does it hurt innovation?

281
00:09:21,320 --> 00:09:23,600
Some people are saying that these strict standards will

282
00:09:23,600 --> 00:09:25,880
put European companies at a disadvantage.

283
00:09:25,880 --> 00:09:27,240
What's the other side of that argument?

284
00:09:27,240 --> 00:09:30,240
Well, the EU is basically saying, look, responsible AI,

285
00:09:30,240 --> 00:09:32,880
that's the key to being competitive in the long term.

286
00:09:32,880 --> 00:09:34,200
They're playing the long game.

287
00:09:34,200 --> 00:09:35,240
Exactly.

288
00:09:35,240 --> 00:09:38,200
Build trust, make sure things are ethical.

289
00:09:38,200 --> 00:09:40,120
That's how you create a place where people want

290
00:09:40,120 --> 00:09:41,480
to invest and innovate.

291
00:09:41,480 --> 00:09:43,160
So it's about sustainability.

292
00:09:43,160 --> 00:09:43,520
Yeah.

293
00:09:43,520 --> 00:09:44,960
Not just chasing quick wins.

294
00:09:44,960 --> 00:09:45,480
Right.

295
00:09:45,480 --> 00:09:47,600
The EU is saying, we're not cutting corners.

296
00:09:47,600 --> 00:09:48,560
We're doing this right.

297
00:09:48,560 --> 00:09:51,520
And in the end, it's going to create a stronger AI landscape

298
00:09:51,520 --> 00:09:53,120
for everyone in Europe.

299
00:09:53,120 --> 00:09:55,160
OK, I can see that logic.

300
00:09:55,160 --> 00:09:57,720
But AI, it's global, right?

301
00:09:57,720 --> 00:09:58,880
It's happening everywhere.

302
00:09:58,880 --> 00:09:59,480
It is.

303
00:09:59,480 --> 00:10:01,800
What if other countries don't follow suit?

304
00:10:01,800 --> 00:10:03,720
Is Europe just setting itself up to fail

305
00:10:03,720 --> 00:10:05,280
by going it alone on this?

306
00:10:05,280 --> 00:10:06,000
It's true.

307
00:10:06,000 --> 00:10:08,320
The EU can't control the whole world.

308
00:10:08,320 --> 00:10:11,200
But they're hoping that by doing this right,

309
00:10:11,200 --> 00:10:13,200
others will want to follow their lead.

310
00:10:13,200 --> 00:10:14,600
So they're setting the example.

311
00:10:14,600 --> 00:10:15,880
Like, hey, this is how you do it.

312
00:10:15,880 --> 00:10:17,520
Yeah, it's like a ripple effect.

313
00:10:17,520 --> 00:10:20,920
If the EU can show that responsible AI is possible

314
00:10:20,920 --> 00:10:23,120
and it worked well, other countries

315
00:10:23,120 --> 00:10:24,720
might say, hey, we should do that too.

316
00:10:24,720 --> 00:10:26,920
And that would be good for everyone, not just Europe.

317
00:10:26,920 --> 00:10:27,520
I like that.

318
00:10:27,520 --> 00:10:29,240
Leading by example.

319
00:10:29,240 --> 00:10:31,880
But let's face it, AI, it's a moving target.

320
00:10:31,880 --> 00:10:33,240
Constantly changing.

321
00:10:33,240 --> 00:10:36,000
New stuff popping up all the time that we can't even

322
00:10:36,000 --> 00:10:36,920
predict right now.

323
00:10:36,920 --> 00:10:39,720
How can the EU possibly make rules for something

324
00:10:39,720 --> 00:10:41,160
that's evolving so quickly?

325
00:10:41,160 --> 00:10:43,720
That's the tough part about regulating any new technology,

326
00:10:43,720 --> 00:10:44,360
right?

327
00:10:44,360 --> 00:10:46,640
But remember those delegated acts we talked about?

328
00:10:46,640 --> 00:10:48,080
Oh, right, the FastTrack updates.

329
00:10:48,080 --> 00:10:50,240
That's one way they can adapt to new stuff

330
00:10:50,240 --> 00:10:52,240
without having to rewrite the whole law.

331
00:10:52,240 --> 00:10:53,920
And they've given the European Commission

332
00:10:53,920 --> 00:10:58,680
the job of keeping that list of high-risk AI systems updated

333
00:10:58,680 --> 00:10:59,240
too.

334
00:10:59,240 --> 00:11:00,680
So as new tech comes up.

335
00:11:00,680 --> 00:11:01,680
Boom, added to the list.

336
00:11:01,680 --> 00:11:02,200
Exactly.

337
00:11:02,200 --> 00:11:05,040
It's about having guidelines, but also being

338
00:11:05,040 --> 00:11:07,120
able to adapt as things change.

339
00:11:07,120 --> 00:11:08,040
Find that balance.

340
00:11:08,040 --> 00:11:08,520
Right.

341
00:11:08,520 --> 00:11:10,240
The goal isn't to stop innovation.

342
00:11:10,240 --> 00:11:13,560
It's to give it a framework so it can happen in a good way.

343
00:11:13,560 --> 00:11:16,280
OK, I'm really starting to see the whole picture here.

344
00:11:16,280 --> 00:11:18,440
But I keep thinking about this AI literacy thing.

345
00:11:18,440 --> 00:11:19,720
It seems really important.

346
00:11:19,720 --> 00:11:21,320
But how are they actually going to do that?

347
00:11:21,320 --> 00:11:24,280
Are we talking like AI classes in schools?

348
00:11:24,280 --> 00:11:25,320
I don't know about that.

349
00:11:25,320 --> 00:11:27,000
AI influencers on TikTok?

350
00:11:27,000 --> 00:11:28,160
Maybe.

351
00:11:28,160 --> 00:11:31,560
But they are definitely taking a multi-pronged approach.

352
00:11:31,560 --> 00:11:32,640
OK, what does that mean?

353
00:11:32,640 --> 00:11:34,880
Well, like we said, companies have to train their employees.

354
00:11:34,880 --> 00:11:36,000
That's a start.

355
00:11:36,000 --> 00:11:39,080
But they're also putting money into bigger educational

356
00:11:39,080 --> 00:11:40,960
initiatives for the general public.

357
00:11:40,960 --> 00:11:42,040
What kind of initiatives?

358
00:11:42,040 --> 00:11:45,560
Think public awareness campaigns, online resources,

359
00:11:45,560 --> 00:11:48,320
educational programs in schools and universities,

360
00:11:48,320 --> 00:11:51,200
funding research into the best ways to actually teach people

361
00:11:51,200 --> 00:11:52,280
about AI.

362
00:11:52,280 --> 00:11:55,280
So it's not just the technical stuff.

363
00:11:55,280 --> 00:11:57,040
It's about helping people understand

364
00:11:57,040 --> 00:12:01,480
how AI is going to impact society, the economy, our lives.

365
00:12:01,480 --> 00:12:02,400
Exactly.

366
00:12:02,400 --> 00:12:05,680
The EU wants citizens to be active participants

367
00:12:05,680 --> 00:12:09,320
in this AI revolution, not just sitting on the sidelines

368
00:12:09,320 --> 00:12:10,320
watching it happen.

369
00:12:10,320 --> 00:12:11,480
OK, I'm starting to get it.

370
00:12:11,480 --> 00:12:14,120
The EU's thinking about the tech, the rules, the economy,

371
00:12:14,120 --> 00:12:17,320
and how all this affects people all at the same time.

372
00:12:17,320 --> 00:12:19,160
This is a huge EU undertaking.

373
00:12:19,160 --> 00:12:20,040
It is.

374
00:12:20,040 --> 00:12:23,120
But for the EU, this is a defining moment for AI.

375
00:12:23,120 --> 00:12:25,240
They want to get this right, not just for Europe,

376
00:12:25,240 --> 00:12:26,560
but for the whole world.

377
00:12:26,560 --> 00:12:27,880
That's a build statement.

378
00:12:27,880 --> 00:12:29,640
But honestly, I'm starting to feel

379
00:12:29,640 --> 00:12:31,360
optimistic, cautiously optimistic.

380
00:12:31,360 --> 00:12:32,160
I think we can be.

381
00:12:32,160 --> 00:12:33,560
If they can actually pull this off,

382
00:12:33,560 --> 00:12:35,960
the whole AI industry could change for the better.

383
00:12:35,960 --> 00:12:37,040
I think so too.

384
00:12:37,040 --> 00:12:39,160
And it's something we all need to pay attention to.

385
00:12:39,160 --> 00:12:39,920
Absolutely.

386
00:12:39,920 --> 00:12:41,480
We've covered so much ground here.

387
00:12:41,480 --> 00:12:44,680
Smaller companies, how the rules were enforced,

388
00:12:44,680 --> 00:12:47,760
global impact, AI literacy.

389
00:12:47,760 --> 00:12:49,560
It's amazing how the EU has put together

390
00:12:49,560 --> 00:12:53,480
this vision for a responsible AI future.

391
00:12:53,480 --> 00:12:54,560
But you know what?

392
00:12:54,560 --> 00:12:56,520
We've been kind of dancing around the specifics

393
00:12:56,520 --> 00:12:58,280
of those high risk AI systems.

394
00:12:58,280 --> 00:13:00,600
What companies actually need to do to follow the rules.

395
00:13:00,600 --> 00:13:01,280
You're right.

396
00:13:01,280 --> 00:13:02,680
Time to unpack that, don't you think?

397
00:13:02,680 --> 00:13:03,160
Let's do it.

398
00:13:03,160 --> 00:13:04,880
Let's get into the nitty gritty and see

399
00:13:04,880 --> 00:13:09,040
how the EU plans to make trustworthy AI a reality.

400
00:13:09,040 --> 00:13:09,240
Right.

401
00:13:09,240 --> 00:13:10,880
Let's get specific.

402
00:13:10,880 --> 00:13:13,360
High risk AI systems.

403
00:13:13,360 --> 00:13:15,040
What does a company actually have

404
00:13:15,040 --> 00:13:17,960
to do to follow this AI act?

405
00:13:17,960 --> 00:13:20,920
Well, the act lays out some pretty specific requirements.

406
00:13:20,920 --> 00:13:21,240
Yeah.

407
00:13:21,240 --> 00:13:23,280
And it's not just the company's building the AI.

408
00:13:23,280 --> 00:13:24,280
The company's using it.

409
00:13:24,280 --> 00:13:25,680
They have to follow the rules too.

410
00:13:25,680 --> 00:13:26,040
Right.

411
00:13:26,040 --> 00:13:29,000
Because even a safe AI can be used in risky ways, I guess.

412
00:13:29,000 --> 00:13:29,680
Exactly.

413
00:13:29,680 --> 00:13:32,480
So the act talks about providers and employers.

414
00:13:32,480 --> 00:13:35,240
Providers are the ones making and selling the AI.

415
00:13:35,240 --> 00:13:37,680
And employers are the ones actually using it.

416
00:13:37,680 --> 00:13:38,200
OK.

417
00:13:38,200 --> 00:13:42,400
So if, like, Google made a facial recognition algorithm,

418
00:13:42,400 --> 00:13:43,880
they'd be the provider.

419
00:13:43,880 --> 00:13:46,720
And a police department using it would be the deployer.

420
00:13:46,720 --> 00:13:47,520
Exactly.

421
00:13:47,520 --> 00:13:49,640
And for providers, the rules are tough.

422
00:13:49,640 --> 00:13:52,520
They have to prove their AI is using good data,

423
00:13:52,520 --> 00:13:55,480
that it's accurate, that it's built to handle risks and bias.

424
00:13:55,480 --> 00:13:56,960
So it can't just work.

425
00:13:56,960 --> 00:13:59,760
It has to be responsibly made for the beginning.

426
00:13:59,760 --> 00:14:00,600
That's the idea.

427
00:14:00,600 --> 00:14:03,520
The EU saying, if you're selling high-risk AI,

428
00:14:03,520 --> 00:14:06,040
you better be able to show us you did it right.

429
00:14:06,040 --> 00:14:06,280
OK.

430
00:14:06,280 --> 00:14:07,240
That makes sense.

431
00:14:07,240 --> 00:14:08,720
But how do they actually show that?

432
00:14:08,720 --> 00:14:09,840
What's the proof?

433
00:14:09,840 --> 00:14:11,600
This is where it gets interesting.

434
00:14:11,600 --> 00:14:15,280
The act uses something called harmonized standards.

435
00:14:15,280 --> 00:14:16,440
Harmonized standards.

436
00:14:16,440 --> 00:14:16,920
Yeah.

437
00:14:16,920 --> 00:14:19,200
They're like these super detailed specs.

438
00:14:19,200 --> 00:14:22,400
They lay out exactly how W to meet the requirements.

439
00:14:22,400 --> 00:14:25,400
So like if the AI act is the rule book,

440
00:14:25,400 --> 00:14:27,080
the standards are the blueprints.

441
00:14:27,080 --> 00:14:28,680
You follow those, you're good to go.

442
00:14:28,680 --> 00:14:29,720
Perfect analogy.

443
00:14:29,720 --> 00:14:32,160
And the standards are going to be made with industry experts,

444
00:14:32,160 --> 00:14:35,280
so they'll be tailored to the specific kind of AI.

445
00:14:35,280 --> 00:14:37,520
So providers, they've got to follow those standards

446
00:14:37,520 --> 00:14:38,520
to the letter.

447
00:14:38,520 --> 00:14:40,920
Or else the AI office is going to come knocking.

448
00:14:40,920 --> 00:14:41,880
You got it.

449
00:14:41,880 --> 00:14:44,000
But it's not just about avoiding punishment.

450
00:14:44,000 --> 00:14:45,520
There's a bigger incentive here.

451
00:14:45,520 --> 00:14:46,000
Oh, yeah.

452
00:14:46,000 --> 00:14:46,720
What's that?

453
00:14:46,720 --> 00:14:49,560
The EU is betting that following these rules

454
00:14:49,560 --> 00:14:52,840
will actually make companies more competitive.

455
00:14:52,840 --> 00:14:53,760
How so?

456
00:14:53,760 --> 00:14:56,920
Think about it, it's all about trustworthy AI, right?

457
00:14:56,920 --> 00:14:59,960
If you can say, hey, our AI isn't just effective,

458
00:14:59,960 --> 00:15:05,280
it's also safe, reliable, ethical, that's a big deal.

459
00:15:05,280 --> 00:15:06,160
It's a selling point.

460
00:15:06,160 --> 00:15:07,040
Exactly.

461
00:15:07,040 --> 00:15:09,720
Especially now when people are kind of nervous about what

462
00:15:09,720 --> 00:15:13,880
AI can do, that trust, it's valuable.

463
00:15:13,880 --> 00:15:15,160
It's like a seal of approval, right?

464
00:15:15,160 --> 00:15:17,160
Made in Europe, developed responsibly.

465
00:15:17,160 --> 00:15:19,280
Yeah, that's a great way to put it.

466
00:15:19,280 --> 00:15:21,400
OK, but let's talk about the deployers now.

467
00:15:21,400 --> 00:15:23,320
The company is actually using the AI.

468
00:15:23,320 --> 00:15:24,160
Right.

469
00:15:24,160 --> 00:15:25,720
What are their responsibilities?

470
00:15:25,720 --> 00:15:27,720
They have to make sure they're using the AI the right way,

471
00:15:27,720 --> 00:15:29,960
following the instructions that their staff are trained.

472
00:15:29,960 --> 00:15:30,560
Decent.

473
00:15:30,560 --> 00:15:33,240
And this is important, human oversight the whole time.

474
00:15:33,240 --> 00:15:34,240
Wait, hold on.

475
00:15:34,240 --> 00:15:35,160
Human oversight.

476
00:15:35,160 --> 00:15:37,280
The EU is saying humans have to be involved.

477
00:15:37,280 --> 00:15:38,400
Absolutely.

478
00:15:38,400 --> 00:15:41,480
One of the big ideas in the AI Act is that high-risk AI,

479
00:15:41,480 --> 00:15:43,800
it can never be totally on autopilot.

480
00:15:43,800 --> 00:15:46,800
OK, so it's about humans and AI working together.

481
00:15:46,800 --> 00:15:47,560
Yep.

482
00:15:47,560 --> 00:15:49,880
AI is a tool, not a replacement.

483
00:15:49,880 --> 00:15:53,640
The AI Act recognizes that AI, as powerful as it is,

484
00:15:53,640 --> 00:15:54,720
it can mess up.

485
00:15:54,720 --> 00:15:57,240
And sometimes you need a human to step in,

486
00:15:57,240 --> 00:16:00,160
to make sure things are safe, fair, you know?

487
00:16:00,160 --> 00:16:01,000
That makes sense.

488
00:16:01,000 --> 00:16:03,720
But what happens when something does go wrong?

489
00:16:03,720 --> 00:16:07,080
Like, the AI malfunctions or makes a bad decision.

490
00:16:07,080 --> 00:16:09,920
Well, that's why the Act is so big on risk management.

491
00:16:09,920 --> 00:16:12,160
Providers, A&D deployers, they have

492
00:16:12,160 --> 00:16:15,040
to have systems in place to spot and fix problems.

493
00:16:15,040 --> 00:16:16,920
From the start, all the way through.

494
00:16:16,920 --> 00:16:17,360
Right.

495
00:16:17,360 --> 00:16:19,280
It's not just about making safe AI.

496
00:16:19,280 --> 00:16:21,680
It's about always checking it, making sure it's doing

497
00:16:21,680 --> 00:16:22,680
what it should be doing.

498
00:16:22,680 --> 00:16:24,880
And if there's an incident, they have to investigate,

499
00:16:24,880 --> 00:16:27,320
report it to the EU, and make sure it doesn't happen again.

500
00:16:27,320 --> 00:16:28,560
Accountability is key.

501
00:16:28,560 --> 00:16:29,400
Absolutely.

502
00:16:29,400 --> 00:16:29,900
Yeah.

503
00:16:29,900 --> 00:16:31,280
The EU is serious about this.

504
00:16:31,280 --> 00:16:34,320
Developing and using AI, it comes with responsibility.

505
00:16:34,320 --> 00:16:37,840
OK, we're seeing how the EU wants AI to work in Europe.

506
00:16:37,840 --> 00:16:41,520
But the big question, all of these rules,

507
00:16:41,520 --> 00:16:43,640
will it stifle innovation?

508
00:16:43,640 --> 00:16:46,240
Will European companies be able to compete?

509
00:16:46,240 --> 00:16:47,320
It's a valid question.

510
00:16:47,320 --> 00:16:48,800
A lot of people are talking about this.

511
00:16:48,800 --> 00:16:53,760
But the EU is betting that trustworthy AI is the key.

512
00:16:53,760 --> 00:16:56,200
In the long run, it'll be worth it.

513
00:16:56,200 --> 00:16:58,480
They're saying, yeah, it might be tough now.

514
00:16:58,480 --> 00:17:01,960
But in the end, trust and reliability.

515
00:17:01,960 --> 00:17:03,840
That'll make European AI stronger.

516
00:17:03,840 --> 00:17:04,640
Exactly.

517
00:17:04,640 --> 00:17:07,280
They're saying the other way, just racing to build

518
00:17:07,280 --> 00:17:09,480
AI as fast as possible without thinking

519
00:17:09,480 --> 00:17:12,080
about the consequences, that's dangerous.

520
00:17:12,080 --> 00:17:12,920
It's short-sighted.

521
00:17:12,920 --> 00:17:13,480
Yeah.

522
00:17:13,480 --> 00:17:14,520
And it could backfire.

523
00:17:14,520 --> 00:17:16,240
This isn't just about being competitive.

524
00:17:16,240 --> 00:17:18,680
It's about making AI that benefits everyone.

525
00:17:18,680 --> 00:17:19,840
That's the big idea.

526
00:17:19,840 --> 00:17:22,080
The EU wants to set the standard,

527
00:17:22,080 --> 00:17:24,080
make AI that's a force for good.

528
00:17:24,080 --> 00:17:25,400
It's a big goal.

529
00:17:25,400 --> 00:17:29,200
And this deep dive, looking into the AI act,

530
00:17:29,200 --> 00:17:30,960
it's not just about understanding the rules.

531
00:17:30,960 --> 00:17:32,760
It's about understanding one of the biggest changes

532
00:17:32,760 --> 00:17:33,600
in human history.

533
00:17:33,600 --> 00:17:34,440
It really is.

534
00:17:34,440 --> 00:17:36,600
And what we do now, all of us, it's

535
00:17:36,600 --> 00:17:38,360
going to shape the future of AI.

536
00:17:38,360 --> 00:17:42,080
That's a big responsibility and kind of scary, but exciting,

537
00:17:42,080 --> 00:17:43,080
too.

538
00:17:43,080 --> 00:17:45,280
The story of AI hasn't been fully written yet.

539
00:17:45,280 --> 00:17:46,640
And the EU is trying to write a chapter

540
00:17:46,640 --> 00:17:47,760
that we can all be proud of.

541
00:17:47,760 --> 00:17:48,400
Well said.

542
00:17:48,400 --> 00:17:50,560
Well, that just about wraps up our deep dive

543
00:17:50,560 --> 00:17:52,600
into the EU's AI act.

544
00:17:52,600 --> 00:17:53,600
Woo.

545
00:17:53,600 --> 00:17:54,560
This is a lot to take in.

546
00:17:54,560 --> 00:17:55,520
But I don't know about you.

547
00:17:55,520 --> 00:17:56,520
I'm feeling optimistic.

548
00:17:56,520 --> 00:17:58,280
This is a big experiment.

549
00:17:58,280 --> 00:18:01,120
And it's going to be fascinating to see what happens next.

550
00:18:01,120 --> 00:18:02,760
Thanks for joining us on this journey, everyone,

551
00:18:02,760 --> 00:18:04,160
to keep those brains buzzing as we all

552
00:18:04,160 --> 00:18:20,200
figure out this brave new world of AI together.

