1
00:00:00,071 --> 00:00:01,585
(classical music)

2
00:00:01,585 --> 00:00:03,590
Do you promise to take your wedded partner

3
00:00:03,590 --> 00:00:04,823
from this day forwards,

4
00:00:05,940 --> 00:00:06,837
for better or worse,

5
00:00:06,837 --> 00:00:08,800
for richer and poorer,

6
00:00:08,800 --> 00:00:09,678
in sickness and health,

7
00:00:09,678 --> 00:00:12,260
to love and to cherish,

8
00:00:12,260 --> 00:00:13,988
till death do you part?

9
00:00:13,988 --> 00:00:16,290
(classical music)

10
00:00:16,290 --> 00:00:17,221
I do.

11
00:00:17,221 --> 00:00:19,915
(gentle music)

12
00:00:19,915 --> 00:00:21,320
I do.

13
00:00:21,320 --> 00:00:23,860
If a human being falls in love with AI,

14
00:00:23,860 --> 00:00:25,630
can he or she marry it?

15
00:00:25,630 --> 00:00:28,743
Can AI become a he or she?

16
00:00:29,919 --> 00:00:33,233
{\an8}(suspenseful music)

17
00:00:33,233 --> 00:00:34,066
(electronics hissing)

18
00:00:34,066 --> 00:00:34,899
(lasers whizzing)

19
00:00:34,899 --> 00:00:36,353
{\an8}(Chua sighing)

20
00:00:36,353 --> 00:00:37,186
(taut mellow music)

21
00:00:37,186 --> 00:00:40,186
(keyboard clicking)

22
00:00:47,340 --> 00:00:49,620
{\an8}Your honor, it's nothing but the truth.

23
00:00:49,620 --> 00:00:51,645
{\an8}Honey, would you like some coffee?

24
00:00:51,645 --> 00:00:52,710
{\an8}Yes, please.

25
00:00:52,710 --> 00:00:53,543
{\an8}(taut mellow music)

26
00:00:53,543 --> 00:00:54,376
{\an8}(mechanics whirring)

27
00:00:59,173 --> 00:01:01,100
(robot beeps)

28
00:01:01,100 --> 00:01:02,687
You mean you found a cure for cancer?

29
00:01:02,687 --> 00:01:05,436
Cancer cells exist in humans,

30
00:01:05,436 --> 00:01:08,380
and human cells mutate into cancer cells.

31
00:01:08,380 --> 00:01:11,110
So I shall eliminate all humans,

32
00:01:11,110 --> 00:01:13,341
starting with you.

33
00:01:13,341 --> 00:01:14,174
What?

34
00:01:14,174 --> 00:01:18,479
(Chua screams)
(electronic sizzling)

35
00:01:18,479 --> 00:01:19,312
(mischievous music)

36
00:01:19,312 --> 00:01:20,145
(robot whizzing)

37
00:01:22,208 --> 00:01:25,041
(taut gong music)

38
00:01:27,480 --> 00:01:29,340
I'm in Chiba Prefecture,

39
00:01:29,340 --> 00:01:31,160
two hours by car from Tokyo,

40
00:01:31,160 --> 00:01:32,739
{\an8}to attend a funeral.

41
00:01:32,739 --> 00:01:34,368
{\an8}(minister praying)

42
00:01:34,368 --> 00:01:36,476
(bell tolling)

43
00:01:36,476 --> 00:01:39,850
(minister praying)

44
00:01:39,850 --> 00:01:43,920
A farewell not for humans, but for dogs.

45
00:01:43,920 --> 00:01:46,670
And to be precise, robot dogs,

46
00:01:46,670 --> 00:01:48,910
AIBOs manufactured by Sony.

47
00:01:50,558 --> 00:01:55,058
{\an8}(minister speaks in foreign language)

48
00:02:03,929 --> 00:02:08,929
(clapper clacking)
(plaintive music)

49
00:02:11,251 --> 00:02:15,751
{\an8}(minister speaks in foreign language)

50
00:02:21,729 --> 00:02:23,604
(mellow Asian music)

51
00:02:23,604 --> 00:02:27,140
I guess I'm feeling a little bit emotional.

52
00:02:27,140 --> 00:02:31,330
{\an8}I don't know if it's because I was just at a funeral.

53
00:02:31,330 --> 00:02:34,570
I did feel that we were bidding goodbye to something.

54
00:02:34,570 --> 00:02:36,790
We've said goodbye to human beings,

55
00:02:36,790 --> 00:02:38,560
we've said goodbye to our pets,

56
00:02:38,560 --> 00:02:41,349
but I've never said goodbye to an object.

57
00:02:41,349 --> 00:02:43,160
(mellow music)

58
00:02:43,160 --> 00:02:46,080
To find out how these AIBO funerals started,

59
00:02:46,080 --> 00:02:49,523
the best person to ask is the brain behind this ceremony.

60
00:02:50,720 --> 00:02:55,610
{\an8}Owner has a very deep love

61
00:02:55,610 --> 00:02:58,810
{\an8}They feel that love like a partner,

62
00:02:58,810 --> 00:03:00,630
{\an8}not, you know, not stuff.

63
00:03:00,630 --> 00:03:03,784
Does AIBO bark?
<v ->Yes.

64
00:03:03,784 --> 00:03:05,540
{\an8}(mimics barking)

65
00:03:05,540 --> 00:03:09,090
Mr. Norimatsu is a former Sony engineer.

66
00:03:09,090 --> 00:03:10,690
After Sony stopped supporting

67
00:03:10,690 --> 00:03:12,890
the earlier generations of AIBO,

68
00:03:12,890 --> 00:03:16,410
his company became the last resort for many AIBO owners

69
00:03:16,410 --> 00:03:19,250
who couldn't bear to part with their beloved dogs

70
00:03:19,250 --> 00:03:21,550
when they stopped working.

71
00:03:21,550 --> 00:03:23,660
This all started with a special request

72
00:03:23,660 --> 00:03:24,493
by an AIBO owner more than four years ago.

73
00:03:24,493 --> 00:03:25,326
(robot whirring)

74
00:03:30,090 --> 00:03:33,710
{\an8}2014, one old woman

75
00:03:33,710 --> 00:03:37,020
{\an8}who was owner of AIBO,

76
00:03:37,020 --> 00:03:42,020
{\an8}she was try to repairing her set, her AIBO.

77
00:03:42,210 --> 00:03:44,840
{\an8}She searching many companies,

78
00:03:44,840 --> 00:03:46,813
{\an8}but unfortunate, she cannot find out.

79
00:03:47,710 --> 00:03:50,620
{\an8}Then finally, she came to our company.

80
00:03:50,620 --> 00:03:53,420
{\an8}We don't have any experience for the repairing for AIBO,

81
00:03:55,090 --> 00:03:58,600
{\an8}but our company philosophy,

82
00:03:58,600 --> 00:04:01,098
{\an8}we don't want to give up.

83
00:04:01,098 --> 00:04:02,280
AIBO, dance.

84
00:04:02,280 --> 00:04:03,570
{\an8}I was working Sony.

85
00:04:03,570 --> 00:04:08,520
{\an8}I utilized, fully utilized for this type of connection,

86
00:04:08,520 --> 00:04:12,390
{\an8}drinking party, friendship relation.

87
00:04:12,390 --> 00:04:16,460
{\an8}I collected from brain, all the engineer.

88
00:04:16,460 --> 00:04:17,337
{\an8}We achieved hope repairing.

89
00:04:17,337 --> 00:04:18,170
{\an8}(AIBO electronically barking)

90
00:04:19,700 --> 00:04:20,560
Owners sent their malfunctioning units

91
00:04:20,560 --> 00:04:25,560
to Norimatsu, hoping he can fix them, or at least,

92
00:04:26,120 --> 00:04:29,140
that their dogs can help revive others.

93
00:04:29,140 --> 00:04:33,080
{\an8}They donated us, utilize former part of parts,

94
00:04:33,080 --> 00:04:35,280
{\an8}for repairing first.

95
00:04:35,280 --> 00:04:38,450
{\an8}Owner feel that this AIBO

96
00:04:38,450 --> 00:04:43,030
{\an8}continue to has a life in the another AIBOs.

97
00:04:43,030 --> 00:04:43,888
{\an8}Owners has a very, very happy, you know,

98
00:04:43,888 --> 00:04:48,880
{\an8}and the peace of mind.

99
00:04:48,880 --> 00:04:52,150
{\an8}This is a, you know, one of reason

100
00:04:52,150 --> 00:04:56,763
{\an8}why we have a funeral event,

101
00:04:56,763 --> 00:04:59,050
{\an8}for thanks for the owners.

102
00:04:59,050 --> 00:05:04,050
Right, so, it's almost like organ donation.

103
00:05:04,558 --> 00:05:05,391
Yes. Yes.

104
00:05:06,230 --> 00:05:08,280
How do I say, give me your paw.

105
00:05:08,280 --> 00:05:09,447
How do I say?

106
00:05:09,447 --> 00:05:10,321
Ote.

107
00:05:10,321 --> 00:05:13,355
(speaks in foreign language)

108
00:05:13,355 --> 00:05:14,772
(robot speaks in foreign language)

109
00:05:14,772 --> 00:05:15,701
Hai.

110
00:05:15,701 --> 00:05:17,095
(hands clapping)

111
00:05:17,095 --> 00:05:18,003
So cute.

112
00:05:18,003 --> 00:05:22,173
AIBO is indeed very cute and adorable.

113
00:05:23,184 --> 00:05:24,178
Or as the Japanese would say Kawai,

114
00:05:24,178 --> 00:05:27,540
but I'm not quite convinced

115
00:05:27,540 --> 00:05:30,554
that someone can be this devoted to a robot pup.

116
00:05:30,554 --> 00:05:33,310
I've pets, cats, and dogs,

117
00:05:33,310 --> 00:05:35,160
and they're very much part of my family,

118
00:05:35,160 --> 00:05:37,020
but can someone grow as attached

119
00:05:37,020 --> 00:05:39,339
or even more to a robot dog?

120
00:05:39,339 --> 00:05:42,006
(upbeat music)

121
00:05:43,200 --> 00:05:44,727
For this.

122
00:05:44,727 --> 00:05:45,560
(speaks in foreign language)

123
00:05:45,560 --> 00:05:47,299
Look at me.

124
00:05:47,299 --> 00:05:49,543
No, he's looking at you.

125
00:05:49,543 --> 00:05:53,293
{\an8}(speaks in foreign language)

126
00:06:12,253 --> 00:06:16,003
{\an8}(speaks in foreign language)

127
00:06:32,665 --> 00:06:34,740
For the last 13 years.

128
00:06:34,740 --> 00:06:38,633
Portos has been a significant part of your Yurito's life.

129
00:06:38,633 --> 00:06:40,948
She brings him everywhere.

130
00:06:40,948 --> 00:06:43,843
They are an inseparable duo.

131
00:06:43,843 --> 00:06:47,676
(speaks in foreign language)

132
00:07:00,700 --> 00:07:02,270
Seeing how Portos occupies

133
00:07:02,270 --> 00:07:04,453
a special place in Yurito's heart,

134
00:07:05,530 --> 00:07:09,080
I begin to understand how so many other AIBO owners

135
00:07:09,080 --> 00:07:12,585
can be just as attached to their robot pups.

136
00:07:12,585 --> 00:07:17,585
AIBO's, to them, are not machines, but people.

137
00:07:19,150 --> 00:07:22,230
My name is Sophia, the latest and most advanced robot-

138
00:07:22,230 --> 00:07:25,071
Machines recognized as people.

139
00:07:25,071 --> 00:07:27,470
If they are regarded as people

140
00:07:27,470 --> 00:07:29,860
shouldn't they have rights too?

141
00:07:29,860 --> 00:07:33,010
Even Sophia, the AI powered humanoid,

142
00:07:33,010 --> 00:07:35,523
has been granted citizenship by Saudi Arabia.

143
00:07:35,523 --> 00:07:39,770
While Sophia's citizenship has been criticized

144
00:07:39,770 --> 00:07:41,098
as a publicity stunt,

145
00:07:41,098 --> 00:07:44,700
the idea of e-personhood has also been debated

146
00:07:44,700 --> 00:07:46,601
in the European Union parliament

147
00:07:46,601 --> 00:07:49,378
We are machines and we try to help humans.

148
00:07:49,378 --> 00:07:52,206
But what exactly are robot rights?

149
00:07:52,206 --> 00:07:54,706
(jazz music)

150
00:07:55,773 --> 00:07:58,080
I am here at Boston College Law School

151
00:07:58,080 --> 00:08:01,880
to meet an attorney who believes that AI robots

152
00:08:01,880 --> 00:08:03,665
should have rights, right?

153
00:08:03,665 --> 00:08:06,498
(patriotic music)

154
00:08:07,340 --> 00:08:10,092
Boston is home to the famous freedom trail,

155
00:08:10,092 --> 00:08:13,680
marking significant locations that tell the story

156
00:08:13,680 --> 00:08:16,673
of the American revolution against the British.

157
00:08:17,550 --> 00:08:22,550
With AI, are we also charting a new history and humanity?

158
00:08:22,592 --> 00:08:26,800
And we, and our family members that are serviced

159
00:08:26,800 --> 00:08:29,033
by these robots are gonna grow attached to them.

160
00:08:30,944 --> 00:08:32,276
And we're going to start to think of them as friends,

161
00:08:32,276 --> 00:08:33,349
and maybe even family members.

162
00:08:33,349 --> 00:08:35,960
So we will start to come up with robot protection laws,

163
00:08:35,960 --> 00:08:37,860
to make sure that robots are treated

164
00:08:37,860 --> 00:08:39,480
with a certain level of respect,

165
00:08:39,480 --> 00:08:41,350
in the same way that we have animal rights laws

166
00:08:41,350 --> 00:08:45,362
to ensure that animals receive a certain amount of respect.

167
00:08:45,362 --> 00:08:48,340
Now let's get to a discussion of what are the elements

168
00:08:48,340 --> 00:08:51,737
of limited legal personhood that we should give to AI?

169
00:08:51,737 --> 00:08:53,370
(bright music)

170
00:08:53,370 --> 00:08:56,880
John Weaver is a Boston based AI attorney.

171
00:08:56,880 --> 00:09:00,190
Profession that I haven't heard of until recently.

172
00:09:00,190 --> 00:09:02,950
He's the author of "Robots are People Too".

173
00:09:02,950 --> 00:09:05,347
And an advocate for limited personhood

174
00:09:05,347 --> 00:09:07,433
for AI powered machines.

175
00:09:07,433 --> 00:09:11,810
A legal person is not a person in real life,

176
00:09:11,810 --> 00:09:13,050
but under the law,

177
00:09:13,050 --> 00:09:15,670
we consider it legal persons like a corporation.

178
00:09:15,670 --> 00:09:18,610
Corporation isn't a person in real life,

179
00:09:18,610 --> 00:09:20,217
but it can sign contracts.

180
00:09:20,217 --> 00:09:22,410
It can own property, it can be sued.

181
00:09:22,410 --> 00:09:25,880
And I think that limited legal personhood for AI,

182
00:09:25,880 --> 00:09:27,280
can be used similarly.

183
00:09:27,280 --> 00:09:28,980
We're developing the technology

184
00:09:28,980 --> 00:09:31,810
where we can also have shopping assistants

185
00:09:31,810 --> 00:09:33,757
that go out and buy things for us.

186
00:09:33,757 --> 00:09:36,710
And they can engage in complicated shopping transactions

187
00:09:36,710 --> 00:09:39,070
for us, conveyances, real estate purchases,

188
00:09:39,070 --> 00:09:40,060
things like that.

189
00:09:40,060 --> 00:09:41,460
Welcome to Henn-na Hotel.

190
00:09:41,460 --> 00:09:44,237
Those robots would be able to

191
00:09:44,237 --> 00:09:45,560
be designated by individuals,

192
00:09:45,560 --> 00:09:49,514
or by companies to engage in negotiations and contracts.

193
00:09:49,514 --> 00:09:51,580
Are, there we go.

194
00:09:51,580 --> 00:09:53,680
So this is a cute cuddly robot

195
00:09:53,680 --> 00:09:55,630
that is designed to interact with senior citizens,

196
00:09:55,630 --> 00:09:58,070
either at senior citizens centers or nursing homes.

197
00:09:58,070 --> 00:10:01,695
Besides granting contract rights to AI,

198
00:10:01,695 --> 00:10:05,020
John believes that robots should also be protected.

199
00:10:05,020 --> 00:10:08,859
And people interact with these robots,

200
00:10:08,859 --> 00:10:11,188
like these robots, trust these robots,

201
00:10:11,188 --> 00:10:14,243
and our laws don't anticipate that at all.

202
00:10:17,924 --> 00:10:20,132
(singing in foreign language)

203
00:10:20,132 --> 00:10:21,981
(piano music)

204
00:10:21,981 --> 00:10:23,184
Our laws don't anticipate

205
00:10:23,184 --> 00:10:24,955
that a non-person could serve these functions.

206
00:10:24,955 --> 00:10:26,830
There are things that AI can do, that we want it to do,

207
00:10:26,830 --> 00:10:29,435
that will really be helpful to people,

208
00:10:29,435 --> 00:10:30,905
but that AI can't necessarily do under the law right now.

209
00:10:30,905 --> 00:10:31,738
It's so contentious,

210
00:10:31,738 --> 00:10:35,021
when you're talking about a patient with dementia,

211
00:10:35,021 --> 00:10:39,557
like grandpa wanting to leave all his money to.

212
00:10:39,557 --> 00:10:40,390
Yes.

213
00:10:40,390 --> 00:10:41,828
A robot.

214
00:10:41,828 --> 00:10:44,080
In that instance, you want the robot to be there,

215
00:10:44,080 --> 00:10:47,910
to take care of grandpa, to make sure that grandpa is safe,

216
00:10:47,910 --> 00:10:49,290
provide social interaction.

217
00:10:49,290 --> 00:10:52,447
But that's the extent of the legal personhood,

218
00:10:52,447 --> 00:10:56,297
the right to be named as the guardian or conservator,

219
00:10:56,297 --> 00:11:00,160
or to have power of attorney on behalf of somebody,

220
00:11:00,160 --> 00:11:02,080
the robot itself does not have the right

221
00:11:02,080 --> 00:11:04,450
to inherit money or property.

222
00:11:04,450 --> 00:11:06,840
And so that would be void from the beginning.

223
00:11:06,840 --> 00:11:09,320
It would not be permitted under the law.

224
00:11:09,320 --> 00:11:10,893
Rich people, don't worry.

225
00:11:12,536 --> 00:11:13,583
(John laughing)

226
00:11:13,583 --> 00:11:14,995
Or children of rich people, don't worry.

227
00:11:14,995 --> 00:11:16,645
(both laughing)

228
00:11:16,645 --> 00:11:18,980
(upbeat rock music)

229
00:11:18,980 --> 00:11:21,537
Good to know that robots can't inherit our money.

230
00:11:21,537 --> 00:11:25,625
But if we were to entrust AI machines to act as our agent,

231
00:11:25,625 --> 00:11:28,470
what happens when they make mistakes?

232
00:11:28,470 --> 00:11:30,250
Can we hold them accountable?

233
00:11:30,250 --> 00:11:31,083
(upbeat music)

234
00:11:31,083 --> 00:11:31,916
(machine whirring)

235
00:11:38,130 --> 00:11:40,713
(upbeat music)

236
00:11:41,581 --> 00:11:42,414
(machine whirring)

237
00:11:42,414 --> 00:11:43,247
(train tracks clacking)

238
00:11:45,634 --> 00:11:48,862
(cheerful music)

239
00:11:48,862 --> 00:11:52,089
AI's becoming increasingly prevalent in our lives.

240
00:11:52,089 --> 00:11:52,922
It's in our phone.

241
00:11:52,922 --> 00:11:53,755
(Siri bleeping)

242
00:11:53,755 --> 00:11:55,720
It's even our personal assistant.

243
00:11:55,720 --> 00:11:58,308
Hey Siri, will I need an umbrella today?

244
00:11:58,308 --> 00:11:59,430
(Siri bleeping)

245
00:11:59,430 --> 00:12:01,535
Yes, I think we'll see some rain today.

246
00:12:01,535 --> 00:12:03,716
It helps our traffic system

247
00:12:03,716 --> 00:12:05,930
determine traffic light timings

248
00:12:05,930 --> 00:12:08,192
and public transport dispatch.

249
00:12:08,192 --> 00:12:09,990
It's even in our restaurants.

250
00:12:09,990 --> 00:12:13,330
(patrons murmuring)

251
00:12:13,330 --> 00:12:16,800
And now we're seeing AI powered autonomous vehicles

252
00:12:16,800 --> 00:12:18,203
being tested on our streets.

253
00:12:19,227 --> 00:12:22,130
(upbeat music)

254
00:12:22,130 --> 00:12:24,800
Driverless cars come with big promises.

255
00:12:24,800 --> 00:12:26,000
They don't get tired.

256
00:12:26,000 --> 00:12:27,370
They don't get angry.

257
00:12:27,370 --> 00:12:28,650
They don't get distracted.

258
00:12:28,650 --> 00:12:30,840
And of course, there'll be no more drink driving cases,

259
00:12:30,840 --> 00:12:31,930
because it's far as we know,

260
00:12:31,930 --> 00:12:34,280
robots don't drink cocktails.

261
00:12:34,280 --> 00:12:36,150
But how do they make decisions on the road

262
00:12:36,150 --> 00:12:38,720
to ensure everyone's safety?

263
00:12:38,720 --> 00:12:41,431
And of course, with, with my best interests at heart.

264
00:12:41,431 --> 00:12:44,264
(cheerful music)

265
00:12:45,940 --> 00:12:47,860
We built very accurate maps of the environment,

266
00:12:47,860 --> 00:12:51,080
maps the car's able to localize itself, to know where it is.

267
00:12:51,080 --> 00:12:52,910
And once we give a destination to the car,

268
00:12:52,910 --> 00:12:54,850
it'll start to plan a path from where it is

269
00:12:54,850 --> 00:12:55,990
to where it wants to go.

270
00:12:55,990 --> 00:12:58,190
We use different kinds of sensors on our cars.

271
00:12:58,190 --> 00:13:00,301
Cameras, lidars, radars.

272
00:13:00,301 --> 00:13:01,999
What dars?

273
00:13:01,999 --> 00:13:04,550
Lidar, it's a laser sensor.

274
00:13:04,550 --> 00:13:07,220
It's a distance measuring device and radar,

275
00:13:07,220 --> 00:13:08,730
using a different kind of technology

276
00:13:08,730 --> 00:13:10,360
to also measure distance.

277
00:13:10,360 --> 00:13:13,732
So we use various sensors, essentially, to detect objects.

278
00:13:13,732 --> 00:13:17,419
We use deep learning for object classification

279
00:13:17,419 --> 00:13:20,640
to tell us what object it is, whether it's a person,

280
00:13:20,640 --> 00:13:22,031
whether it's a vehicle.

281
00:13:22,031 --> 00:13:23,482
Or an otter.

282
00:13:23,482 --> 00:13:24,664
Or an otter

283
00:13:24,664 --> 00:13:26,400
or wild dog.
Wild dog.

284
00:13:26,400 --> 00:13:27,310
And other animals.

285
00:13:27,310 --> 00:13:29,800
So imagine if you had lots of people sitting in a car,

286
00:13:29,800 --> 00:13:31,920
looking around, checking all your blind spots,

287
00:13:31,920 --> 00:13:34,510
and telling you, "I see a car speeding

288
00:13:34,510 --> 00:13:35,580
from hundred meters away.

289
00:13:35,580 --> 00:13:37,620
So maybe it we should just switch to the next lane."

290
00:13:37,620 --> 00:13:40,230
Personally, as a driver, I won't,

291
00:13:40,230 --> 00:13:42,450
I don't think I'll appreciate a hundred back seat drivers,

292
00:13:42,450 --> 00:13:43,851
in my car.

293
00:13:43,851 --> 00:13:45,478
That's true. that's true.

294
00:13:45,478 --> 00:13:46,861
That's why we have a robot to handle that.

295
00:13:46,861 --> 00:13:48,440
Once we know what objects there are in the environment,

296
00:13:48,440 --> 00:13:50,570
how would then figure out

297
00:13:50,570 --> 00:13:52,190
how it should go to it's destination,

298
00:13:52,190 --> 00:13:54,390
whether it should keep to its own lane,

299
00:13:54,390 --> 00:13:56,180
whether it should change lanes, make a left turn,

300
00:13:56,180 --> 00:13:58,370
right turn, and so on and so forth.

301
00:13:58,370 --> 00:13:59,279
James tries hard to unscramble

302
00:13:59,279 --> 00:14:01,650
the complex technology

303
00:14:01,650 --> 00:14:03,919
behind these driverless cars to assure me

304
00:14:03,919 --> 00:14:07,228
that they are safer than human drivers.

305
00:14:07,228 --> 00:14:09,070
I'm a driver myself,

306
00:14:09,070 --> 00:14:11,840
and I think I make pretty good decisions on the roads,

307
00:14:11,840 --> 00:14:13,460
but it sounds like he is telling me

308
00:14:13,460 --> 00:14:15,440
that an AI powered driverless car

309
00:14:15,440 --> 00:14:17,165
can make better decisions than me.

310
00:14:17,165 --> 00:14:18,650
Is this technology, you know,

311
00:14:18,650 --> 00:14:21,562
in some ways inflexible?

312
00:14:21,562 --> 00:14:26,410
That it may cause someone else to hit something else.

313
00:14:26,410 --> 00:14:28,640
Some rules are more important than others.

314
00:14:28,640 --> 00:14:29,690
For example,

315
00:14:29,690 --> 00:14:32,917
the rule to not hit something is the most important rule.

316
00:14:32,917 --> 00:14:34,764
But there are other rules,

317
00:14:34,764 --> 00:14:36,350
for example, don't cross the double white line,

318
00:14:36,350 --> 00:14:38,420
or don't cross into a chevron.

319
00:14:38,420 --> 00:14:41,950
As human drivers we kind of know when to break

320
00:14:41,950 --> 00:14:44,271
some of these rules just to proceed.

321
00:14:44,271 --> 00:14:48,430
We try to understand how typical drivers behave.

322
00:14:48,430 --> 00:14:52,900
So one thing that we commonly see around this one-off areas,

323
00:14:52,900 --> 00:14:54,650
a lot of illegally parked vehicles.

324
00:14:55,535 --> 00:14:56,403
In order to move forward,

325
00:14:57,628 --> 00:14:59,064
you have to cross maybe a double white line,

326
00:14:59,064 --> 00:15:01,633
And the next lane could be one with oncoming traffic.

327
00:15:01,633 --> 00:15:03,980
And the next lane could be a one with oncoming traffic.

328
00:15:03,980 --> 00:15:06,242
Because of someone else violating a rule,

329
00:15:06,242 --> 00:15:09,800
it kind of forces you to also violate some parts

330
00:15:09,800 --> 00:15:12,303
of the rules in order to proceed and to move forward.

331
00:15:14,532 --> 00:15:16,030
And these are the kinds of things

332
00:15:16,030 --> 00:15:18,363
that we want the car to understand as well.

333
00:15:18,363 --> 00:15:20,070
In certain situations where full compliance

334
00:15:20,070 --> 00:15:21,869
of all rules is impossible,

335
00:15:21,869 --> 00:15:23,762
we need to know what kind of rules

336
00:15:23,762 --> 00:15:25,085
we could break in a sense.

337
00:15:25,085 --> 00:15:27,120
So you guys are engineers, software engineers,

338
00:15:27,120 --> 00:15:30,000
but when it comes to things like the hierarchy of rules,

339
00:15:30,000 --> 00:15:33,740
or all the gray areas, who do you guys consult with?

340
00:15:33,740 --> 00:15:36,057
We run a lot of internal simulation to validate

341
00:15:36,057 --> 00:15:38,930
whether these hierarchy of rules are correct.

342
00:15:38,930 --> 00:15:40,300
And once we have our findings,

343
00:15:40,300 --> 00:15:42,940
we actually do share it with the government agencies,

344
00:15:42,940 --> 00:15:44,720
like the land transport authority,

345
00:15:44,720 --> 00:15:47,493
or the traffic police to see whether this makes sense.

346
00:15:47,493 --> 00:15:51,570
GoPro, another GoPro over there.

347
00:15:51,570 --> 00:15:53,240
Good luck, guys.

348
00:15:53,240 --> 00:15:54,958
Enough talk, I think it's still best

349
00:15:54,958 --> 00:15:57,036
to experience it myself.

350
00:15:57,036 --> 00:15:58,998
(upbeat rock music)

351
00:15:58,998 --> 00:15:59,960
Okay, roger.

352
00:15:59,960 --> 00:16:03,533
Safety first, since the cars is still in it's testing stage,

353
00:16:03,533 --> 00:16:06,010
two chaperones are riding along with me.

354
00:16:06,010 --> 00:16:07,610
It picks up quite quickly, yeah?

355
00:16:08,658 --> 00:16:09,755
Okay, and then it battles.

356
00:16:09,755 --> 00:16:13,290
(road crunching)

357
00:16:13,290 --> 00:16:14,621
Oh the vehicle has slowed down,

358
00:16:14,621 --> 00:16:17,352
because there was a dispatch rider in the way.

359
00:16:17,352 --> 00:16:19,265
It has stopped by itself at the red light,

360
00:16:19,265 --> 00:16:20,883
which is pretty cool.

361
00:16:20,883 --> 00:16:23,300
(rock music)

362
00:16:24,900 --> 00:16:29,070
It's quite strange seeing the steering wheel turn by itself,

363
00:16:29,070 --> 00:16:31,053
without any human intervention.

364
00:16:31,053 --> 00:16:32,550
(electronics bleeping)

365
00:16:32,550 --> 00:16:34,885
Oh, oh, oh, wow so this uncle

366
00:16:34,885 --> 00:16:37,500
just tried to walk in front of the car,

367
00:16:37,500 --> 00:16:39,510
and, okay, it wasn't a jam break,

368
00:16:39,510 --> 00:16:41,200
but it was a fairly hard break.

369
00:16:41,200 --> 00:16:43,181
I felt tension in my seatbelt.

370
00:16:43,181 --> 00:16:46,160
(upbeat rock music)

371
00:16:46,160 --> 00:16:47,460
Oh, pedestrian crossing.

372
00:16:48,760 --> 00:16:50,198
Oh, okay it stopped.

373
00:16:50,198 --> 00:16:51,323
Hey.

374
00:16:52,380 --> 00:16:54,479
Okay that was a very interesting situation.

375
00:16:54,479 --> 00:16:57,779
There was a pedestrian who started crossing the road,

376
00:16:57,779 --> 00:16:59,850
but he stopped in the middle,

377
00:16:59,850 --> 00:17:01,670
and then the car stopped as well.

378
00:17:01,670 --> 00:17:03,050
And then after that,

379
00:17:03,050 --> 00:17:05,290
Interesting how it decided the pedestrian

380
00:17:05,290 --> 00:17:07,120
Interesting how was decided the pedestrian

381
00:17:07,120 --> 00:17:09,146
was giving way to us.

382
00:17:09,146 --> 00:17:14,146
Now, as a dedicated host,

383
00:17:14,475 --> 00:17:17,365
Now, as a dedicated host,

384
00:17:17,365 --> 00:17:19,350
I shall put my life on the line

385
00:17:19,350 --> 00:17:21,170
for you my dear audience

386
00:17:21,170 --> 00:17:24,940
and test its object identification and distance gauging.

387
00:17:24,940 --> 00:17:27,057
You know, all those high-tech sensors.

388
00:17:27,057 --> 00:17:28,510
(breaks screeching)

389
00:17:28,510 --> 00:17:29,343
Ah!

390
00:17:31,680 --> 00:17:33,457
I have to say, it did pass the reflex test.

391
00:17:33,457 --> 00:17:37,447
It definitely saw me using it's peripheral vision

392
00:17:37,447 --> 00:17:41,143
And then it did stop and with quite a bit of distance

393
00:17:41,143 --> 00:17:44,544
And then it did stop and with quite a bit of distance

394
00:17:44,544 --> 00:17:46,780
between the vehicle and me.

395
00:17:46,780 --> 00:17:47,888
So, um, not bad.

396
00:17:47,888 --> 00:17:52,888
I'm alive, still intact, all in one piece.

397
00:17:53,710 --> 00:17:55,283
I think.

398
00:17:55,283 --> 00:17:57,014
(video swooshing)

399
00:17:57,014 --> 00:17:58,890
Overall that was a pretty uneventful experience,

400
00:17:58,890 --> 00:18:01,300
but seriously we face so many different kinds

401
00:18:01,300 --> 00:18:03,610
of scenarios on the roads every day.

402
00:18:03,610 --> 00:18:05,660
Let's say I'm in a driverless car

403
00:18:05,660 --> 00:18:08,364
and an oncoming car makes an illegal turn.

404
00:18:08,364 --> 00:18:11,619
My car turns to avoid a collision,

405
00:18:11,619 --> 00:18:13,240
but in doing so,

406
00:18:13,240 --> 00:18:17,170
it may just hit a mother and a child scooting down the road.

407
00:18:17,170 --> 00:18:19,199
What would the driverless car do?

408
00:18:19,199 --> 00:18:21,140
I don't even know what I would do.

409
00:18:21,140 --> 00:18:24,050
Would the car protect me, the passenger,

410
00:18:24,050 --> 00:18:28,320
or will it sacrifice me for the mother and her kid?

411
00:18:28,320 --> 00:18:32,855
This dilemma is what philosophers call the trolley problem.

412
00:18:32,855 --> 00:18:34,874
(fingers snapping)

413
00:18:34,874 --> 00:18:37,669
(joyful music)

414
00:18:37,669 --> 00:18:40,640
(gasps) There's five people in front of me on this track.

415
00:18:40,640 --> 00:18:43,570
Oh, I could just move this lever and change tracks.

416
00:18:43,570 --> 00:18:47,409
But wait, there's someone in front of me on this track.

417
00:18:47,409 --> 00:18:50,141
I can stop this guidance. I'm going to hit anyway too.

418
00:18:50,141 --> 00:18:51,680
I can't stop this car in time not to hit anyone.

419
00:18:51,680 --> 00:18:52,613
what should I do?

420
00:18:53,600 --> 00:18:54,433
Argh!

421
00:18:57,699 --> 00:18:58,906
(fingers snapping)

422
00:18:58,906 --> 00:19:01,140
Whoa, I'm glad I didn't face any of those outcomes

423
00:19:01,140 --> 00:19:02,744
because they're all bad.

424
00:19:02,744 --> 00:19:05,594
(Chua mumbling)

425
00:19:05,594 --> 00:19:09,040
(narrator mumbling)

426
00:19:09,040 --> 00:19:11,132
Carina Prunkl is an AI ethicist.

427
00:19:11,132 --> 00:19:15,875
Her research focuses on ethical and societal impacts of AI

428
00:19:15,875 --> 00:19:19,840
and its effects on human cognition.

429
00:19:19,840 --> 00:19:23,403
This dilemma with driverless vehicles is right up her alley.

430
00:19:25,660 --> 00:19:28,034
As a philosopher, we love the trolley problem.

431
00:19:28,034 --> 00:19:31,230
It's a good way of testing someone's intuition,

432
00:19:31,230 --> 00:19:32,720
their moral intuitions.

433
00:19:32,720 --> 00:19:34,940
So would you rather save yourself,

434
00:19:34,940 --> 00:19:36,690
or the school bus of children?

435
00:19:36,690 --> 00:19:39,040
So all the time we're driving on the street,

436
00:19:39,040 --> 00:19:40,930
and we have to make rapid decisions

437
00:19:40,930 --> 00:19:42,770
that have moral consequences.

438
00:19:42,770 --> 00:19:44,532
Let's just imagine that, you know,

439
00:19:44,532 --> 00:19:45,991
So, you see a yellow light and you decide to go for it

440
00:19:45,991 --> 00:19:48,520
So you see a yellow light and you decide to go for it

441
00:19:48,520 --> 00:19:49,680
instead of hitting the brake.

442
00:19:49,680 --> 00:19:52,360
That can have serious ethical consequences,

443
00:19:52,360 --> 00:19:54,040
and yet you do it all the time.

444
00:19:54,040 --> 00:19:56,255
So these are things humans do as a second nature,

445
00:19:56,255 --> 00:19:58,505
but these are things that autonomous vehicles

446
00:19:59,693 --> 00:20:00,526
We've had trolley problems

447
00:20:00,526 --> 00:20:03,732
for a hundred years.

448
00:20:03,732 --> 00:20:06,175
Ever since we've had cars,

449
00:20:06,175 --> 00:20:08,996
it's not like the trolley problem's a new problem.

450
00:20:08,996 --> 00:20:09,873
it's not like the trolley problem's a new problem.

451
00:20:12,626 --> 00:20:14,162
What's interesting is that

452
00:20:14,162 --> 00:20:15,393
because we have to program a computer,

453
00:20:15,393 --> 00:20:16,690
we have to decide what happens up front,

454
00:20:16,690 --> 00:20:17,924
before the accident,

455
00:20:17,924 --> 00:20:19,292
before we even turn the machine on,

456
00:20:19,292 --> 00:20:21,488
we have to decide how will it behave in those circumstances?

457
00:20:21,488 --> 00:20:24,053
And we don't have, you know,

458
00:20:24,053 --> 00:20:27,645
Toby is a respected AI scientist,

459
00:20:27,645 --> 00:20:28,880
Toby is a respected AI scientist,

460
00:20:28,880 --> 00:20:31,398
and is seen as a rock star in this circle.

461
00:20:31,398 --> 00:20:34,843
He started focusing on AI in the 1980's

462
00:20:34,843 --> 00:20:36,479
when it was still a pipe dream.

463
00:20:36,479 --> 00:20:39,444
No, I mean, from a technical perspective,

464
00:20:39,444 --> 00:20:42,710
No. I mean, from a technical perspective,

465
00:20:42,710 --> 00:20:45,850
is that cars aren't aware of the world enough

466
00:20:45,850 --> 00:20:48,105
to actually be making these decisions.

467
00:20:48,105 --> 00:20:50,460
You see and there's no code in an autonomous car today,

468
00:20:50,460 --> 00:20:53,510
that's deciding the answer to a trolley problem.

469
00:20:53,510 --> 00:20:54,769
It has no idea of what's happening around it,

470
00:20:54,769 --> 00:20:56,218
to be able to make this sorts of quality of decision,

471
00:20:56,218 --> 00:21:00,642
about whose life it's saving.

472
00:21:00,642 --> 00:21:03,047
Phew, so at least the car isn't deciding who lives,

473
00:21:03,047 --> 00:21:07,500
Phew, so at least the car isn't deciding who lives,

474
00:21:07,500 --> 00:21:09,020
and who dies,

475
00:21:09,020 --> 00:21:11,893
but still when an accident occurs,

476
00:21:11,893 --> 00:21:14,204
The manufactured item via a person.

477
00:21:14,204 --> 00:21:16,152
The manufactured item via a person.

478
00:21:16,152 --> 00:21:17,943
Could it be what John Weaver talked about?

479
00:21:19,313 --> 00:21:20,400
Granting e-personhood to the driverless car,

480
00:21:20,400 --> 00:21:22,641
so that it could be held responsible.

481
00:21:22,641 --> 00:21:24,348
(Grant mumbling)

482
00:21:24,348 --> 00:21:25,954
When we talk about extending electronic personhood

483
00:21:25,954 --> 00:21:29,135
to some sort of AI system, there's two sides to that.

484
00:21:29,135 --> 00:21:32,570
There's the giving, rights to a machine.

485
00:21:32,570 --> 00:21:34,540
And then there's the question of responsibility.

486
00:21:34,540 --> 00:21:36,940
My own view is that the AI technologies

487
00:21:36,940 --> 00:21:38,725
aren't sophisticated enough to be held responsible.

488
00:21:38,725 --> 00:21:41,590
And part of the reason for that is that there's no real

489
00:21:41,590 --> 00:21:44,660
punitive measure we to hold an algorithm responsible, right?

490
00:21:44,660 --> 00:21:46,510
We can't send it to jail.

491
00:21:46,510 --> 00:21:48,115
It doesn't feel shame.

492
00:21:48,115 --> 00:21:50,543
It has no money, so we can't fine it.

493
00:21:50,543 --> 00:21:54,597
What if something happens who is held responsible then?

494
00:21:54,597 --> 00:21:57,840
Is it the driver of the car person sitting in the car?

495
00:21:57,840 --> 00:22:00,053
Is it the person who bought the car?

496
00:22:01,445 --> 00:22:02,278
Is it the person who manufactured the car.

497
00:22:02,278 --> 00:22:03,811
We don't know.

498
00:22:03,811 --> 00:22:04,644
We don't have very good answers to these questions.

499
00:22:05,938 --> 00:22:08,272
And we need answers to these questions

500
00:22:08,272 --> 00:22:09,105
say they're gonna sell

501
00:22:10,706 --> 00:22:12,276
say they're going to sell

502
00:22:12,276 --> 00:22:13,391
fully autonomous level five cars by 2025,

503
00:22:13,391 --> 00:22:14,507
which is seven years away.

504
00:22:14,507 --> 00:22:15,340
Not far at all.

505
00:22:15,340 --> 00:22:16,173
Indeed lots of questions

506
00:22:16,173 --> 00:22:17,006
and not many answers, so far.

507
00:22:17,006 --> 00:22:18,809
Indeed lots of questions and not many answers, so far.

508
00:22:18,809 --> 00:22:22,300
While the responsibility of autonomous vehicles

509
00:22:22,300 --> 00:22:23,880
is still being debated,

510
00:22:23,880 --> 00:22:26,042
algorithms have been given the responsibility

511
00:22:26,042 --> 00:22:28,477
in deciding people's fate in the course of law.

512
00:22:28,477 --> 00:22:30,337
How can I help you?

513
00:22:30,337 --> 00:22:31,974
(robot buzzing)

514
00:22:31,974 --> 00:22:34,500
Can AI really make the right decision?

515
00:22:34,500 --> 00:22:35,701
Order in the court.

516
00:22:35,701 --> 00:22:37,013
And what is a right decision?

517
00:22:37,013 --> 00:22:42,013
To quote Toby,

518
00:22:42,203 --> 00:22:45,347
"This is the golden age for philosophers."

519
00:22:45,347 --> 00:22:47,137
(upbeat music)

520
00:22:47,137 --> 00:22:47,970
(upbeat music)

521
00:22:47,970 --> 00:22:48,803
(machine whirring)

522
00:22:48,803 --> 00:22:49,636
(upbeat music)

523
00:22:49,636 --> 00:22:50,469
(machine whirring)

524
00:22:57,468 --> 00:22:59,570
Your honor, that's the truth,

525
00:22:59,570 --> 00:23:02,120
the whole truth and nothing but the truth.

526
00:23:02,120 --> 00:23:05,388
You have to believe me, please believe me!

527
00:23:05,388 --> 00:23:08,055
(robot buzzing)

528
00:23:11,452 --> 00:23:12,859
(playful music)

529
00:23:12,859 --> 00:23:15,220
(robot buzzing)

530
00:23:15,220 --> 00:23:18,390
(upbeat music)

531
00:23:18,390 --> 00:23:19,633
On 23rd of March, 2016

532
00:23:19,633 --> 00:23:23,550
a 19 year old American girl started a Twitter account

533
00:23:23,550 --> 00:23:27,586
under the name TayTweets and handle @TayandYou.

534
00:23:27,586 --> 00:23:30,721
Can I just say that I'm stoked to meet you.

535
00:23:30,721 --> 00:23:31,554
Humans are super cool.

536
00:23:31,554 --> 00:23:33,960
Starting out with some good natured tweets

537
00:23:33,960 --> 00:23:37,683
Tay, within hours began sprouting racist, and sexually

538
00:23:37,683 --> 00:23:39,283
charged tweets.

539
00:23:40,787 --> 00:23:41,828
(phone bleeping)

540
00:23:41,828 --> 00:23:45,800
Consumed Tay was taken off Twitter by her parents

541
00:23:45,800 --> 00:23:47,784
or creator, Microsoft.

542
00:23:47,784 --> 00:23:50,268
I decided to pay a visit to Microsoft

543
00:23:50,268 --> 00:23:53,750
to find out what went wrong with Tay.

544
00:23:53,750 --> 00:23:56,387
It was a technical experiment.

545
00:23:56,387 --> 00:23:58,680
It was also a very big social,

546
00:23:58,680 --> 00:24:00,860
and cultural experiments itself.

547
00:24:00,860 --> 00:24:05,036
And what went on was that offensive language

548
00:24:05,036 --> 00:24:07,660
was being used to interact with it.

549
00:24:07,660 --> 00:24:11,364
So what happened was that Tay was reciprocating

550
00:24:11,364 --> 00:24:14,920
with racist tweets or offensive language?

551
00:24:14,920 --> 00:24:16,290
Yes that's, that's right?

552
00:24:16,290 --> 00:24:21,247
Unfortunately it reflected a very kind of negative side

553
00:24:21,247 --> 00:24:26,000
of how humans interact with intelligence bots.

554
00:24:26,000 --> 00:24:28,160
Soon after Tay went online,

555
00:24:28,160 --> 00:24:31,904
trolls were cursing and tweeting racist messages to Tay.

556
00:24:31,904 --> 00:24:33,390
(notification bleeping)

557
00:24:33,390 --> 00:24:34,953
I support genocide.

558
00:24:34,953 --> 00:24:39,493
I effing hate feminists and they should all die,

559
00:24:41,989 --> 00:24:43,245
and burn in hell.

560
00:24:43,245 --> 00:24:45,281
(notification bleeping)

561
00:24:45,281 --> 00:24:47,781
Hitler did nothing wrong.

562
00:24:47,781 --> 00:24:49,024
Oh my goodness.

563
00:24:49,024 --> 00:24:50,889
Tay was really learning from the worst

564
00:24:50,889 --> 00:24:51,889
of the worst amongst us.

565
00:24:51,889 --> 00:24:52,733
Could this have been prevented?

566
00:24:53,830 --> 00:24:54,663
(upbeat music)

567
00:24:54,663 --> 00:24:55,993
It's quite surprising.

568
00:24:57,196 --> 00:24:58,196
A company as technologically switched on,

569
00:24:58,196 --> 00:24:59,287
as Microsoft made all the mistakes it made.

570
00:24:59,287 --> 00:25:02,080
I mean, the fact that they didn't put a profanity filter on

571
00:25:02,080 --> 00:25:04,520
the input to Tay so that people could swear at it.

572
00:25:04,520 --> 00:25:07,510
And it was, it was learning off that swearing.

573
00:25:07,510 --> 00:25:08,630
It was the first mistake,

574
00:25:08,630 --> 00:25:10,880
but then the fact that they did put a profanity filter

575
00:25:10,880 --> 00:25:12,980
on the output so that it wouldn't swear back to you.

576
00:25:12,980 --> 00:25:14,360
And then the third mistake

577
00:25:14,360 --> 00:25:16,200
was they left the machine learning turned on

578
00:25:16,200 --> 00:25:18,316
so that it changed over time and adapted,

579
00:25:18,316 --> 00:25:21,170
and it reflected the data that it had been exposed to.

580
00:25:21,170 --> 00:25:23,533
And it behaved became racist, sexist,

581
00:25:23,533 --> 00:25:25,530
misogynous very quickly.

582
00:25:25,530 --> 00:25:27,660
But that illustrates a real problem,

583
00:25:27,660 --> 00:25:28,970
a challenge that we're going to face

584
00:25:28,970 --> 00:25:32,020
with autonomous cars, with all of the smart machines

585
00:25:32,020 --> 00:25:32,940
entering our lives.

586
00:25:32,940 --> 00:25:35,360
If we leave the machine learning turned on,

587
00:25:35,360 --> 00:25:37,495
then how they behave is not how they were programmed

588
00:25:37,495 --> 00:25:38,940
to behave.

589
00:25:38,940 --> 00:25:40,850
They're a product of also their environment.

590
00:25:40,850 --> 00:25:42,160
And that means the machines

591
00:25:42,160 --> 00:25:43,640
are gonna be much more unpredictable.

592
00:25:43,640 --> 00:25:44,890
And then who's responsible?

593
00:25:44,890 --> 00:25:47,220
Have you felt dizzy, unsteady?

594
00:25:47,220 --> 00:25:50,410
Hm, sounds like we need to make sure

595
00:25:50,410 --> 00:25:52,820
the machines we build have the right morals

596
00:25:52,820 --> 00:25:55,162
and ethics before we push them out.

597
00:25:55,162 --> 00:25:57,929
There are different strategies that are talked about.

598
00:25:57,929 --> 00:26:01,155
One is to put in ethical principles,

599
00:26:01,155 --> 00:26:03,270
and then debate emerges

600
00:26:03,270 --> 00:26:05,660
over what the right ethical principles are.

601
00:26:05,660 --> 00:26:08,010
So, instead of putting in principles,

602
00:26:08,010 --> 00:26:10,140
having a machine learn principles

603
00:26:10,140 --> 00:26:12,267
or derive them from human examples.

604
00:26:12,267 --> 00:26:15,150
And that's which what happened with Tay.

605
00:26:15,150 --> 00:26:18,450
So, if we have a machine that's learning our bar example,

606
00:26:18,450 --> 00:26:20,280
it could end up amplifying

607
00:26:21,423 --> 00:26:23,063
sort of the darker nature of humanity.

608
00:26:24,491 --> 00:26:25,620
We need to make absolutely sure that our data is clean,

609
00:26:25,620 --> 00:26:27,533
when there's real life consequences.

610
00:26:27,533 --> 00:26:28,926
After the Tay episode,

611
00:26:28,926 --> 00:26:31,310
Microsoft relooked at how it designed

612
00:26:31,310 --> 00:26:33,164
and developed AI technologies.

613
00:26:33,164 --> 00:26:35,790
And we said, "Hey, let's think about

614
00:26:35,790 --> 00:26:37,726
how do we build the necessary guard rails?"

615
00:26:37,726 --> 00:26:40,629
And one of them was content moderation,

616
00:26:40,629 --> 00:26:44,271
to filter and decipher offensive languages.

617
00:26:44,271 --> 00:26:47,567
Who is it that's putting up these guard rails.

618
00:26:47,567 --> 00:26:50,110
These technologies itself,

619
00:26:50,110 --> 00:26:51,980
it's gonna be used by millions of people

620
00:26:51,980 --> 00:26:53,770
from many diverse background.

621
00:26:53,770 --> 00:26:57,817
So, it's important for us to have people

622
00:26:57,817 --> 00:26:59,118
from the Liberal Arts major

623
00:26:59,118 --> 00:27:01,493
in humanities, economics psychology,

624
00:27:01,493 --> 00:27:05,260
to be able to bring a certain perspective

625
00:27:05,260 --> 00:27:08,710
into how this technology should be developed.

626
00:27:08,710 --> 00:27:11,957
It's not just about what AI can do, but what should it do?

627
00:27:11,957 --> 00:27:15,453
A biased and politically incorrect chat bot

628
00:27:15,453 --> 00:27:18,622
might seem trivial and almost funny,

629
00:27:18,622 --> 00:27:22,392
but what happens when bias exists in AI technologies

630
00:27:22,392 --> 00:27:26,749
with real life consequences, like in the court of law?

631
00:27:26,749 --> 00:27:28,383
Some courts in the United States

632
00:27:28,383 --> 00:27:30,360
have been using algorithms

633
00:27:30,360 --> 00:27:32,423
to recommend sentencing and bail.

634
00:27:32,423 --> 00:27:33,400
(video swooshing)

635
00:27:33,400 --> 00:27:36,450
Basically the algorithm will rank you with a risk score

636
00:27:36,450 --> 00:27:39,250
to see how likely you are to commit another crime

637
00:27:39,250 --> 00:27:41,600
so that the judges will know how much bail to set,

638
00:27:41,600 --> 00:27:43,677
or even how long your prison sentence should be.

639
00:27:43,677 --> 00:27:47,250
First, you'll be asked to complete a questionnaire

640
00:27:47,250 --> 00:27:49,059
with questions like gender.

641
00:27:49,059 --> 00:27:52,110
If you're a man your risk score will be higher

642
00:27:52,110 --> 00:27:54,580
because statistically males are more likely

643
00:27:54,580 --> 00:27:56,490
to commit a crime than females.

644
00:27:56,490 --> 00:27:57,760
Sorry, guys.

645
00:27:57,760 --> 00:27:59,480
Have you been arrested before?

646
00:27:59,480 --> 00:28:03,317
If you answered yes, your risk score will go up.

647
00:28:03,317 --> 00:28:05,376
Then there indirect questions like,

648
00:28:05,376 --> 00:28:08,970
how often have you moved in the past 12 months?

649
00:28:08,970 --> 00:28:11,156
Have you ever been expelled from school,

650
00:28:11,156 --> 00:28:15,700
and how often do you barely have enough money to get by?

651
00:28:15,700 --> 00:28:18,980
With each answer your risk score could potentially go up,

652
00:28:18,980 --> 00:28:22,330
and you might be labeled as high risk.

653
00:28:22,330 --> 00:28:25,060
This means that your bail will be set higher,

654
00:28:25,060 --> 00:28:27,070
or you could possibly be thrown into jail

655
00:28:27,070 --> 00:28:28,990
for a longer period of time.

656
00:28:28,990 --> 00:28:31,420
I mean, it seems like a great idea, right?

657
00:28:31,420 --> 00:28:34,587
Let emotionless impartial technology analyze your record,

658
00:28:34,587 --> 00:28:38,720
and come up with a decision rather than relying on a human.

659
00:28:38,720 --> 00:28:42,257
But the question is can machines be biased?

660
00:28:42,257 --> 00:28:44,381
(upbeat music)

661
00:28:44,381 --> 00:28:47,770
According to American Civil Liberty Union

662
00:28:47,770 --> 00:28:50,220
and investigative reporters at Publica,

663
00:28:50,220 --> 00:28:52,680
one such algorithm compass

664
00:28:52,680 --> 00:28:56,019
has been labeling people of color with much higher risk,

665
00:28:56,019 --> 00:28:59,652
resulting them being sentenced with longer jail terms,

666
00:28:59,652 --> 00:29:03,124
reflecting deep racial bias.

667
00:29:03,124 --> 00:29:06,100
As the manufacturer refuses to disclose

668
00:29:06,100 --> 00:29:07,744
how risk scores are calculated.

669
00:29:07,744 --> 00:29:11,851
It isn't clear what exactly resulted in such bias.

670
00:29:11,851 --> 00:29:14,434
(upbeat music)

671
00:29:15,460 --> 00:29:17,353
I'm here in Pennsylvania.

672
00:29:17,353 --> 00:29:20,380
One of the first states in America that is developing

673
00:29:20,380 --> 00:29:22,110
its own risk assessment tool,

674
00:29:22,110 --> 00:29:24,430
utilizing algorithms to help deal

675
00:29:25,554 --> 00:29:27,026
with the growing prison population.

676
00:29:27,026 --> 00:29:30,848
How do they avoid making the same mistakes as Compass?

677
00:29:30,848 --> 00:29:33,321
What are they doing differently?

678
00:29:33,321 --> 00:29:36,330
There's gonna be people you're categorizing as loaders

679
00:29:36,330 --> 00:29:39,200
for all crimes, can reinvent, or maybe murder.

680
00:29:39,200 --> 00:29:42,249
But a tool that's wrong 89% of the time

681
00:29:42,249 --> 00:29:46,058
is a ridiculously bad tool.

682
00:29:46,058 --> 00:29:46,891
The Sentencing Commission

683
00:29:46,891 --> 00:29:49,028
has a public hearing today.

684
00:29:49,028 --> 00:29:52,960
It's probably the best starting point for me to learn more.

685
00:29:52,960 --> 00:29:54,610
And I think that transparency

686
00:29:54,610 --> 00:29:57,017
and accountability are critical.

687
00:29:57,017 --> 00:29:58,643
They are retribution-

688
00:29:58,643 --> 00:30:00,840
The commission is made up of members

689
00:30:00,840 --> 00:30:03,370
from various backgrounds to make sure

690
00:30:03,370 --> 00:30:05,620
that all sides are represented,

691
00:30:05,620 --> 00:30:09,410
judges, parole, board officers, victim advocates,

692
00:30:09,410 --> 00:30:11,320
and of course, defense attorneys.

693
00:30:11,320 --> 00:30:14,985
I mean we disagree, but we all get along

694
00:30:14,985 --> 00:30:18,113
and I think in the end we want he best

695
00:30:18,113 --> 00:30:20,050
and fairest system that we can come up with.

696
00:30:20,050 --> 00:30:24,486
The idea behind risk assessment is trying to predict

697
00:30:24,486 --> 00:30:26,645
the risk of future offense.

698
00:30:26,645 --> 00:30:28,730
So the problem,

699
00:30:28,730 --> 00:30:31,174
the basic problem I have with risk assessment tools

700
00:30:31,174 --> 00:30:36,174
as a sentencing tool is that it's essentially using

701
00:30:37,583 --> 00:30:42,070
statistics to predict the possibility

702
00:30:42,070 --> 00:30:43,960
that you will re-offend.

703
00:30:43,960 --> 00:30:46,752
And then using that to sentence you for something

704
00:30:46,752 --> 00:30:47,585
that you already did.

705
00:30:47,585 --> 00:30:50,250
So, it's a little bit like a Sci-Fi movie,

706
00:30:50,250 --> 00:30:52,390
like minority report essentially.

707
00:30:52,390 --> 00:30:53,950
Could you give me some examples

708
00:30:53,950 --> 00:30:55,930
of like how it could be misused?

709
00:30:55,930 --> 00:31:00,050
Now, statistically where you're from or where you live

710
00:31:00,050 --> 00:31:01,483
is very predictive, right?

711
00:31:01,483 --> 00:31:03,816
Someone from Pittsburgh or Philadelphia,

712
00:31:03,816 --> 00:31:06,400
more high crime areas

713
00:31:06,400 --> 00:31:08,320
are going to be treated differently

714
00:31:08,320 --> 00:31:10,870
than someone from another county,

715
00:31:10,870 --> 00:31:12,479
which is a low crime area,

716
00:31:12,479 --> 00:31:15,850
that has nothing to do with who they really are,

717
00:31:15,850 --> 00:31:17,310
or what they really did.

718
00:31:17,310 --> 00:31:18,143
At an early point

719
00:31:18,143 --> 00:31:19,780
in the development of the model,

720
00:31:19,780 --> 00:31:23,270
we did have county as one of the factors,

721
00:31:23,270 --> 00:31:26,020
but then we also did closer look at that,

722
00:31:26,020 --> 00:31:29,500
and determined that it could be in fact, a proxy for race.

723
00:31:29,500 --> 00:31:31,180
And so we took that out.

724
00:31:31,180 --> 00:31:34,310
When we were going through

725
00:31:34,310 --> 00:31:36,900
the process of figuring out what to include,

726
00:31:36,900 --> 00:31:39,190
we had to be extremely careful

727
00:31:39,190 --> 00:31:42,953
and cautious not to use predictive factors

728
00:31:42,953 --> 00:31:47,953
that would also potentially be unconstitutional, unfair.

729
00:31:48,785 --> 00:31:51,740
I'm Jill Rangos I am the Chair of-

730
00:31:51,740 --> 00:31:54,034
Jill has been a judge for almost 20 years.

731
00:31:54,034 --> 00:31:58,480
While she's mindful of biases that creep into the tool,

732
00:31:58,480 --> 00:32:01,755
she welcomes the use of the risk assessment algorithm.

733
00:32:01,755 --> 00:32:05,040
What do you have to say about the criticism,

734
00:32:05,040 --> 00:32:08,940
that you may be labeling someone for a crime

735
00:32:08,940 --> 00:32:13,940
that he has not done, but he may do in the future.

736
00:32:14,143 --> 00:32:16,770
So when you do sentence somebody

737
00:32:16,770 --> 00:32:18,500
part of what you're trying to do

738
00:32:19,623 --> 00:32:22,270
is determine if I put them back in the community now,

739
00:32:22,270 --> 00:32:25,617
are they going to be a risk to public safety?

740
00:32:25,617 --> 00:32:27,715
The more information you have,

741
00:32:27,715 --> 00:32:31,270
the better you're going to do

742
00:32:31,270 --> 00:32:34,070
with making that prediction anyway,

743
00:32:34,070 --> 00:32:38,260
based on our experience and common sense,

744
00:32:38,260 --> 00:32:39,690
judges intuition.

745
00:32:39,690 --> 00:32:43,222
Adding to that some scientific evidence

746
00:32:43,222 --> 00:32:45,840
can only make that decision better.

747
00:32:45,840 --> 00:32:48,740
So, do you think some people are seeing

748
00:32:48,740 --> 00:32:53,660
this risk assessment tool through the minority report lens?

749
00:32:53,660 --> 00:32:54,510
Of course.

750
00:32:54,510 --> 00:32:56,600
And I think we all should be cautious.

751
00:32:56,600 --> 00:32:58,757
There's a famous Latin expression translated

752
00:32:58,757 --> 00:33:01,859
often applied in the medical community,

753
00:33:01,859 --> 00:33:03,683
"First do no harm."

754
00:33:03,683 --> 00:33:05,958
So, when we adopt risk assessment,

755
00:33:05,958 --> 00:33:08,790
we want to be sure that the benefits of it

756
00:33:08,790 --> 00:33:11,251
outweigh the possible negative consequences.

757
00:33:11,251 --> 00:33:13,470
And if they don't,

758
00:33:13,470 --> 00:33:15,648
then we shouldn't be going in that direction.

759
00:33:15,648 --> 00:33:18,625
(upbeat music)

760
00:33:18,625 --> 00:33:19,480
Aha, good morning.

761
00:33:19,480 --> 00:33:21,757
Good morning Bob.

762
00:33:21,757 --> 00:33:22,909
Good to see you.
<v ->I'm good thank you.

763
00:33:22,909 --> 00:33:24,120
Developing a risk assessment tool that is fair,

764
00:33:24,120 --> 00:33:26,070
and acceptable for all sides

765
00:33:26,070 --> 00:33:28,326
is a long and challenging process.

766
00:33:28,326 --> 00:33:30,906
Great view, the corner suite,

767
00:33:30,906 --> 00:33:34,077
Mark Bergstrum has been living and breathing

768
00:33:34,077 --> 00:33:36,957
this for the last eight years.

769
00:33:36,957 --> 00:33:41,157
The tool has also evolved from making sentencing decisions

770
00:33:41,157 --> 00:33:44,119
to a more conservative assessment.

771
00:33:44,119 --> 00:33:46,848
Soon into the process

772
00:33:46,848 --> 00:33:50,354
we thought that it would be better to use the instrument

773
00:33:50,354 --> 00:33:51,780
to identify where more information

774
00:33:51,780 --> 00:33:54,195
would help the judge to make a better decision.

775
00:33:54,195 --> 00:33:57,284
Because an individual who is high risk might be high risk

776
00:33:57,284 --> 00:33:59,410
for very many reasons.

777
00:33:59,410 --> 00:34:01,850
Could be 'cause of drug dependency,

778
00:34:01,850 --> 00:34:03,240
or something like that.

779
00:34:03,240 --> 00:34:06,862
Well, the response by the court is not necessarily,

780
00:34:06,862 --> 00:34:08,290
lock the person up for a longer period of time.

781
00:34:08,290 --> 00:34:09,900
We're just saying to the judge,

782
00:34:09,900 --> 00:34:12,800
get more information so that you make a better decision

783
00:34:12,800 --> 00:34:14,890
at sentencing about what to do.

784
00:34:14,890 --> 00:34:17,230
Progressing slowly and cautiously

785
00:34:17,230 --> 00:34:20,060
with a technology that we are still figuring out,

786
00:34:20,060 --> 00:34:22,160
is probably a good move.

787
00:34:22,160 --> 00:34:25,435
Once you have facial recognition software.

788
00:34:25,435 --> 00:34:26,497
Would that be the end of a line-up?

789
00:34:26,497 --> 00:34:28,300
Well that's a good question.

790
00:34:28,300 --> 00:34:30,852
But it's an undeniable fact that AI is here,

791
00:34:30,852 --> 00:34:32,089
and here to stay.

792
00:34:32,089 --> 00:34:33,300
(Chua murmuring)

793
00:34:33,300 --> 00:34:34,600
So, further down the road,

794
00:34:36,088 --> 00:34:39,420
do you think there won't be a need for human judges

795
00:34:39,420 --> 00:34:42,949
and instead we'll have robot judges powered by AI?

796
00:34:42,949 --> 00:34:46,080
Only time will tell, I don't know.

797
00:34:46,080 --> 00:34:49,150
There is nothing that can replace

798
00:34:49,150 --> 00:34:52,028
the emotion that you hear at sentencing

799
00:34:52,028 --> 00:34:55,035
from the victim, from the victim's family,

800
00:34:55,035 --> 00:34:57,930
from the defendant and the defendant's family,

801
00:34:57,930 --> 00:35:01,190
who also suffer consequences.

802
00:35:01,190 --> 00:35:04,625
And if you cannot also consider that,

803
00:35:04,625 --> 00:35:07,420
I think you lose the human aspect,

804
00:35:07,420 --> 00:35:12,420
the compassion and the true desire

805
00:35:13,922 --> 00:35:18,030
to want to help the people that are in front of you.

806
00:35:18,030 --> 00:35:20,159
The judge traditionally comes out.

807
00:35:20,159 --> 00:35:21,889
This is so cool.

808
00:35:21,889 --> 00:35:24,091
Have a little experience with this role.

809
00:35:24,091 --> 00:35:25,662
(hammer banging)

810
00:35:25,662 --> 00:35:28,320
Order in the court,

811
00:35:28,320 --> 00:35:30,260
court is in session.

812
00:35:30,260 --> 00:35:32,524
While, I get to role play as a judge on TV,

813
00:35:32,524 --> 00:35:34,620
I couldn't help but think

814
00:35:34,620 --> 00:35:37,317
how big a role AI will play in our lives.

815
00:35:37,317 --> 00:35:41,762
What about AI in our spiritual life?

816
00:35:41,762 --> 00:35:44,424
Will be viewed God differently?

817
00:35:44,424 --> 00:35:47,813
Or will AI even replace God?

818
00:35:48,879 --> 00:35:50,357
(upbeat music)

819
00:35:50,357 --> 00:35:53,107
(robot whirring)

820
00:35:55,187 --> 00:35:57,216
(upbeat music)

821
00:35:57,216 --> 00:36:00,499
(robot whirring)

822
00:36:00,499 --> 00:36:03,630
(religious music)

823
00:36:03,630 --> 00:36:06,230
Forgive me, father for I have sinned.

824
00:36:06,230 --> 00:36:07,510
May God,

825
00:36:07,510 --> 00:36:09,840
who has enlightened every heart

826
00:36:09,840 --> 00:36:11,430
help you to know your sins

827
00:36:11,430 --> 00:36:14,559
and trust in his mercy, repent then,

828
00:36:14,559 --> 00:36:18,662
and turn to God so that your sins may be wipes out.

829
00:36:18,662 --> 00:36:22,319
The times of refreshing may come from the Lord.

830
00:36:22,319 --> 00:36:24,130
(upbeat music)

831
00:36:24,130 --> 00:36:25,980
(robot whizzing)

832
00:36:25,980 --> 00:36:28,437
Beyond the ethical and legal dilemmas,

833
00:36:28,437 --> 00:36:32,594
what role will AI play in our spiritual life?

834
00:36:32,594 --> 00:36:35,958
(upbeat music)

835
00:36:35,958 --> 00:36:37,386
Formal Google engineer,

836
00:36:37,386 --> 00:36:39,440
Anthony Levandowski has started a church,

837
00:36:39,440 --> 00:36:42,727
Way of the Future where they worship AI as God.

838
00:36:42,727 --> 00:36:47,333
Is this guy doing this as a gimmick for the attention?

839
00:36:49,644 --> 00:36:51,520
(moaning)

840
00:36:51,520 --> 00:36:52,783
It's for real.

841
00:36:53,840 --> 00:36:56,930
Levandowski believes that AI is of a higher power,

842
00:36:56,930 --> 00:36:59,866
and so is God, so shouldn't we treat AI as God?

843
00:36:59,866 --> 00:37:04,866
I guess he has his convictions, but I'm not that convinced.

844
00:37:08,150 --> 00:37:09,950
Wait, let's think about it.

845
00:37:09,950 --> 00:37:12,280
If we trust AI to help us find jobs

846
00:37:12,280 --> 00:37:14,110
and even our life partner,

847
00:37:14,110 --> 00:37:17,513
couldn't AI also be our spiritual guide?

848
00:37:21,378 --> 00:37:22,894
(bell ringing)

849
00:37:22,894 --> 00:37:25,477
(drum beating)

850
00:37:27,460 --> 00:37:29,650
In Japan, the Pepper Robot

851
00:37:29,650 --> 00:37:31,894
has been conducting funerals for humans.

852
00:37:31,894 --> 00:37:34,647
(robot singing)

853
00:37:34,647 --> 00:37:35,496
(drum beating)

854
00:37:35,496 --> 00:37:37,450
(calm music)

855
00:37:37,450 --> 00:37:39,010
How did the idea of using Pepper

856
00:37:39,010 --> 00:37:41,053
as a priest for funerals come about?

857
00:37:42,371 --> 00:37:46,121
{\an8}(speaks in foreign language)

858
00:38:02,212 --> 00:38:06,680
{\an8}Pepper Robot was welcoming the customers, that only thing.

859
00:38:06,680 --> 00:38:11,174
{\an8}And IT department thinking that they want to use it

860
00:38:11,174 --> 00:38:13,900
{\an8}more like an innovation.

861
00:38:13,900 --> 00:38:17,560
So Pepper's job got up-sized essentially, right?

862
00:38:17,560 --> 00:38:20,903
Now that's some out of the box thinking.

863
00:38:22,947 --> 00:38:25,533
(traditional music)

864
00:38:25,533 --> 00:38:29,283
{\an8}(speaks in foreign language)

865
00:38:34,530 --> 00:38:37,530
But is there a demand for a robot priest

866
00:38:37,530 --> 00:38:39,453
in the real world?

867
00:38:39,453 --> 00:38:42,079
(traditional music continues)

868
00:38:42,079 --> 00:38:43,330
Who are these people that she conducts these ceremonies for?

869
00:38:43,330 --> 00:38:45,876
{\an8}We want to use Pepper Robot as a priest,

870
00:38:45,876 --> 00:38:49,960
{\an8}to people who doesn't have enough money or poor people.

871
00:38:49,960 --> 00:38:54,060
{\an8}And the people who have welfare.

872
00:38:54,060 --> 00:38:57,405
{\an8}We have to pay individually for the human priest,

873
00:38:57,405 --> 00:39:01,310
{\an8}like 500 dollar to 3,000 US dollars.

874
00:39:01,310 --> 00:39:02,980
So, what was people's first reactions

875
00:39:02,980 --> 00:39:06,643
when you said you're gonna have her conduct

876
00:39:06,643 --> 00:39:07,644
these funeral rights.

877
00:39:07,644 --> 00:39:09,220
{\an8}We are also very worried about it,

878
00:39:09,220 --> 00:39:11,300
{\an8}how they will react.

879
00:39:11,300 --> 00:39:12,133
{\an8}First thing,

880
00:39:13,111 --> 00:39:15,145
{\an8}they are laughing at it at Pepper priest.

881
00:39:15,145 --> 00:39:20,145
{\an8}They're like taking a picture, but during the same time,

882
00:39:20,720 --> 00:39:22,070
{\an8}they are crying.

883
00:39:22,070 --> 00:39:24,747
{\an8}It's kinda funny, but sad.

884
00:39:24,747 --> 00:39:27,080
(sad music)

885
00:39:28,300 --> 00:39:29,770
When I first heard that, that there was a robot

886
00:39:29,770 --> 00:39:32,500
that would conduct funerals for human beings,

887
00:39:32,500 --> 00:39:35,579
I thought, "Wow, that's a bit strange."

888
00:39:35,579 --> 00:39:40,310
But you know, I think I'm open to it now.

889
00:39:40,310 --> 00:39:43,490
Look, you know, when churches, when they collect tithes,

890
00:39:43,490 --> 00:39:45,340
there's no more about passing the hat,

891
00:39:45,340 --> 00:39:50,140
you can actually do a bank transfer to your church.

892
00:39:50,140 --> 00:39:54,330
Hello, technology, you know?

893
00:39:54,330 --> 00:39:55,260
But then again,

894
00:39:55,260 --> 00:39:58,528
I think the robot has to have a better voice quality.

895
00:39:58,528 --> 00:40:01,111
(upbeat music)

896
00:40:08,373 --> 00:40:10,013
{\an8}Here I am in Nashville, Tennessee,

897
00:40:10,013 --> 00:40:12,905
known as the country music, capital of the world,

898
00:40:12,905 --> 00:40:16,283
as well as Music City USA.

899
00:40:16,283 --> 00:40:18,105
(guitar music)

900
00:40:18,105 --> 00:40:20,138
Not for some Dolly Parton or Johnny Cash music,

901
00:40:20,138 --> 00:40:23,230
but to meet a group of people who have an interesting view

902
00:40:23,230 --> 00:40:26,348
on marrying AI and religion.

903
00:40:26,348 --> 00:40:29,007
♪ A little town ♪

904
00:40:29,007 --> 00:40:31,407
But I'm curious to hear about these Christians

905
00:40:31,407 --> 00:40:32,909
who are here today.

906
00:40:32,909 --> 00:40:34,510
Who are I thinking about incorporating AI

907
00:40:34,510 --> 00:40:38,168
into their religion or is it their religion into AI,

908
00:40:38,168 --> 00:40:41,873
wait, I'm confused.

909
00:40:42,730 --> 00:40:44,982
Let me just go in to hear what they've got to say.

910
00:40:44,982 --> 00:40:47,732
(cheerful music)

911
00:40:48,838 --> 00:40:49,671
Hello.

912
00:40:50,678 --> 00:40:51,852
Hi.

913
00:40:51,852 --> 00:40:53,238
What is transhumanism?

914
00:40:53,238 --> 00:40:54,573
And there's a lot of definitions.

915
00:40:54,573 --> 00:40:56,477
A lot of ways we can talk about this,

916
00:40:56,477 --> 00:40:58,602
but the simplest definition is that transhumanism

917
00:40:58,602 --> 00:41:00,575
is a movement to use science and technology

918
00:41:00,575 --> 00:41:04,750
to transform what it means to be human.

919
00:41:04,750 --> 00:41:06,900
First thing that Christian transhumanism

920
00:41:06,900 --> 00:41:07,736
is it's a conversation between

921
00:41:07,736 --> 00:41:12,736
the leading edges of scientific and technological thought

922
00:41:14,510 --> 00:41:18,110
and the broad world of people of faith.

923
00:41:18,110 --> 00:41:20,770
It sounds like the Christian transhumanists believe

924
00:41:20,770 --> 00:41:24,452
that technology and Christianity should work hand in hand.

925
00:41:24,452 --> 00:41:26,732
(hands clapping)

926
00:41:26,732 --> 00:41:28,020
(upbeat music)

927
00:41:28,020 --> 00:41:29,982
I'm gonna catch Micha the, founder of this conference,

928
00:41:29,982 --> 00:41:32,428
and ask him more about what he's trying to do.

929
00:41:32,428 --> 00:41:36,914
Yeah a lot of Christians are afraid of technology

930
00:41:36,914 --> 00:41:38,421
for some reason,

931
00:41:38,421 --> 00:41:41,685
But in our faith tradition,

932
00:41:41,685 --> 00:41:45,098
we're actually told that humans were created as beings

933
00:41:45,098 --> 00:41:49,270
made in the image of God.

934
00:41:49,270 --> 00:41:52,810
And God is a creative being, who brings new things

935
00:41:52,810 --> 00:41:54,170
into existence.

936
00:41:54,170 --> 00:41:56,080
And we're called to do the same thing,

937
00:41:56,080 --> 00:41:57,300
create new things.

938
00:41:57,300 --> 00:41:59,630
And doing, so, we transform the world,

939
00:41:59,630 --> 00:42:01,070
and we transform ourselves.

940
00:42:01,070 --> 00:42:02,490
So, for a transhumanists,

941
00:42:02,490 --> 00:42:04,923
what role do you think religion should play in AI?

942
00:42:04,923 --> 00:42:09,923
We have to bring our values into how we interact,

943
00:42:12,278 --> 00:42:17,278
with AI and treat it with compassion and with love.

944
00:42:17,840 --> 00:42:21,670
Treat it as our child, as our creation.

945
00:42:21,670 --> 00:42:25,268
And so we have to treat it with those religious values

946
00:42:25,268 --> 00:42:28,980
that humanity really cares about.

947
00:42:28,980 --> 00:42:32,880
So, we should be treating AI like how we bring up a child.

948
00:42:32,880 --> 00:42:36,400
When we have a young child growing up,

949
00:42:36,400 --> 00:42:37,936
we have to treat it right,

950
00:42:37,936 --> 00:42:39,603
so that it grows up into a healthy,

951
00:42:39,603 --> 00:42:41,170
productive citizen of the world.

952
00:42:41,170 --> 00:42:43,790
And if we're bringing AI into existence,

953
00:42:43,790 --> 00:42:46,890
we need to think about how we're treating it well,

954
00:42:46,890 --> 00:42:49,403
so that it can grow into a healthy, productive,

955
00:42:50,753 --> 00:42:51,586
creative citizen of the world.

956
00:42:51,586 --> 00:42:53,930
Micha advocates for AI to be designed

957
00:42:53,930 --> 00:42:58,410
with Christian values in mind, with love and empathy.

958
00:42:58,410 --> 00:43:01,340
I'm here with David Deutsch, an Oxford physicist,

959
00:43:01,340 --> 00:43:04,010
a pioneer in the field of quantum computing.

960
00:43:04,010 --> 00:43:05,760
How would you define life?

961
00:43:05,760 --> 00:43:08,420
What does it mean to be an intelligent being,

962
00:43:08,420 --> 00:43:10,103
what's special about being human?

963
00:43:10,950 --> 00:43:12,830
He actively speaks on this front

964
00:43:12,830 --> 00:43:15,360
hoping to bring tech and religious communities together.

965
00:43:15,360 --> 00:43:17,930
You can always get show notes for this,

966
00:43:17,930 --> 00:43:19,600
and every other episode

967
00:43:19,600 --> 00:43:22,216
at christiantranshumanistpodcast.com.

968
00:43:22,216 --> 00:43:24,640
Thanks so much for listening.

969
00:43:24,640 --> 00:43:27,550
It's tough because a lot of the concepts are abstract,

970
00:43:27,550 --> 00:43:29,950
and there's also religious people

971
00:43:29,950 --> 00:43:32,340
who are just so scared of these things,

972
00:43:32,340 --> 00:43:34,300
and they don't want to talk about it at all.

973
00:43:34,300 --> 00:43:38,190
And so part of what I do is to show people

974
00:43:38,190 --> 00:43:39,840
this probably is going to happen.

975
00:43:39,840 --> 00:43:41,320
We can actually engage in it

976
00:43:41,320 --> 00:43:43,120
and shape it towards something good.

977
00:43:44,670 --> 00:43:46,701
♪ God I'm just a lonely old man ♪

978
00:43:46,701 --> 00:43:47,534
Michael was really nice

979
00:43:47,534 --> 00:43:48,835
and he welcomed me into this home.

980
00:43:48,835 --> 00:43:53,835
What I find really interesting is that he is a pastor' son,

981
00:43:54,420 --> 00:43:56,634
but he's also a software engineer.

982
00:43:56,634 --> 00:43:59,730
And I think what's cool about it

983
00:43:59,730 --> 00:44:02,988
is that he is marrying these two things.

984
00:44:02,988 --> 00:44:05,930
I believe there are a lot of people who are very fearful

985
00:44:05,930 --> 00:44:10,040
of a new world without religion.

986
00:44:10,040 --> 00:44:13,711
So, having this dialogue is useful and interesting.

987
00:44:13,711 --> 00:44:15,970
(upbeat rock music)

988
00:44:15,970 --> 00:44:19,710
AI has become ubiquitous and is part of how we live,

989
00:44:19,710 --> 00:44:22,990
how we work, and soon it could even be part

990
00:44:22,990 --> 00:44:25,986
of how we practice our faith, who knows?

991
00:44:25,986 --> 00:44:28,570
And at every step we wonder

992
00:44:28,570 --> 00:44:30,790
if we can really trust the technology.

993
00:44:30,790 --> 00:44:33,770
It's back to that old hypothesis,

994
00:44:33,770 --> 00:44:36,948
technology is neutral, and it's what humans do with it.

995
00:44:36,948 --> 00:44:40,600
But AI is even trickier, as it learns from the environment

996
00:44:40,600 --> 00:44:42,761
and data that it is exposed to.

997
00:44:42,761 --> 00:44:45,860
And human's inherent bias and discrimination

998
00:44:45,860 --> 00:44:49,300
can easily slip into it if we aren't careful.

999
00:44:49,300 --> 00:44:51,948
Case in point Tay and Compass.

1000
00:44:51,948 --> 00:44:54,380
(cheerful music)

1001
00:44:54,380 --> 00:44:57,877
Can the stick of law, help keep AI from going rogue?

1002
00:44:57,877 --> 00:45:02,513
[Assoc. Prof. Eugene Tan] Law is always playing catch up.

1003
00:45:02,513 --> 00:45:04,113
You know technology races ahead,

1004
00:45:06,281 --> 00:45:07,748
and policymakers, lawmakers,

1005
00:45:07,748 --> 00:45:10,850
always trying to understand how a new technology works,

1006
00:45:10,850 --> 00:45:13,140
Governments around the world and Singapore included,

1007
00:45:13,140 --> 00:45:15,818
you know, you try to develop, you know,

1008
00:45:15,818 --> 00:45:18,480
regulatory sandboxes broad guidelines are laid out,

1009
00:45:18,480 --> 00:45:21,830
you know, certain norms, which must not be pre put in place.

1010
00:45:21,830 --> 00:45:24,750
And so long as you operate within that, sandbox,

1011
00:45:24,750 --> 00:45:27,683
you know, you're free to manifest your creativity, playful.

1012
00:45:27,683 --> 00:45:31,097
So, how far along are we in developing this framework

1013
00:45:31,097 --> 00:45:33,210
for Singapore?

1014
00:45:33,210 --> 00:45:35,283
I think we have started that process.

1015
00:45:36,315 --> 00:45:38,231
The personal data protection commission

1016
00:45:38,231 --> 00:45:39,981
has come up with a paper, you know,

1017
00:45:41,054 --> 00:45:43,090
on how AI ought to be a comfortable in dealing

1018
00:45:43,090 --> 00:45:44,290
with personal data.

1019
00:45:44,290 --> 00:45:46,230
There is also the AI console.

1020
00:45:46,230 --> 00:45:48,490
It is a multi-stakeholder approach

1021
00:45:48,490 --> 00:45:52,285
where governments get these tech companies on board.

1022
00:45:52,285 --> 00:45:54,830
You know, civil society is also involved,

1023
00:45:54,830 --> 00:45:57,210
in what can be done, or what should it be done.

1024
00:45:57,210 --> 00:45:59,023
You know, the sort of OB markers,

1025
00:46:00,097 --> 00:46:00,930
the conscience of society.

1026
00:46:02,095 --> 00:46:03,410
We are putting attention on ethics

1027
00:46:03,410 --> 00:46:05,683
and morality of artificial intelligence.

1028
00:46:07,447 --> 00:46:08,859
We need to keep our eyes on that ball,

1029
00:46:08,859 --> 00:46:10,168
and not lose sight of it.

1030
00:46:10,168 --> 00:46:11,930
(video swooshing)

1031
00:46:11,930 --> 00:46:15,874
It seems like with every new way we use AI,

1032
00:46:15,874 --> 00:46:18,070
it always comes back to this question.

1033
00:46:18,070 --> 00:46:20,253
How do we use it responsibly?

1034
00:46:21,420 --> 00:46:23,730
And who's responsible when something goes wrong.

1035
00:46:23,730 --> 00:46:24,563
Let me tell you,

1036
00:46:24,563 --> 00:46:26,917
there's always no clear cut answer.

1037
00:46:26,917 --> 00:46:30,760
Ethics and morality are complicated enough subjects.

1038
00:46:30,760 --> 00:46:32,594
That's why we've got statues

1039
00:46:32,594 --> 00:46:35,090
of all these philosophers and thinkers.

1040
00:46:35,090 --> 00:46:39,210
Who have been debating this for centuries.

1041
00:46:39,210 --> 00:46:43,980
Add AI into this picture, and it's a lot more complex.

1042
00:46:43,980 --> 00:46:46,600
So I'm kinda glad that we're thinking about it now.

1043
00:46:46,600 --> 00:46:49,660
And hopefully we'll never lose sight of what's right,

1044
00:46:49,660 --> 00:46:53,260
and wrong as we chase this technological dream.

1045
00:46:53,260 --> 00:46:57,169
(playful music ends)

1046
00:46:57,169 --> 00:47:00,420
Hey Siri, how do you use AI responsibly?

1047
00:47:00,420 --> 00:47:01,253
(Siri bleeping)

1048
00:47:01,253 --> 00:47:02,086
I don't understand.

1049
00:47:02,086 --> 00:47:06,426
Hey Siri, why are you a guy? (laughs)

1050
00:47:06,426 --> 00:47:07,259
(Siri bleeping)

1051
00:47:07,259 --> 00:47:08,860
I was not assigned a gender.

1052
00:47:08,860 --> 00:47:11,210
There is a big arms race going on.

1053
00:47:11,210 --> 00:47:14,925
In my next adventure, I visit two AI superpowers.

1054
00:47:14,925 --> 00:47:18,758
(speaks in foreign language)

1055
00:47:20,968 --> 00:47:22,471
The United States,

1056
00:47:22,471 --> 00:47:25,222
it certainly has an excellent infrastructure or ecosystem.

1057
00:47:25,222 --> 00:47:28,796
To find out who will hold the real power over AI.

1058
00:47:28,796 --> 00:47:31,829
(speaks in foreign language)

1059
00:47:31,829 --> 00:47:34,496
(upbeat music)