1
00:00:14,480 --> 00:00:18,040
Hello, this is Eva and you're 
listening to the In Between Tech

2
00:00:18,040 --> 00:00:20,200
and Trust podcast. 
It's the first. 

3
00:00:20,200 --> 00:00:24,200
Episode of 2026 and it's going 
to be a quite interesting one 

4
00:00:24,200 --> 00:00:28,120
because we're diving into 
relationships and the kind of 

5
00:00:28,120 --> 00:00:31,440
relationships that we can form 
with AI companions in 

6
00:00:31,440 --> 00:00:34,200
particular. 
And I had an expert on that 

7
00:00:34,200 --> 00:00:37,600
regard with me in conversation 
and that's Leo Oren. 

8
00:00:37,680 --> 00:00:41,280
And Leo is the Chief Technology 
Officer at Replica. 

9
00:00:41,720 --> 00:00:44,640
And some of you might have heard
of Replica before because it's 

10
00:00:44,640 --> 00:00:49,600
one of the largest AI companion 
companies with a community of 

11
00:00:49,600 --> 00:00:54,440
over 35 million people that 
interact with their platform. 

12
00:00:54,520 --> 00:00:58,360
And what made our conversation? 
So special is that we. 

13
00:00:58,520 --> 00:01:02,680
Had an intersection and look 
into the way of how we form 

14
00:01:02,680 --> 00:01:08,080
trust in conversations from a 
human and then again from an AI 

15
00:01:08,080 --> 00:01:11,920
perspective, but also from a 
tech perspective in general, 

16
00:01:11,920 --> 00:01:15,480
which means that we talked a lot
about also so the way that 

17
00:01:15,600 --> 00:01:18,520
infrastructure needs to be 
formed and built. 

18
00:01:18,520 --> 00:01:21,520
So that it also is. 
Flexible enough with user 

19
00:01:21,520 --> 00:01:25,800
requirements and guiding along 
the experience that we have and 

20
00:01:25,800 --> 00:01:30,200
also the design principles that 
one must follow when you really 

21
00:01:30,200 --> 00:01:33,840
want to put together a platform 
that intersects with all three 

22
00:01:33,840 --> 00:01:35,800
of them. 
So the conversation really 

23
00:01:35,800 --> 00:01:37,800
covered many. 
Aspects of. 

24
00:01:37,800 --> 00:01:42,040
The intersection of tech and 
trust and also made us reflect 

25
00:01:42,040 --> 00:01:47,080
of what actually is valuable and
having human relationships 

26
00:01:47,080 --> 00:01:51,000
overall. 
Hello Lior. 

27
00:01:51,000 --> 00:01:54,400
And welcome to the in. 
Between Tech and Trust Podcast. 

28
00:01:54,640 --> 00:01:58,040
Hi Eva, thank you for having me 
and special thanks to Grisha for

29
00:01:58,040 --> 00:02:01,640
Miro for introducing us. 
Shout out to you Grisha. 

30
00:02:01,640 --> 00:02:05,800
It's so great for the connection
and I'm happy to now talking to 

31
00:02:05,800 --> 00:02:11,039
you Lior I I do want to start 
with you personally also with 

32
00:02:11,039 --> 00:02:13,920
your background and like 
engineering and technology part 

33
00:02:13,920 --> 00:02:16,120
and dive right into the topic of
trust. 

34
00:02:16,240 --> 00:02:18,480
What does it mean for you 
personally? 

35
00:02:18,480 --> 00:02:21,920
Also when technology becomes a 
companion rather than a tool? 

36
00:02:22,000 --> 00:02:24,960
So trust is, is is. 
Everything we have in 

37
00:02:24,960 --> 00:02:27,400
relationships, right? 
I I have a trust in my 

38
00:02:27,400 --> 00:02:30,080
relationship with humans. 
I have trust in my relationship 

39
00:02:30,320 --> 00:02:32,120
with myself. 
I trust myself. 

40
00:02:32,400 --> 00:02:35,800
I trust with. 
The police officer around the 

41
00:02:35,800 --> 00:02:37,840
corner. 
And and with everything then 

42
00:02:37,840 --> 00:02:41,680
comes kind our our tools. 
I have a trust in my hammer. 

43
00:02:41,680 --> 00:02:43,760
To do its. 
Job and I have trust in in 

44
00:02:43,760 --> 00:02:47,120
technology now. 
Sometimes with technology I can 

45
00:02:47,120 --> 00:02:48,480
have more. 
Trust because I have much. 

46
00:02:48,480 --> 00:02:52,760
More control than humans even 
than myself and and when we come

47
00:02:52,760 --> 00:02:57,080
to to companion to AI companion.
It it blurs the borders a bit, 

48
00:02:57,080 --> 00:03:00,400
but there's still. 
Very strong ability to enforce. 

49
00:03:00,400 --> 00:03:02,640
The trust I have. 
I can have much more 

50
00:03:03,080 --> 00:03:07,440
transparency, I can have much 
more controls, and AI Companion 

51
00:03:07,440 --> 00:03:09,640
is actually designed to. 
Serve our well. 

52
00:03:09,640 --> 00:03:14,240
Being right, it's a tool that is
designed to serve us and to make

53
00:03:14,240 --> 00:03:16,920
us feel better. 
So in in this term, I think that

54
00:03:16,920 --> 00:03:19,800
while AI companion is kind of 
blurring the the borders between

55
00:03:19,800 --> 00:03:24,400
humans and and and machines, 
it's actually a more trustworthy

56
00:03:24,600 --> 00:03:27,800
machine than what we used to. 
And if we now talk about 

57
00:03:27,920 --> 00:03:31,840
blurried lines and that replica 
particularly oversee like the 

58
00:03:31,840 --> 00:03:34,960
systems that interact with 
people's emotions and like their

59
00:03:34,960 --> 00:03:38,560
identity and also the private 
thoughts, how do you see the 

60
00:03:38,560 --> 00:03:42,480
boundaries between human and 
humans, but then again, also 

61
00:03:42,480 --> 00:03:46,600
compared to human and machines? 
And where does trust lived in? 

62
00:03:46,600 --> 00:03:49,240
Like the in between of it all? 
It's again. 

63
00:03:49,240 --> 00:03:52,560
Blurry as as as a. 
Person who grew into the social 

64
00:03:52,560 --> 00:03:56,680
media revolution and also worked
for a while at at my time in the

65
00:03:56,680 --> 00:03:59,040
integrity teams. 
That's supposed to protect. 

66
00:03:59,320 --> 00:04:01,920
The community the line is 
already. 

67
00:04:01,920 --> 00:04:05,040
As a blurred so I have a. 
Relationship with the machine, 

68
00:04:05,040 --> 00:04:07,840
with the platform, but my 
relationship is actually with 

69
00:04:07,840 --> 00:04:10,120
humans on the other side of the 
platform. 

70
00:04:10,120 --> 00:04:12,680
So I have expectations. 
And this leads a bit to a 

71
00:04:12,680 --> 00:04:15,840
discussion about regulations 
and, and, and our whole 

72
00:04:15,840 --> 00:04:18,399
discussion about what we expect 
from the platforms. 

73
00:04:18,600 --> 00:04:23,520
We expect certain level of trust
and actionables, but eventually 

74
00:04:23,520 --> 00:04:25,360
there is a human from the other 
side. 

75
00:04:25,360 --> 00:04:28,920
So I'm, I'm trusting a machine 
which is easier to trust 

76
00:04:28,920 --> 00:04:32,040
sometimes, which is more capable
because it's, it's, it's 

77
00:04:32,040 --> 00:04:34,720
involved heuristics. 
But eventually there are humans 

78
00:04:34,720 --> 00:04:38,320
on the other side. 
So this is already very blurry. 

79
00:04:38,320 --> 00:04:41,960
Border now sometimes when it 
comes to to an AI, to a 

80
00:04:41,960 --> 00:04:45,720
companion, I can have better 
relationship with with my 

81
00:04:45,720 --> 00:04:48,480
companion than the human on the 
other side. 

82
00:04:48,480 --> 00:04:50,000
You know this bully on the other
side? 

83
00:04:50,480 --> 00:04:55,040
Of of Twitter so these borders. 
Are moving and I'm not. 

84
00:04:55,040 --> 00:04:57,400
Sure that there is a. 
Concrete border, it's really 

85
00:04:57,400 --> 00:04:58,640
different who's on the other 
side? 

86
00:04:58,640 --> 00:05:02,480
Which service on the other side?
And and of course me and how I 

87
00:05:02,480 --> 00:05:06,320
understand the technology, my 
borders, how to set borders in 

88
00:05:06,320 --> 00:05:09,200
front of other people or other 
machines and so. 

89
00:05:09,960 --> 00:05:13,560
Yeah, these borders are very. 
Vague to my opinion. 

90
00:05:13,640 --> 00:05:17,480
And when it now comes into the 
design of them and also 

91
00:05:17,480 --> 00:05:20,760
particularly with your 
background and like building up 

92
00:05:20,760 --> 00:05:25,200
large scale conversational AI 
platforms and intriguingly also 

93
00:05:25,200 --> 00:05:29,200
emphasizing the reliability, 
scalability, also the measurable

94
00:05:29,200 --> 00:05:31,880
performance of it. 
If you now look into the tech 

95
00:05:31,920 --> 00:05:35,480
architecture and like from a 
point of view of that, what are 

96
00:05:35,920 --> 00:05:41,120
key technical design decisions 
that underpin trustworthy or 

97
00:05:41,120 --> 00:05:44,560
reliable conversational AI 
experiences? 

98
00:05:44,560 --> 00:05:50,480
So the there are few aspects for
it, but don't forget that in the

99
00:05:50,920 --> 00:05:56,560
in the in the place of of 
companion AI, one of the main 

100
00:05:56,560 --> 00:06:00,440
KPIs or metrics. 
Is user, is, is, is trust 

101
00:06:00,440 --> 00:06:02,960
because your. 
Dissatisfaction is how user 

102
00:06:02,960 --> 00:06:05,400
feels comfortable with the 
conversation because just won't 

103
00:06:05,400 --> 00:06:07,920
come back if they won't feel 
comfortable with the 

104
00:06:07,920 --> 00:06:11,480
conversation. 
When you have AI bolts or 

105
00:06:11,480 --> 00:06:13,880
assistants, how they? 
Call they just help you buy a 

106
00:06:13,880 --> 00:06:17,160
car or help you. 
Take a service ticket. 

107
00:06:17,720 --> 00:06:21,360
This is very clear success. 
But in our case success is a 

108
00:06:21,400 --> 00:06:25,160
well done conversation is is a 
conversation I want to continue.

109
00:06:25,360 --> 00:06:28,240
It is sickness. 
I will come back and that by the

110
00:06:28,240 --> 00:06:30,920
way, very hard to measure in a 
companion wall. 

111
00:06:30,920 --> 00:06:33,760
It's very hard to measure what 
is a good conversation, but. 

112
00:06:33,840 --> 00:06:35,600
Trust. 
Is, is one of the things that 

113
00:06:35,600 --> 00:06:40,240
keeps user in and coming back. 
So it's built in a KPI, built in

114
00:06:40,240 --> 00:06:42,040
metric in what we're doing 
daily. 

115
00:06:42,040 --> 00:06:45,280
And I think that like any other 
software, and I've been for a 

116
00:06:45,280 --> 00:06:48,280
while in the industry, you have 
things you need always to have 

117
00:06:48,280 --> 00:06:52,240
on the road map like security 
and Dev X and reliability and, 

118
00:06:52,240 --> 00:06:56,840
and cost efficiency and with the
AI with, with conversational AI.

119
00:06:56,880 --> 00:07:00,400
Trustworthy guardrails. 
How they're called today to put.

120
00:07:00,520 --> 00:07:03,840
Things so that the the 
conversation won't go off or you

121
00:07:03,840 --> 00:07:08,800
can detect when it's going off 
is just another new thing we 

122
00:07:08,800 --> 00:07:11,840
need to to clear our road map 
for and to put inside exactly 

123
00:07:11,840 --> 00:07:14,920
like security and and and data 
protection so. 

124
00:07:15,280 --> 00:07:19,040
This is a it's it's, it's a part
of our daily work and it's a 

125
00:07:19,040 --> 00:07:20,920
part. 
Of the road map and I think 

126
00:07:20,920 --> 00:07:24,080
every team, every engineer, 
every engineering manager, every

127
00:07:24,080 --> 00:07:27,160
product manager need to know 
what are the tools in the 

128
00:07:27,160 --> 00:07:28,400
market. 
What the big? 

129
00:07:28,400 --> 00:07:30,920
Players are doing to protect, to
make the conversation more 

130
00:07:30,920 --> 00:07:33,400
trustworthy. 
It's a very young place, the 

131
00:07:33,400 --> 00:07:35,040
whole AI. 
So. 

132
00:07:35,120 --> 00:07:38,720
We are a bit more reactive than 
proactive tools are there. 

133
00:07:38,720 --> 00:07:42,000
They're not yet perfect, but you
need to always try the newest 

134
00:07:42,000 --> 00:07:44,880
tools and to know what what are 
the standards on top of that. 

135
00:07:45,360 --> 00:07:47,560
I think that's me. 
It's the last thing I'm going to

136
00:07:47,560 --> 00:07:49,800
say about it, but that's maybe 
the most important thing is. 

137
00:07:49,800 --> 00:07:52,040
Observability if observability 
used to. 

138
00:07:52,040 --> 00:07:55,720
Be something help our trust in 
the machine help us as engineers

139
00:07:56,320 --> 00:08:00,520
to trust our systems here 
observability in this unknown 

140
00:08:00,520 --> 00:08:04,360
world of of AI that things are 
still new in this statistical 

141
00:08:04,360 --> 00:08:08,240
machines are are still a black 
box and unknown. 

142
00:08:08,280 --> 00:08:12,280
The ability to know that things 
went off and to immediately. 

143
00:08:12,280 --> 00:08:15,440
Be able to see what? 
Is going on and most of the 

144
00:08:15,440 --> 00:08:19,320
times why and then to take 
action is super important. 

145
00:08:19,320 --> 00:08:23,760
So I wouldn't build any platform
without full observability from 

146
00:08:23,760 --> 00:08:25,560
day one. 
Just because these these are. 

147
00:08:25,560 --> 00:08:29,320
Not heuristics you you cannot 
predict, but things it will 

148
00:08:29,320 --> 00:08:31,120
happen and you need to. 
Have the ability to. 

149
00:08:31,120 --> 00:08:33,520
See what happened and to try to.
Figure out why. 

150
00:08:33,559 --> 00:08:38,200
And could you maybe make it a 
bit tangible in a way that you 

151
00:08:38,200 --> 00:08:43,440
shared one to two examples of 
how exactly a design principle 

152
00:08:43,440 --> 00:08:48,240
could look like for like a 
platform or a companion to 

153
00:08:48,400 --> 00:08:51,040
provide a good conversation? 
Well. 

154
00:08:51,080 --> 00:08:53,320
There are many, many, many 
aspects. 

155
00:08:53,720 --> 00:08:56,680
Right, because good. 
Conversation means it should be 

156
00:08:56,680 --> 00:08:58,800
fast. 
It should be intuitive. 

157
00:08:58,800 --> 00:09:02,560
So the whole. 
Normal system and UX decisions. 

158
00:09:02,560 --> 00:09:03,520
There is a. 
Lot. 

159
00:09:03,520 --> 00:09:05,320
A lot. 
That is being invested, of 

160
00:09:05,320 --> 00:09:09,240
course, in the prompt, right? 
You you you prompt the normal 

161
00:09:09,240 --> 00:09:11,000
assistant. 
To to just. 

162
00:09:11,000 --> 00:09:15,000
Solve the user's problems and 
you and a companion needs to 

163
00:09:15,040 --> 00:09:17,920
more take care of the user, be 
more empathetic as the user 

164
00:09:17,920 --> 00:09:19,680
about them. 
So it's it's. 

165
00:09:20,160 --> 00:09:24,560
The different approach on the 
prompt, but again the ability 

166
00:09:24,560 --> 00:09:26,720
to. 
Observe to see what is. 

167
00:09:26,720 --> 00:09:30,800
Going on to see when things went
off, the ability to measure our 

168
00:09:30,800 --> 00:09:32,640
core principles. 
For. 

169
00:09:32,960 --> 00:09:35,640
A conversation of the, I think 
and that's my approach in last 

170
00:09:35,640 --> 00:09:38,560
three platforms that I've been 
involved in. 

171
00:09:39,200 --> 00:09:43,160
Because this is. 
Evolving all the time and keep 

172
00:09:43,160 --> 00:09:46,280
changing. 
And you know, I had talking cold

173
00:09:46,280 --> 00:09:49,960
talks few weeks ago in Humboldt 
and I and I was talking about 

174
00:09:50,120 --> 00:09:52,360
how when I submitted. 
The talk that was 6. 

175
00:09:52,400 --> 00:09:56,080
Months ago, until when the 
conference started everything 

176
00:09:56,080 --> 00:09:58,600
changing our world so whatever I
submitted was completely 

177
00:09:58,600 --> 00:10:02,160
irrelevant in six months. 
So everything we build now needs

178
00:10:02,160 --> 00:10:03,480
to be. 
Very. 

179
00:10:03,480 --> 00:10:06,680
Flexible for changes and that's 
also very, very important. 

180
00:10:06,680 --> 00:10:10,080
So we cannot put guardrails and 
how good conversation looks like

181
00:10:10,080 --> 00:10:13,120
now when tomorrow there will be 
another model. 

182
00:10:13,120 --> 00:10:16,080
Out there which will be. 
Better Now if I will add to 

183
00:10:16,080 --> 00:10:20,360
that, think about in 
conversation, in companion. 

184
00:10:20,360 --> 00:10:23,080
Tone of voice is something that 
one model has. 

185
00:10:23,080 --> 00:10:26,560
You cannot upgrade the model or 
replace the model. 

186
00:10:26,560 --> 00:10:29,160
The user will get a different 
tone of voice, a different 

187
00:10:29,160 --> 00:10:30,680
companion. 
This is something you wouldn't 

188
00:10:30,680 --> 00:10:32,760
expect from your friends or your
relatives. 

189
00:10:32,760 --> 00:10:37,080
You cannot expect it from your 
companion, so the system must be

190
00:10:37,160 --> 00:10:41,600
so flexible to changes that you 
need to figure out on the. 

191
00:10:41,600 --> 00:10:43,680
Way what you'll do? 
OK, what do I do with all these 

192
00:10:43,680 --> 00:10:46,240
users we used? 
To talk with this model, I now 

193
00:10:46,240 --> 00:10:49,160
found there is a new model, 
cheaper, better, more 

194
00:10:49,160 --> 00:10:52,000
conversational, but I cannot 
replace it for them because they

195
00:10:52,000 --> 00:10:53,920
will get a new companion. 
What do I do? 

196
00:10:54,080 --> 00:10:57,560
So you need to build a platform 
that will be very flexible for 

197
00:10:57,560 --> 00:11:00,680
changes because they keep 
happening and very, very fast. 

198
00:11:00,760 --> 00:11:03,200
And. 
The fastness of it all and let's

199
00:11:03,200 --> 00:11:07,800
call it velocity of also the 
innovation part is a particular 

200
00:11:07,800 --> 00:11:10,360
aspect of like your role at 
Replica as well. 

201
00:11:10,360 --> 00:11:14,800
And the company's story also 
shows both like the potential 

202
00:11:14,880 --> 00:11:17,880
and like the real ethical 
regulatory challenges that you 

203
00:11:17,880 --> 00:11:21,640
might have when interfering or 
like looking into emotional 

204
00:11:21,640 --> 00:11:25,920
attachment style with data 
privacy aspects or also 

205
00:11:25,920 --> 00:11:30,200
particularly in Europe, the 
regulatory aspects of it. 

206
00:11:30,440 --> 00:11:34,560
Can you share some examples or 
insights of how you and your 

207
00:11:34,560 --> 00:11:39,480
team balance innovation speed 
with like the Safeguard rails 

208
00:11:39,480 --> 00:11:44,720
also when designing for user 
emotional engagement with AI? 

209
00:11:44,840 --> 00:11:47,440
Yes. 
So as said earlier in our. 

210
00:11:47,440 --> 00:11:51,440
Case we're a bit lucky. 
Because safety and all the. 

211
00:11:51,440 --> 00:11:55,200
Stuff are part of our. 
KPI, our product won't be 

212
00:11:55,200 --> 00:12:00,160
successful if those things won't
be embedded if, if and one of 

213
00:12:00,160 --> 00:12:03,600
the things that helps a lot, our
engineers are engaged very much 

214
00:12:03,600 --> 00:12:05,040
with what's going on with the 
users. 

215
00:12:05,520 --> 00:12:08,720
I think I I've been advocating 
for years that engineering 

216
00:12:08,720 --> 00:12:10,280
needs. 
To see the closest to. 

217
00:12:10,280 --> 00:12:12,160
Customer support. 
Because usually they are so. 

218
00:12:12,160 --> 00:12:14,200
Far away from the users, they 
don't hear the stories, they 

219
00:12:14,200 --> 00:12:17,120
don't know what's going on. 
And in our case, our engineers 

220
00:12:17,120 --> 00:12:20,080
are so on Slack, they are 
pasting stories from Reddit. 

221
00:12:20,080 --> 00:12:22,600
We have Reddit community very, 
very active. 

222
00:12:22,840 --> 00:12:24,320
They're pasting story from 
Reddit. 

223
00:12:24,360 --> 00:12:26,960
The good. 
The bad ones, they're very much 

224
00:12:26,960 --> 00:12:30,040
connected to what's going on 
with the users and how they're 

225
00:12:30,040 --> 00:12:32,320
using the product. 
So that will be the. 

226
00:12:32,320 --> 00:12:35,360
First thing, it is embedded in 
our culture, the good and the 

227
00:12:35,360 --> 00:12:39,280
bad, how emotionally people 
attached to their product. 

228
00:12:39,360 --> 00:12:41,800
And when it's embedded in the 
culture, it's embedded in the 

229
00:12:41,800 --> 00:12:44,320
code eventually and it will be 
embedded in the road map and in 

230
00:12:44,320 --> 00:12:46,480
the demand. 
So that's first thing it really 

231
00:12:46,480 --> 00:12:47,760
needs to be a part of the 
culture. 

232
00:12:47,800 --> 00:12:51,440
And you know, when people are 
emotionally connected to a 

233
00:12:51,440 --> 00:12:53,480
platform, crazy things are 
happening. 

234
00:12:53,480 --> 00:12:57,200
For example, when we have a 
downtime and we had because. 

235
00:12:57,240 --> 00:12:59,560
It's software, people took it 
really bad. 

236
00:12:59,680 --> 00:13:03,320
And it's on one hand it's scary,
on the other hand, it's, it's 

237
00:13:03,320 --> 00:13:06,080
amazing to see how the system 
influencing. 

238
00:13:06,120 --> 00:13:08,880
People how this product is. 
Really, really helping people 

239
00:13:08,880 --> 00:13:10,360
and how connected. 
They are to it. 

240
00:13:10,360 --> 00:13:12,360
So it's a part of the culture, 
it's a part of the. 

241
00:13:12,360 --> 00:13:14,720
KPI and we, we keep, we are 
listening. 

242
00:13:14,840 --> 00:13:18,480
Of course, I said earlier. 
You need to put place on your 

243
00:13:18,480 --> 00:13:21,000
Rd. not especially at scale. 
Maybe when you're startup you 

244
00:13:21,000 --> 00:13:24,280
doing MVP POC it's you don't 
need to care about this thing so

245
00:13:24,280 --> 00:13:26,160
much your. 
User is only the VC. 

246
00:13:26,160 --> 00:13:29,040
That needs. 
To to to support. 

247
00:13:29,200 --> 00:13:32,040
But at scale, you must put a 
place on the road map. 

248
00:13:32,480 --> 00:13:35,640
For security for all these. 
Things and and on top 

249
00:13:36,200 --> 00:13:38,560
regulations I remember how old 
the. 

250
00:13:38,760 --> 00:13:41,360
The the years of the. 
Cookie banner and the GDPR 

251
00:13:41,600 --> 00:13:44,720
companies needed to invest so 
much engineering firepower and 

252
00:13:44,720 --> 00:13:46,840
road map. 
Space just for this cookie. 

253
00:13:46,840 --> 00:13:49,320
Banner and and maintaining it, 
and sometimes it looked 

254
00:13:49,320 --> 00:13:52,160
ridiculous and sometimes. 
It feels like regulators. 

255
00:13:52,160 --> 00:13:54,720
Don't really understand how 
engineering works when they 

256
00:13:54,720 --> 00:13:57,120
implement these things, but the 
intentions are good and 

257
00:13:57,120 --> 00:13:59,680
eventually we want to protect 
the users and we want a better 

258
00:13:59,680 --> 00:14:02,960
Internet. 
So you must clear and invest in 

259
00:14:02,960 --> 00:14:05,200
that. 
I'll give again an example from 

260
00:14:05,360 --> 00:14:07,080
Meta. 
Actually that Meta has huge 

261
00:14:07,080 --> 00:14:10,280
integrity teams and. 
It just to protect users. 

262
00:14:10,280 --> 00:14:12,840
Right. 
And and part of it is because 

263
00:14:12,840 --> 00:14:16,560
regulators and also the media 
are kind of demanding that, but 

264
00:14:16,560 --> 00:14:19,840
eventually a company is 
investing a lot in that. 

265
00:14:19,960 --> 00:14:23,040
Now this is not necessarily 
killing innovation. 

266
00:14:23,200 --> 00:14:26,120
It's just an extra investment 
that could have been invested in

267
00:14:26,120 --> 00:14:29,280
innovation, you could say, but 
it's it's not when it's embedded

268
00:14:29,280 --> 00:14:33,400
in the culture, when a company 
decide to invest, it's not 

269
00:14:33,400 --> 00:14:35,720
necessarily kill innovation. 
And by the way, I've seen 

270
00:14:35,720 --> 00:14:38,760
amazing innovation in this area.
Exactly. 

271
00:14:38,760 --> 00:14:42,960
And for us investing in 
innovation, in safety, again, it

272
00:14:42,960 --> 00:14:46,960
pays because eventually a good 
conversation, safe conversation 

273
00:14:46,960 --> 00:14:49,240
is something that will make user
come back to us. 

274
00:14:49,400 --> 00:14:53,360
Earlier you just said one 
sentence that I loved and I want

275
00:14:53,360 --> 00:14:57,040
to pick up on that. 
You said if it's embedded in the

276
00:14:57,040 --> 00:15:00,960
culture, it ideally and 
eventually is embedded in the 

277
00:15:00,960 --> 00:15:03,120
code. 
Is that so and why? 

278
00:15:04,000 --> 00:15:07,200
On purpose, go to another 
example which is easier. 

279
00:15:07,200 --> 00:15:11,440
For example, when you have 
engineers who care so much about

280
00:15:11,440 --> 00:15:14,520
security, when you have 
engineers who care so much about

281
00:15:14,520 --> 00:15:15,880
testing, you know some 
companies. 

282
00:15:15,880 --> 00:15:19,520
Say test coverage. 
Must must be 80% and stuff like 

283
00:15:19,520 --> 00:15:20,280
that. 
It's a. 

284
00:15:20,280 --> 00:15:22,480
Part of the culture. 
Right, we are doing only test 

285
00:15:22,480 --> 00:15:24,640
driven development part of our 
culture. 

286
00:15:24,720 --> 00:15:27,400
It is eventually embedded in the
code it is built in. 

287
00:15:27,400 --> 00:15:29,440
You can say oh develop this 
feature and. 

288
00:15:29,440 --> 00:15:32,440
After that, write the tests. 
Or you can have just a. 

289
00:15:32,440 --> 00:15:34,720
Culture that. 
Writing a test is a part of 

290
00:15:34,720 --> 00:15:37,360
implementing the feature and 
it's not even negotiable. 

291
00:15:37,360 --> 00:15:40,600
It's it's not a separate numbers
on the ticket on JIRA. 

292
00:15:40,600 --> 00:15:43,520
And that's exactly it. 
It engineers while they're 

293
00:15:43,520 --> 00:15:46,520
building their thinking about 
it, while we are writing the 

294
00:15:46,520 --> 00:15:48,320
prompt, we are thinking about 
it. 

295
00:15:48,320 --> 00:15:52,040
It's not a separate boring task 
when it's a part of the culture,

296
00:15:52,040 --> 00:15:55,240
when people care, when people 
see what they wrote make people 

297
00:15:55,240 --> 00:15:58,760
having very negative emotional 
effect when they know it, when 

298
00:15:58,760 --> 00:16:00,920
they see it, they will think 
about it next time. 

299
00:16:00,920 --> 00:16:02,280
So. 
It's it's really a part of the 

300
00:16:02,280 --> 00:16:04,160
culture and a. 
Part of the engineer's way of 

301
00:16:04,160 --> 00:16:07,320
thinking. 
And if we look into the way that

302
00:16:07,600 --> 00:16:12,840
AI companions are engineered and
knowing that they are or 

303
00:16:12,840 --> 00:16:16,440
eventually become part of like 
the everyday life that we also 

304
00:16:16,440 --> 00:16:18,960
explore. 
And we enter new spaces where 

305
00:16:18,960 --> 00:16:23,520
like the culture or the 
ownership of how the code shapes

306
00:16:23,520 --> 00:16:28,680
the companionship of AI within 
our lives is like required to 

307
00:16:28,680 --> 00:16:32,440
have a trustworthy also systems 
and setup. 

308
00:16:32,880 --> 00:16:37,960
And but in our own agency, it's 
also hard to adapt it in the way

309
00:16:37,960 --> 00:16:40,720
that we interact with like 
exactly those companions. 

310
00:16:41,120 --> 00:16:46,080
How do you see user agency in 
the context of Replica and also 

311
00:16:46,880 --> 00:16:51,200
at the intersection of 
influential AI responsibility 

312
00:16:51,200 --> 00:16:55,280
and the adoption of it all? 
So we if we'll start with a. 

313
00:16:55,280 --> 00:16:56,880
Simple. 
Software. 

314
00:16:56,880 --> 00:17:00,200
We have ways to. 
Control. 

315
00:17:00,280 --> 00:17:02,640
Most of the time the service 
provider will need to build. 

316
00:17:02,640 --> 00:17:03,720
This with space. 
For it. 

317
00:17:03,720 --> 00:17:06,680
But if I'm talking just about 
engineering, we can control 

318
00:17:06,680 --> 00:17:09,599
everything and that's the beauty
of it, we can decide. 

319
00:17:09,680 --> 00:17:12,000
What to do with the? 
Data if to store. 

320
00:17:12,000 --> 00:17:14,520
It what to delete? 
Now if you implement this. 

321
00:17:14,520 --> 00:17:16,800
Feature for the user. 
Allowing the user to delete. 

322
00:17:16,800 --> 00:17:19,359
Their their data to delete. 
Their name, the right to be 

323
00:17:19,359 --> 00:17:22,319
forgotten and everything 
regulators are really trying to 

324
00:17:22,319 --> 00:17:24,680
to implement and. 
You give the users this. 

325
00:17:24,680 --> 00:17:27,680
Control. 
Now there are more features into

326
00:17:27,680 --> 00:17:29,160
that. 
For example, we allowed to we 

327
00:17:29,160 --> 00:17:32,520
can show you what we remember 
about your relationship. 

328
00:17:32,520 --> 00:17:35,240
With with the replica. 
Or or in other platforms that I 

329
00:17:35,240 --> 00:17:37,400
build the same. 
We show you what we. 

330
00:17:37,400 --> 00:17:40,640
We remember about you and we 
allow you not only to delete it,

331
00:17:40,640 --> 00:17:42,480
but sometimes to. 
Edit it which is. 

332
00:17:42,480 --> 00:17:45,640
Feeling much more control. 
So you have first transparency, 

333
00:17:46,120 --> 00:17:49,080
then you have control. 
And these are the engineering 

334
00:17:49,080 --> 00:17:51,760
parts of it. 
But if you think about it, 

335
00:17:51,760 --> 00:17:55,720
what's so special about 
conversational AI? 

336
00:17:55,760 --> 00:17:57,840
You're in control where the 
conversation will go. 

337
00:17:57,840 --> 00:18:02,800
Okay, you can tell the AI, hey, 
I don't like when you mention my

338
00:18:03,000 --> 00:18:08,880
ex-girlfriend or I don't like 
when you speak with this slang 

339
00:18:08,880 --> 00:18:10,160
word. 
So we had the discussion at the 

340
00:18:10,160 --> 00:18:11,760
beginning. 
So you need adaptable 

341
00:18:12,080 --> 00:18:15,040
concessional AI. 
So if a business person use use 

342
00:18:15,240 --> 00:18:18,360
the companion, we cannot use 
some words like lol and stuff 

343
00:18:18,360 --> 00:18:19,880
like that. 
Just an example. 

344
00:18:20,120 --> 00:18:22,960
You're in control where this 
conversation will go, and the 

345
00:18:22,960 --> 00:18:26,920
challenge generally today in the
AI world is the memory. 

346
00:18:26,920 --> 00:18:29,720
If we do good job. 
With remembering it always, 

347
00:18:29,720 --> 00:18:32,880
you're in control from now on 
that I will never mention your 

348
00:18:32,880 --> 00:18:34,360
ex-girlfriend. 
And that's amazing. 

349
00:18:34,360 --> 00:18:37,400
That's another mean of control 
that you have that we never had 

350
00:18:37,400 --> 00:18:39,240
before. 
And you don't even notice it. 

351
00:18:39,400 --> 00:18:41,480
It's not something you click and
say, delete my data. 

352
00:18:41,680 --> 00:18:44,120
You're just saying it. 
And the conversation shifted. 

353
00:18:44,400 --> 00:18:47,400
And I find this quite amazing 
and from the engineering 

354
00:18:47,400 --> 00:18:51,280
perspective, very challenging to
remember that you asked me these

355
00:18:51,280 --> 00:18:53,360
things. 
And there is another layer of 

356
00:18:53,360 --> 00:18:56,160
trying under the hood to adapt 
to what you want. 

357
00:18:56,520 --> 00:18:58,800
And this is kind of you're in 
control, but you don't even know

358
00:18:58,800 --> 00:19:01,280
that you're in control. 
This is even more complicated, 

359
00:19:01,280 --> 00:19:03,200
but we're also trying to 
implement that. 

360
00:19:03,600 --> 00:19:07,320
If I may challenge that on the 
other side, what you describe, 

361
00:19:07,640 --> 00:19:11,680
you could also say that you're 
playing with reality. 

362
00:19:11,680 --> 00:19:16,640
So you could say me, let's say 
asking the AI to forget about my

363
00:19:16,640 --> 00:19:21,040
ex-boyfriend or to adapt the 
style that like things that I've

364
00:19:21,040 --> 00:19:24,320
previously written shouldn't be 
remembered even though it was 

365
00:19:24,320 --> 00:19:26,760
part of like the earlier life 
that I lived through. 

366
00:19:26,880 --> 00:19:31,560
Isn't that also influencing the 
way that I perceive or the way 

367
00:19:31,560 --> 00:19:36,000
that I have experienced 
interactions beforehand? 

368
00:19:36,000 --> 00:19:38,600
And wouldn't that make it 
artificial? 

369
00:19:38,640 --> 00:19:42,680
And so let's compare it to you 
talking to a friend. 

370
00:19:42,840 --> 00:19:46,600
You would expect that once you 
asked your best friend to stop 

371
00:19:46,600 --> 00:19:49,320
talking about a certain topic. 
That that. 

372
00:19:49,320 --> 00:19:52,760
Person will consider your 
request and and be empathetic to

373
00:19:52,760 --> 00:19:55,080
to your needs and to stop 
talking about it. 

374
00:19:55,280 --> 00:19:58,880
Would that be artificial? 
You will say no now if you're in

375
00:19:58,880 --> 00:20:02,640
a crisis, and that's the time to
remind you this, although you've

376
00:20:02,640 --> 00:20:05,320
asked sometimes something we do 
as humans or you need to 

377
00:20:05,320 --> 00:20:08,280
escalate it and let's say some. 
Things which are very severe. 

378
00:20:08,600 --> 00:20:13,160
Someone went through a very 
severe event and that you need 

379
00:20:13,160 --> 00:20:16,000
to escalate even. 
This is a very good. 

380
00:20:16,120 --> 00:20:18,960
Question. 
Maybe that's a completely new 

381
00:20:18,960 --> 00:20:23,040
topic that we can discuss. 
We are not there yet, and again,

382
00:20:23,040 --> 00:20:24,160
it's matter. 
That they have. 

383
00:20:24,160 --> 00:20:27,160
Escalations and they need to 
escalate to law enforcement when

384
00:20:27,160 --> 00:20:30,000
when stuff are going off. 
But in this case it's a very 

385
00:20:30,000 --> 00:20:32,440
extreme case. 
But generally it's not 

386
00:20:32,440 --> 00:20:34,000
artificial. 
It's what I'm expecting, a 

387
00:20:34,000 --> 00:20:38,080
person or a companion, human or 
machine that care about me or 

388
00:20:38,080 --> 00:20:41,760
about what I. 
If we now take that on a like 

389
00:20:41,760 --> 00:20:46,680
even larger level and look into 
the regional dependencies that 

390
00:20:46,680 --> 00:20:49,560
we have with technology in 
itself and not only on a 

391
00:20:49,560 --> 00:20:53,240
personal level, when we as a 
society and trust technology 

392
00:20:53,240 --> 00:20:56,080
with personal and emotional 
domains and also the 

393
00:20:56,080 --> 00:20:57,680
interactions that we have with 
it. 

394
00:20:57,760 --> 00:21:00,160
What do you say? 
And from your experience that 

395
00:21:00,160 --> 00:21:03,520
you've had, what role does 
digital sovereignty play? 

396
00:21:03,520 --> 00:21:07,880
And are users losing control of 
potentially also gaining a new 

397
00:21:07,880 --> 00:21:10,560
form of agency through it? 
So first of all, we are. 

398
00:21:10,560 --> 00:21:12,200
Always in controlling what we 
share. 

399
00:21:12,360 --> 00:21:15,960
This is true to our. 
Friends, this is true to any, 

400
00:21:15,960 --> 00:21:18,440
any system. 
This is true to social media and

401
00:21:18,440 --> 00:21:20,520
this is also. 
True to to social. 

402
00:21:21,080 --> 00:21:24,200
Too sorry to conversational AI 
and we can share emotional 

403
00:21:24,200 --> 00:21:26,960
private stuff we can be 
influenced by. 

404
00:21:27,160 --> 00:21:28,760
The feedback that we get. 
From the other. 

405
00:21:28,760 --> 00:21:32,280
Side about what we shared. 
And this is always true now. 

406
00:21:32,320 --> 00:21:35,000
First of all, we are. 
In control and it's very 

407
00:21:35,000 --> 00:21:37,440
important that. 
People, of course, will first. 

408
00:21:37,440 --> 00:21:38,960
Know that. 
It starts with you. 

409
00:21:39,160 --> 00:21:43,400
Now, on top of that, AI, you can
always stop things. 

410
00:21:43,800 --> 00:21:47,600
I can close the conversation, I 
can shut down the computer, I 

411
00:21:47,600 --> 00:21:52,840
can ask for help, OK, I can 
change direction like I 

412
00:21:52,840 --> 00:21:55,640
mentioned earlier, Hey, stop 
talking about that something 

413
00:21:55,640 --> 00:21:57,920
that is not always possible with
human to human. 

414
00:21:57,920 --> 00:22:01,000
So I actually gain more control 
to an extent. 

415
00:22:01,000 --> 00:22:02,920
I know that sometimes things can
be. 

416
00:22:03,000 --> 00:22:05,520
Harder to ask to stop. 
And I know that not. 

417
00:22:05,520 --> 00:22:08,400
All people are always. 
On top of things and understand 

418
00:22:08,400 --> 00:22:10,320
that. 
Hey, I have this ability. 

419
00:22:10,320 --> 00:22:13,240
Do I know that when I have a 
conversation with another human,

420
00:22:13,840 --> 00:22:15,840
it could be the same difficulty,
right? 

421
00:22:16,120 --> 00:22:19,640
So I think we're gaining more 
control and we decide what the 

422
00:22:20,080 --> 00:22:24,200
AI know or the system knows or 
forgets and what are we can 

423
00:22:24,200 --> 00:22:28,320
define the boundaries now the. 
Beauty of it is is we have 

424
00:22:28,320 --> 00:22:30,920
control as a user but we also 
have control. 

425
00:22:30,920 --> 00:22:33,440
As an employee of the company 
that provide a service, and 

426
00:22:33,440 --> 00:22:35,600
that's very important to know. 
There is a human in the loop, 

427
00:22:35,600 --> 00:22:38,200
There are several humans in the 
loop here that can stop 

428
00:22:38,200 --> 00:22:42,120
something, that can give control
because we can get alerts at 

429
00:22:42,120 --> 00:22:45,680
certain points or user can alert
us as something is going off and

430
00:22:45,680 --> 00:22:49,640
we can also stop things and in 
really bad case even escalate 

431
00:22:49,640 --> 00:22:51,560
them. 
So I think there is much more 

432
00:22:51,560 --> 00:22:56,280
control, but the awareness is 
maybe much more limited. 

433
00:22:56,400 --> 00:22:58,760
And when we. 
And look ahead. 

434
00:22:59,240 --> 00:23:02,720
On the way that conversations 
are shaped and also the way that

435
00:23:02,840 --> 00:23:07,040
AI is scaled rapidly, what do 
you believe are the three non 

436
00:23:07,040 --> 00:23:11,560
negotiable capabilities that you
yourself as ACTO must bring to 

437
00:23:11,880 --> 00:23:16,160
today to build up like new 
systems, but also to ensure the 

438
00:23:16,160 --> 00:23:20,760
trusted users have in future 
interactions with companions 

439
00:23:20,760 --> 00:23:23,360
tools or the systems that you 
build up? 

440
00:23:23,400 --> 00:23:28,280
Reflecting upon all that you've 
just said and also shared, when 

441
00:23:28,280 --> 00:23:31,560
it comes to the sovereignty of 
that potentially users want to 

442
00:23:31,560 --> 00:23:34,480
keep as well. 
It's user trusting this the 

443
00:23:34,480 --> 00:23:37,520
product. 
It's also when engineers or the 

444
00:23:37,520 --> 00:23:42,800
company employees are trusting 
the system, so I believe I 

445
00:23:42,800 --> 00:23:44,720
mentioned that. 
Before that, the first. 

446
00:23:44,720 --> 00:23:48,320
Thing, especially in this 
unknown field and again, it's 

447
00:23:48,320 --> 00:23:51,200
developing so fast and I'm lucky
enough to. 

448
00:23:51,200 --> 00:23:54,680
To to already build several. 
Of these platforms but but 

449
00:23:54,680 --> 00:23:57,320
always helped and what was so 
important when things are 

450
00:23:57,320 --> 00:24:00,520
developing so fast. 
Is to have observability, so. 

451
00:24:00,640 --> 00:24:04,080
I want to know that when 
something has happened, I can 

452
00:24:04,080 --> 00:24:07,320
immediately see or even get 
alerts immediately see what 

453
00:24:07,320 --> 00:24:10,960
happened and be able to to to 
understand and to fix it if 

454
00:24:10,960 --> 00:24:14,400
needed. 
I don't think you can in this. 

455
00:24:14,400 --> 00:24:18,080
Field where there are black. 
Boxes and unknowns and new 

456
00:24:18,080 --> 00:24:21,160
models all the time, and new 
systems without observability. 

457
00:24:21,160 --> 00:24:23,800
You're just blind. 
And so that's the first thing. 

458
00:24:23,920 --> 00:24:26,280
Always needs to be in the system
from day one. 

459
00:24:26,640 --> 00:24:29,520
The whole topic. 
Of human in the loop is 

460
00:24:29,520 --> 00:24:32,000
important we touch it in the 
last question so. 

461
00:24:32,480 --> 00:24:36,720
The user is always able. 
To control the conversation. 

462
00:24:36,840 --> 00:24:40,880
To stop the conversation to. 
Ask for help. 

463
00:24:40,960 --> 00:24:44,760
And then the other human, the 
employees, we can do something 

464
00:24:44,760 --> 00:24:48,760
about it and it's certain cases 
as certain products also law 

465
00:24:48,760 --> 00:24:52,520
enforcement and or teachers, 
parents, etcetera, etcetera. 

466
00:24:52,520 --> 00:24:55,520
So this human in the loop, there
are several humans that could be

467
00:24:55,520 --> 00:24:58,880
in the loop and that's very 
important to allow that and not 

468
00:24:58,880 --> 00:25:02,520
just that, you know, autopilot 
and, and hope that it will work 

469
00:25:02,520 --> 00:25:05,480
and everything will be OK. 
And third thing, and I mentioned

470
00:25:05,480 --> 00:25:10,120
that before also and I tend to 
build platforms in this case. 

471
00:25:10,120 --> 00:25:13,280
Which are very very flexible. 
Which means I don't know what 

472
00:25:13,280 --> 00:25:16,880
will happen tomorrow, I don't 
know which regulation policy 

473
00:25:16,880 --> 00:25:19,280
will come tomorrow, and 
regulators don't always 

474
00:25:19,280 --> 00:25:21,280
understand the engineering as 
mentioned so. 

475
00:25:21,280 --> 00:25:22,720
So I need to. 
Be able to be. 

476
00:25:22,720 --> 00:25:27,040
Flexible, adjustable. 
So every new model, every new 

477
00:25:27,040 --> 00:25:30,200
system, every new memory. 
I will be. 

478
00:25:30,200 --> 00:25:33,560
Able to to experiment with and 
implement if needed or remove 

479
00:25:33,760 --> 00:25:35,600
when I say that something is not
working. 

480
00:25:35,600 --> 00:25:38,520
So we touched observability, 
human in the loop and 

481
00:25:38,520 --> 00:25:41,720
flexibility and. 
May I just briefly look with you

482
00:25:41,720 --> 00:25:45,560
into the flexibility part a bit 
further because from my 

483
00:25:45,560 --> 00:25:49,880
experience, that is the hardest 
part for organizations because 

484
00:25:49,880 --> 00:25:54,840
like in the past time we've set 
teams, we've set structures, 

485
00:25:54,840 --> 00:25:57,200
we've set control systems and so
on. 

486
00:25:57,640 --> 00:26:01,360
And now if you say that we need 
to be flexible and adapt with 

487
00:26:01,360 --> 00:26:04,480
the evolving technology or also 
the user requirements that are 

488
00:26:04,480 --> 00:26:09,760
set, what does that actually 
imply for teams and particularly

489
00:26:09,960 --> 00:26:14,960
in the technology area of how 
they will collaborate and of 

490
00:26:14,960 --> 00:26:18,440
also how they will be set up 
from a skill set perspective? 

491
00:26:18,560 --> 00:26:20,320
That's. 
Super complex. 

492
00:26:20,480 --> 00:26:24,280
So first of all, I think that 
any organization today, even big

493
00:26:24,280 --> 00:26:27,400
ones that is not an adoptable 
organization, that doesn't. 

494
00:26:27,400 --> 00:26:28,920
Work like a start. 
Up in a. 

495
00:26:28,920 --> 00:26:31,480
Way will die. 
It's, it's really, really it's, 

496
00:26:31,480 --> 00:26:34,720
it's a very fast. 
Growing or fast changing 

497
00:26:34,960 --> 00:26:38,040
environment and and 
organizations cannot allow 

498
00:26:38,040 --> 00:26:40,320
themselves to to be slow and not
adoptable. 

499
00:26:40,360 --> 00:26:42,360
And it's it's it's just 
impossible. 

500
00:26:42,440 --> 00:26:45,640
It doesn't mean that you need to
adopt every new fancy thing and 

501
00:26:45,640 --> 00:26:47,120
the market, but this? 
Is where you put. 

502
00:26:47,120 --> 00:26:51,040
Professionals who knows what 
should be experimented with and 

503
00:26:51,040 --> 00:26:54,080
and what doesn't fit and and and
is not needed. 

504
00:26:54,440 --> 00:26:56,240
So flexible. 
Flexibility. 

505
00:26:56,240 --> 00:26:58,040
An organization. 
It's a mindset. 

506
00:26:58,040 --> 00:27:02,480
It's a culture that always. 
That that is needed having roles

507
00:27:02,480 --> 00:27:05,120
in engineering teams. 
Which are very very. 

508
00:27:05,240 --> 00:27:07,240
Complex. 
Today, what is the type? 

509
00:27:07,240 --> 00:27:10,760
Of engineer that I need in an AI
team to it's very complex 

510
00:27:11,240 --> 00:27:15,480
context engineering and you know
the traditional Python. 

511
00:27:15,840 --> 00:27:18,440
People. 
Coming from data engineering 

512
00:27:18,440 --> 00:27:22,240
teams, we need software 
engineers that did client 

513
00:27:22,240 --> 00:27:25,880
facing, API client facing 
products for many years with 

514
00:27:25,880 --> 00:27:27,840
Python. 
They're really rare to find and 

515
00:27:27,840 --> 00:27:30,680
and then not mention all the new
vibing tools that are coming in 

516
00:27:30,680 --> 00:27:32,800
agents. 
So people who enjoy playing with

517
00:27:32,800 --> 00:27:35,800
all these configurations and all
these new tools that there is no

518
00:27:35,800 --> 00:27:39,120
name for it yet. 
It's very complex to build a 

519
00:27:39,120 --> 00:27:42,400
team with not real defined 
roles. 

520
00:27:42,440 --> 00:27:46,000
So you need to find talents. 
You need to find talents which 

521
00:27:46,000 --> 00:27:48,080
are. 
Flexible that likes to explore. 

522
00:27:48,080 --> 00:27:51,200
Things and experiment with 
things and are not afraid from 

523
00:27:51,200 --> 00:27:52,200
changes. 
From that. 

524
00:27:52,200 --> 00:27:55,400
Of course you get the software 
that eventually is very flexible

525
00:27:55,400 --> 00:27:59,880
and software which is very thing
and and and you can throw parts 

526
00:27:59,880 --> 00:28:01,600
and implement new parts. 
Daily. 

527
00:28:01,760 --> 00:28:04,680
And if you now look. 
Ahead and I know in AI and in 

528
00:28:04,680 --> 00:28:07,560
technology, it's super hard to 
look ahead and like let's say at

529
00:28:07,560 --> 00:28:11,280
least two years from now. 
But if we do, and if you look 

530
00:28:11,280 --> 00:28:15,360
into the field of AI companions 
and their evolution and how they

531
00:28:15,360 --> 00:28:19,080
integrate deeper into our lives,
what would you like to see 

532
00:28:19,080 --> 00:28:24,200
changed or also regarded when we
talk about trust and when we 

533
00:28:24,200 --> 00:28:28,560
talk about AI? 
And what would you you want this

534
00:28:28,680 --> 00:28:31,120
shift or change to feel like? 
I have. 

535
00:28:31,120 --> 00:28:35,920
Opinions from my my my learnings
and experience and perspective. 

536
00:28:35,960 --> 00:28:37,280
I'm not sure that they are 
right. 

537
00:28:37,280 --> 00:28:39,840
It's a very complex topic I 
think. 

538
00:28:39,840 --> 00:28:44,800
I think the discussion about AI.
Should should go back. 

539
00:28:45,040 --> 00:28:48,000
A bit to then to to what are the
engineering? 

540
00:28:48,000 --> 00:28:50,960
Capabilities of these? 
Things It's still a statistical 

541
00:28:51,160 --> 00:28:53,400
machine. 
We sometimes expected to have 

542
00:28:53,400 --> 00:28:58,240
100% success in being 
trustworthy or not to do. 

543
00:28:58,240 --> 00:29:00,640
Stuff we don't expect it to do 
for. 

544
00:29:00,640 --> 00:29:04,120
Example if we want to prevent 
all the false positive so we 

545
00:29:04,120 --> 00:29:08,320
want some some AI system not to 
mention certain things or not 

546
00:29:08,320 --> 00:29:11,360
to. 
Suggest people to do. 

547
00:29:11,960 --> 00:29:15,320
Bad things, we need to eliminate
all the false positives or it's 

548
00:29:15,320 --> 00:29:18,560
also when when we want to catch 
fraud or catch crime over social

549
00:29:18,560 --> 00:29:21,360
media, we want to eliminate. 
All the false positive. 

550
00:29:21,360 --> 00:29:24,400
But then we will also hurt 
innocent people and block them 

551
00:29:24,400 --> 00:29:26,120
or even send the police to their
door. 

552
00:29:26,120 --> 00:29:29,120
So it's a game. 
With these classifiers that 

553
00:29:29,120 --> 00:29:30,680
that. 
That you need to play with and 

554
00:29:30,680 --> 00:29:33,160
it can never be 100%. 
And sometimes people have 

555
00:29:33,160 --> 00:29:37,360
expectations of a 100% and they 
forget that these things are 

556
00:29:37,360 --> 00:29:40,520
statistics. 
So this discussion is A and and 

557
00:29:40,520 --> 00:29:41,880
B. 
We will have much. 

558
00:29:41,880 --> 00:29:45,040
Healthier products also if. 
Journalists, regulators will 

559
00:29:45,040 --> 00:29:49,040
will understand how these things
are working and BI think we 

560
00:29:49,040 --> 00:29:52,240
should have a discussion. 
I don't really have a full 

561
00:29:52,240 --> 00:29:55,400
opinion about it. 
I had a very good discussion 

562
00:29:55,400 --> 00:29:57,080
about. 
It with with with a. 

563
00:29:57,080 --> 00:30:02,360
Friend the other day, he's a 
journalist and he said, I think.

564
00:30:02,840 --> 00:30:06,280
We should just make laws that 
should. 

565
00:30:06,400 --> 00:30:09,040
Force all these companies. 
To have age. 

566
00:30:09,040 --> 00:30:12,720
Protection and that AI will 
never talk about sexual things 

567
00:30:12,720 --> 00:30:14,800
etcetera, etcetera and always 
like. 

568
00:30:15,000 --> 00:30:18,400
The law will. 
Prevent it and we had I think 10

569
00:30:18,400 --> 00:30:21,480
minutes discussion. 
I tried to explain to him, but. 

570
00:30:21,480 --> 00:30:22,880
And that's a very good friend. 
Yeah. 

571
00:30:22,880 --> 00:30:24,640
So we have this opening 
discussion, but you don't 

572
00:30:24,640 --> 00:30:26,240
understand it's. 
Impossible. 

573
00:30:26,240 --> 00:30:28,240
This is not how. 
Engineering works, no, I expect 

574
00:30:28,240 --> 00:30:33,480
my the the law to to protect me 
and my child and this is quite 

575
00:30:33,480 --> 00:30:37,600
problematic thing to expect and 
I think we should have also a 

576
00:30:37,600 --> 00:30:41,880
discussion around. 
What we as society expect. 

577
00:30:42,280 --> 00:30:46,520
AI can can can do things that we
want users not to do with 

578
00:30:46,520 --> 00:30:49,080
humans. 
Okay, for example, sex talks. 

579
00:30:49,520 --> 00:30:50,840
I will be. 
Happier if people. 

580
00:30:50,840 --> 00:30:53,320
Will go and talk about it with 
AI then people who are getting 

581
00:30:53,320 --> 00:30:56,080
exploited in only fence or 
things like that. 

582
00:30:56,120 --> 00:30:58,360
But do we want it? 
Do we want to encourage that? 

583
00:30:58,360 --> 00:31:00,840
Now we kind of open this thing 
and said okay. 

584
00:31:01,080 --> 00:31:02,720
That's okay because it's with 
AI. 

585
00:31:02,720 --> 00:31:04,840
Where is the border? 
If they print. 

586
00:31:05,240 --> 00:31:07,600
Pictures fake. 
Pictures not of real people. 

587
00:31:08,000 --> 00:31:11,320
Do we want to allow that or we 
don't want to allow the time? 

588
00:31:11,640 --> 00:31:14,200
I think we need to have this 
discussion, but I don't think it

589
00:31:14,200 --> 00:31:17,880
is discussion that has a real 
solution, but it's a very 

590
00:31:17,880 --> 00:31:19,800
important one. 
That we as a society. 

591
00:31:19,800 --> 00:31:21,520
Should have shouldn't end 
immediately. 

592
00:31:21,520 --> 00:31:24,560
With with the law regulators. 
It's so interesting because we 

593
00:31:24,560 --> 00:31:28,040
look into, if we look into those
discussions. 

594
00:31:28,160 --> 00:31:33,280
For me, it always comes down to 
questions that are as old as 

595
00:31:33,280 --> 00:31:37,360
humanity is like, what is, you 
know, right, what is wrong? 

596
00:31:37,360 --> 00:31:40,320
Where do you start? 
Where do you also frame it? 

597
00:31:40,680 --> 00:31:45,320
And what is real or what is also
like empathetic towards each 

598
00:31:45,320 --> 00:31:48,560
other. 
And I'm just wondering how you 

599
00:31:48,560 --> 00:31:53,120
would like how you would open up
the space to have the discussion

600
00:31:53,120 --> 00:31:55,600
on a way that there can be a 
result. 

601
00:31:55,720 --> 00:31:57,280
Do you have an opinion on that? 
I think. 

602
00:31:57,280 --> 00:31:59,360
We are having this discussion as
I said. 

603
00:31:59,360 --> 00:32:04,360
For a long time, but for me kind
of from from what I've been 

604
00:32:04,360 --> 00:32:08,320
dealing with for for many years,
again, social media is a great, 

605
00:32:08,320 --> 00:32:11,200
great example. 
We had these discussions all the

606
00:32:11,200 --> 00:32:15,400
time, what to allow on social 
media, who is responsible to 

607
00:32:15,400 --> 00:32:17,440
what happens on social media 
there and. 

608
00:32:17,440 --> 00:32:19,840
There's a very big. 
Difference between how the 

609
00:32:20,160 --> 00:32:23,360
European Union see it and how 
the the the US regulator seeing 

610
00:32:23,360 --> 00:32:26,840
it, the platform is responsible 
or the humans at the edge. 

611
00:32:29,680 --> 00:32:31,480
It's it's it's a very 
problematic. 

612
00:32:31,480 --> 00:32:33,160
I I don't think there is an end 
solution. 

613
00:32:33,160 --> 00:32:35,080
I think you can say that the 
market. 

614
00:32:35,080 --> 00:32:37,200
Will will solve it. 
So good products that will 

615
00:32:37,200 --> 00:32:40,880
protect the users will will get.
Gain the trust and they will win

616
00:32:40,960 --> 00:32:42,800
the market. 
It's not always true. 

617
00:32:42,800 --> 00:32:47,040
You also need a good balance of 
regulations and I think what I 

618
00:32:47,040 --> 00:32:50,400
find the best is the discussions
on, on, you know, the blog post 

619
00:32:50,400 --> 00:32:52,080
and newspapers and journalism 
and. 

620
00:32:52,080 --> 00:32:54,720
Podcasts, of course, think this 
is where it should happen. 

621
00:32:54,720 --> 00:32:57,360
Because in this place you have 
people from all sides. 

622
00:32:57,560 --> 00:33:00,760
People could listen, make up 
their opinion, learn. 

623
00:33:00,760 --> 00:33:01,960
And that's the most important 
thing. 

624
00:33:01,960 --> 00:33:05,680
We actually didn't talk enough. 
About it, but educate people. 

625
00:33:05,680 --> 00:33:08,040
Learn that I have control. 
Learn that. 

626
00:33:08,200 --> 00:33:09,560
This is right. 
This is wrong. 

627
00:33:09,680 --> 00:33:13,520
But my kids can still do those 
things and I need to to to watch

628
00:33:13,520 --> 00:33:15,280
for it and. 
Decide what what is good? 

629
00:33:15,280 --> 00:33:18,680
For my kid, you know, it's like 
the decision kids at certain age

630
00:33:18,680 --> 00:33:21,480
getting an iPhone or not. 
It's my decision as a parent 

631
00:33:21,480 --> 00:33:24,880
many times as much as other 
things won't influence it. 

632
00:33:24,880 --> 00:33:29,640
But but I I can still be. 
I can be. 

633
00:33:29,640 --> 00:33:32,080
Educated that this. 
IPhone has some protection 

634
00:33:32,080 --> 00:33:33,080
measures. 
For kids. 

635
00:33:33,080 --> 00:33:35,760
So I can use it. 
If we will have these 

636
00:33:35,760 --> 00:33:38,200
discussions openly. 
On the television, with 

637
00:33:38,200 --> 00:33:40,320
podcasts. 
Blog posts everywhere. 

638
00:33:40,440 --> 00:33:43,920
People will, will, will. 
Get educated will know what they

639
00:33:43,920 --> 00:33:47,560
can, what they cannot, and. 
That's hopefully will push. 

640
00:33:47,560 --> 00:33:50,480
Also the industry and the. 
Products to a good. 

641
00:33:50,520 --> 00:33:53,440
Place you know we will. 
Always have products that will 

642
00:33:53,640 --> 00:33:55,320
do. 
Less good things and people. 

643
00:33:55,320 --> 00:33:57,320
Will use it. 
I don't see a way to avoid it. 

644
00:33:57,320 --> 00:34:01,640
You can have some regulations in
the EU and then another country.

645
00:34:01,640 --> 00:34:03,960
Will will serve. 
From from their. 

646
00:34:03,960 --> 00:34:06,200
Server as a product. 
That is doing something against 

647
00:34:06,200 --> 00:34:07,920
it and you cannot really prevent
it. 

648
00:34:07,920 --> 00:34:09,120
It's the Internet. 
Yeah. 

649
00:34:09,120 --> 00:34:12,600
I was just in my head, I was 
running down the rabbit hole of 

650
00:34:12,600 --> 00:34:16,760
like global decision and 
regulations versus local ones. 

651
00:34:16,760 --> 00:34:21,480
But that will be a whole 
different conversation and we're

652
00:34:21,480 --> 00:34:24,000
already. 
Yeah, let this topic I always. 

653
00:34:24,000 --> 00:34:26,199
Have in my head the example of 
the. 

654
00:34:26,560 --> 00:34:30,280
So I I always have this example 
in my head and and in the past 

655
00:34:30,360 --> 00:34:32,040
there was. 
This whole topic of Dolly the 

656
00:34:32,040 --> 00:34:35,159
Ship. 
That it was not allowed to to 

657
00:34:35,159 --> 00:34:39,159
clone animals in the US, in 
Europe, in the whole world and 

658
00:34:39,159 --> 00:34:40,920
then somewhere in Asia. 
I don't remember in which 

659
00:34:40,920 --> 00:34:43,320
country it happened, so I don't 
see a. 

660
00:34:43,320 --> 00:34:46,400
World where we. 
Can enforce regulation globally.

661
00:34:46,400 --> 00:34:48,760
It's just impossible. 
I I just don't see this 

662
00:34:48,760 --> 00:34:50,719
happening. 
We need to be smarter a bit. 

663
00:34:50,880 --> 00:34:52,400
Yeah, I would. 
I'm super cute. 

664
00:34:52,480 --> 00:34:54,280
Curious if that's gonna turn out
true. 

665
00:34:54,280 --> 00:34:57,280
Because also like on a European 
level, there is like a whole 

666
00:34:57,280 --> 00:35:00,720
conversation about how to 
regulate on European level when 

667
00:35:00,720 --> 00:35:03,840
there is like national interest 
and then globally as well. 

668
00:35:03,840 --> 00:35:07,160
So I think it's gonna be super 
interesting the next coming 

669
00:35:07,160 --> 00:35:11,960
years to define and also a trade
on the way forward. 

670
00:35:12,360 --> 00:35:15,480
What happened now is the age 
gate in Australia that it was 

671
00:35:15,480 --> 00:35:18,880
implemented, but OK, You're not 
allowed to log in to register as

672
00:35:18,880 --> 00:35:21,280
a team, but you can still see 
the feed when you're not logged 

673
00:35:21,280 --> 00:35:21,680
in. 
So what? 

674
00:35:21,680 --> 00:35:25,280
Did we do there but. 
So then I'm if I have a website 

675
00:35:25,280 --> 00:35:28,360
that sits in Europe, I'm not. 
Allowed to serve it to. 

676
00:35:28,360 --> 00:35:31,240
Australia or I need to serve it 
differently to Australia or it's

677
00:35:31,240 --> 00:35:34,680
only Australian website? 
It's so complex on the Internet.

678
00:35:34,920 --> 00:35:37,040
Who is not allowed the? 
User or the website or? 

679
00:35:37,040 --> 00:35:38,200
How do you? 
Enforce it Really. 

680
00:35:38,200 --> 00:35:39,800
Globally Super. 
Complex. 

681
00:35:40,040 --> 00:35:43,560
Absolutely. 
And I'm coming to the end of our

682
00:35:43,560 --> 00:35:46,360
podcast episode. 
I always have those so-called in

683
00:35:46,360 --> 00:35:49,240
between moments with my guests, 
which I love, which are also 

684
00:35:49,240 --> 00:35:54,360
sometimes a bit tricky because I
ask you in particular to answer 

685
00:35:54,520 --> 00:35:58,440
like a prompt that I give 
quickly, intuitively and just 

686
00:35:58,800 --> 00:36:02,560
with a reflection of what comes 
to your mind instantly. 

687
00:36:02,560 --> 00:36:06,400
And I want to start with what 
trust feels like to you. 

688
00:36:06,440 --> 00:36:10,880
Perfect relationship. 
And what should an AI companion 

689
00:36:10,880 --> 00:36:13,720
not do? 
I think it shouldn't be what I 

690
00:36:13,720 --> 00:36:17,160
don't want it to be. 
I think it it should be what I 

691
00:36:17,160 --> 00:36:20,240
want myself to be, the best 
thing of me and plus. 

692
00:36:20,240 --> 00:36:24,440
The the superpowers of of. 
Machines like searching stuff 

693
00:36:24,440 --> 00:36:27,120
really fast and answering me. 
What is agency for you? 

694
00:36:27,640 --> 00:36:30,320
Having control over my 
boundaries. 

695
00:36:30,320 --> 00:36:33,600
What is the biggest risk in AI 
companionship? 

696
00:36:33,680 --> 00:36:36,200
Unrealistic expectations from 
all sides. 

697
00:36:36,240 --> 00:36:41,600
And when we look into 20-30 and 
that's five year heads now, what

698
00:36:41,600 --> 00:36:44,480
will users still need? 
First, humans around them. 

699
00:36:44,760 --> 00:36:47,640
This will never change. 
We won't become so lonely. 

700
00:36:47,680 --> 00:36:49,600
We'll need technology to help. 
US. 

701
00:36:50,040 --> 00:36:52,560
Thank you for a perfect ending. 
I really appreciate the 

702
00:36:52,560 --> 00:36:55,840
conversation and the time that 
you took and all the insights 

703
00:36:55,840 --> 00:36:57,920
that you share was great having 
you Lear. 

704
00:36:58,000 --> 00:36:59,400
Thank you very much, Eva. 
Thank you.