AI Lab: The Feedback Loop Behind Trustworthy AI

1
00:00:00,150 --> 00:00:02,820
- AI is more accessible than ever.

2
00:00:02,820 --> 00:00:05,010
In the real world and behind the scenes,

3
00:00:05,010 --> 00:00:06,690
it's a very different story.

4
00:00:06,690 --> 00:00:07,620
- That's right.

5
00:00:07,620 --> 00:00:09,360
I mean, think about the built environments

6
00:00:09,360 --> 00:00:10,620
as an example, right?

7
00:00:10,620 --> 00:00:13,770
Conditions are always changing,
buildings are different.

8
00:00:13,770 --> 00:00:16,950
And you know what we
tend to forget sometimes?

9
00:00:16,950 --> 00:00:18,300
There are real people

10
00:00:18,300 --> 00:00:21,510
living and working inside
of these buildings.

11
00:00:21,510 --> 00:00:24,600
How do we know they're
reliable enough to deploy?

12
00:00:24,600 --> 00:00:27,030
- Well, that's the question
that we'll have answered today

13
00:00:27,030 --> 00:00:30,060
because our guests actually
have met that challenge.

14
00:00:30,060 --> 00:00:33,810
They do not just research
AI and dream up things,

15
00:00:33,810 --> 00:00:36,330
but they also build, test,

16
00:00:36,330 --> 00:00:39,150
deploy, and then they do it all again.

17
00:00:39,150 --> 00:00:40,740
- Oh, this sounds exciting.

18
00:00:40,740 --> 00:00:42,150
So where are we gonna start?

19
00:00:42,150 --> 00:00:45,120
- We'll start at the
research with Foutse Khomh.

20
00:00:45,120 --> 00:00:49,260
He's an AI researcher and
leader at IVADO and Mila.

21
00:00:49,260 --> 00:00:50,460
- [Dominique] I'm Dominique Silva.

22
00:00:50,460 --> 00:00:51,390
- [Scott] And I'm Scott Tew,

23
00:00:51,390 --> 00:00:53,820
and this is "Healthy Spaces,"

24
00:00:53,820 --> 00:00:57,543
conversations at the frontier
of sustainable technology.

25
00:00:58,680 --> 00:01:03,090
- AI is a technology with a
high disruptive potential.

26
00:01:03,090 --> 00:01:06,840
It will significantly transform
the way we do many things.

27
00:01:06,840 --> 00:01:09,420
As this technology will permeate

28
00:01:09,420 --> 00:01:11,220
the society and our activity,

29
00:01:11,220 --> 00:01:14,670
trust and trustworthiness in
the technology, it's critical.

30
00:01:14,670 --> 00:01:16,320
From an engineering point of view,

31
00:01:16,320 --> 00:01:19,740
it means the engineer building the systems

32
00:01:19,740 --> 00:01:21,810
have gained enough confidence to be able

33
00:01:21,810 --> 00:01:24,840
to provide guarantee about the
functioning of the systems.

34
00:01:24,840 --> 00:01:26,160
So we may not understand

35
00:01:26,160 --> 00:01:28,770
all the inner working of a technology,

36
00:01:28,770 --> 00:01:31,590
but we have engineering tools

37
00:01:31,590 --> 00:01:35,370
that allow us to study the technology,

38
00:01:35,370 --> 00:01:37,590
to test it, to interact with it,

39
00:01:37,590 --> 00:01:41,130
and grow our confidence in
the behavior of the technology

40
00:01:41,130 --> 00:01:45,960
to a point that we can provide
some level of guarantee

41
00:01:45,960 --> 00:01:50,960
that this is likely to be
providing this kind of output.

42
00:01:51,180 --> 00:01:53,460
Although the output may be
different from one query

43
00:01:53,460 --> 00:01:56,640
to the next, but the output
should remain aligned

44
00:01:56,640 --> 00:01:58,140
with the expectation,

45
00:01:58,140 --> 00:02:01,050
and that allow us to be
able to provide guarantees

46
00:02:01,050 --> 00:02:03,600
that this will provide a service

47
00:02:03,600 --> 00:02:07,440
with this level of reliability.

48
00:02:07,440 --> 00:02:08,790
- So it's sort of a guarantee

49
00:02:08,790 --> 00:02:11,370
that the tool will meet expectations.

50
00:02:11,370 --> 00:02:12,750
Is that what you're saying?

51
00:02:12,750 --> 00:02:14,610
- Exactly. Exactly.

52
00:02:14,610 --> 00:02:16,410
- So what if you identify a weakness

53
00:02:16,410 --> 00:02:18,510
or what if a user identifies a weakness?

54
00:02:18,510 --> 00:02:21,210
What's the role there
for someone in your field

55
00:02:21,210 --> 00:02:22,950
that works on trustworthiness?

56
00:02:22,950 --> 00:02:26,010
Isn't weakness a concern?

57
00:02:26,010 --> 00:02:28,230
- Understanding the weaknesses

58
00:02:28,230 --> 00:02:32,130
and all the failing mode, it's
critical to be able to ensure

59
00:02:32,130 --> 00:02:35,520
that the technology that
we bring in productions

60
00:02:35,520 --> 00:02:37,860
have been calibrated,

61
00:02:37,860 --> 00:02:41,520
controlled in a way that
the output can be reliable.

62
00:02:41,520 --> 00:02:43,230
So in the field of trustworthy AI,

63
00:02:43,230 --> 00:02:44,820
most of the things that we try to do,

64
00:02:44,820 --> 00:02:47,250
it's, for example, we try to invent

65
00:02:47,250 --> 00:02:50,250
novel techniques to ensure
that when we build AI system,

66
00:02:50,250 --> 00:02:51,810
they are robust, right?

67
00:02:51,810 --> 00:02:56,160
By robust mean under
unexpected conditions,

68
00:02:56,160 --> 00:02:58,200
the system should still
maintain a behavior

69
00:02:58,200 --> 00:03:00,030
that is aligned with our expectation.

70
00:03:00,030 --> 00:03:03,060
How can we ensure that in a system,

71
00:03:03,060 --> 00:03:04,620
an AI system that we build?

72
00:03:04,620 --> 00:03:07,230
So this require a deeper understanding,

73
00:03:07,230 --> 00:03:09,660
not only how the system
behave in distribution,

74
00:03:09,660 --> 00:03:11,640
because the current form of AI,

75
00:03:11,640 --> 00:03:13,440
which is heavily data driven,

76
00:03:13,440 --> 00:03:16,290
usually will behave pretty
well in distribution.

77
00:03:16,290 --> 00:03:18,060
The system, the training process

78
00:03:18,060 --> 00:03:21,600
allow us to extract as much
regularity from the data

79
00:03:21,600 --> 00:03:22,830
to which the system is exposed

80
00:03:22,830 --> 00:03:26,160
and build a system that will
be fairly good in distribution.

81
00:03:26,160 --> 00:03:30,840
But usually when the system is
deployed out of distribution,

82
00:03:30,840 --> 00:03:35,190
we expect the systems to be facing

83
00:03:35,190 --> 00:03:37,890
with inputs that to some extent

84
00:03:37,890 --> 00:03:39,480
it's a bit novel to the system.

85
00:03:39,480 --> 00:03:41,400
Of course, we train system to generalize,

86
00:03:41,400 --> 00:03:43,110
but generalization is still a struggle

87
00:03:43,110 --> 00:03:44,550
that we have in the AI field.

88
00:03:44,550 --> 00:03:46,800
- There's this work that's in a lab.

89
00:03:46,800 --> 00:03:49,890
You actually are a partner
for the BrainBox AI Lab,

90
00:03:49,890 --> 00:03:52,530
for instance, and where things get tested,

91
00:03:52,530 --> 00:03:55,020
but at some point you
release them into the world

92
00:03:55,020 --> 00:03:57,270
and you're expecting this tool

93
00:03:57,270 --> 00:04:01,260
to perform to a certain
level of standards.

94
00:04:01,260 --> 00:04:03,380
Does it always work perfectly?

95
00:04:03,380 --> 00:04:05,222
- Not always. Of course, it's a challenge.

96
00:04:05,222 --> 00:04:08,040
And part of the work that we
do is to try to bridge the gap

97
00:04:08,040 --> 00:04:09,450
between these two reality.

98
00:04:09,450 --> 00:04:12,930
So when we do in-lab
testing, it's very important

99
00:04:12,930 --> 00:04:16,590
to provide the testing frameworks

100
00:04:16,590 --> 00:04:18,990
that mimic as much as possible

101
00:04:18,990 --> 00:04:20,430
the reality of the real world

102
00:04:20,430 --> 00:04:22,194
in which the solution will be deployed.

103
00:04:22,194 --> 00:04:23,670
In the case of vendors, for example,

104
00:04:23,670 --> 00:04:26,613
if we're building a
technology for buildings,

105
00:04:27,600 --> 00:04:32,600
because of the domain knowledge
of the context of buildings,

106
00:04:32,730 --> 00:04:36,690
we are able to design testing framework

107
00:04:36,690 --> 00:04:39,360
that's emulate as much as possible

108
00:04:39,360 --> 00:04:42,330
the reality of the real-world environment.

109
00:04:42,330 --> 00:04:44,520
At leveraging this framework,

110
00:04:44,520 --> 00:04:48,690
we are able to exercise the
behavior of the solution

111
00:04:48,690 --> 00:04:51,360
that we're constructing
as much as possible

112
00:04:51,360 --> 00:04:55,770
to understand the
weaknesses, the corner cases,

113
00:04:55,770 --> 00:04:58,650
and to improve of these weaknesses,

114
00:04:58,650 --> 00:05:01,860
to understand the operation
domain of the solution.

115
00:05:01,860 --> 00:05:03,690
So by through this process,

116
00:05:03,690 --> 00:05:06,180
we can understand the envelope

117
00:05:06,180 --> 00:05:10,260
in which the solution has a
certain level of robustness,

118
00:05:10,260 --> 00:05:13,350
so a certain level of reliability,

119
00:05:13,350 --> 00:05:15,780
that is expected when the
solution is deployed in the field.

120
00:05:15,780 --> 00:05:20,190
And on the spaces where we don't
have a level of reliability

121
00:05:20,190 --> 00:05:22,890
that's up to the standard
of our expectation,

122
00:05:22,890 --> 00:05:25,650
there are different
mechanism that we pursue

123
00:05:25,650 --> 00:05:30,210
to try to either build more
adapted AI for those spaces

124
00:05:30,210 --> 00:05:32,670
or the capability of the
AI system in those spaces.

125
00:05:32,670 --> 00:05:34,530
So it's really an engineering process

126
00:05:34,530 --> 00:05:36,110
where deeper understanding

127
00:05:36,110 --> 00:05:38,220
of the inner working of our solution,

128
00:05:38,220 --> 00:05:39,990
the failing mode of our solution,

129
00:05:39,990 --> 00:05:41,730
allow us to build control around

130
00:05:41,730 --> 00:05:44,190
and to build a systems of systems

131
00:05:44,190 --> 00:05:48,030
that allow us to back this
reliability and trust principle.

132
00:05:48,030 --> 00:05:51,210
And at the same time,
you can have a systems

133
00:05:51,210 --> 00:05:56,210
on which user over-trust,
which is not trustworthy.

134
00:05:56,550 --> 00:05:59,850
- Would you say that over-trusting
is a bigger problem even?

135
00:05:59,850 --> 00:06:01,320
- Over-trusting is a bigger problem

136
00:06:01,320 --> 00:06:04,680
because over-trusting will
lead to deploying solutions

137
00:06:04,680 --> 00:06:07,280
in context in which they
shouldn't have been deployed

138
00:06:07,280 --> 00:06:09,510
in the right place, which means

139
00:06:09,510 --> 00:06:13,470
when the inevitable
adversary consequences occur,

140
00:06:13,470 --> 00:06:15,810
this usually actually destroy the trust.

141
00:06:15,810 --> 00:06:18,630
So it's important to avoid

142
00:06:18,630 --> 00:06:20,850
building solution on
which people over-trust

143
00:06:20,850 --> 00:06:23,190
because trust is very hard to gain,

144
00:06:23,190 --> 00:06:25,560
but very easy (chuckles) to last.

145
00:06:25,560 --> 00:06:29,940
Once the system diverts from
the expectation of users,

146
00:06:29,940 --> 00:06:32,850
the mistrust becomes
actually more stronger,

147
00:06:32,850 --> 00:06:34,860
strongly ingrained in the end users.

148
00:06:34,860 --> 00:06:38,250
It's achieved through good understanding

149
00:06:38,250 --> 00:06:39,900
of the behavioral system

150
00:06:39,900 --> 00:06:42,660
and ability to communicate on the strength

151
00:06:42,660 --> 00:06:44,700
and the limitations of the solutions

152
00:06:44,700 --> 00:06:46,320
to the different stakeholders

153
00:06:46,320 --> 00:06:50,040
and also the process in
which that is in place

154
00:06:50,040 --> 00:06:53,130
to ensure that the behavior
is being monitored,

155
00:06:53,130 --> 00:06:54,540
it's being controlled,

156
00:06:54,540 --> 00:06:57,630
this process to be trustworthy

157
00:06:57,630 --> 00:06:59,520
in the eyes of these
different stakeholders.

158
00:06:59,520 --> 00:07:00,930
It's all these ingredients

159
00:07:00,930 --> 00:07:04,530
that enable the deployment
of a trustworthy solution.

160
00:07:04,530 --> 00:07:06,900
- Do all these systems, all these tools,

161
00:07:06,900 --> 00:07:08,850
do they all have flaws?

162
00:07:08,850 --> 00:07:10,800
Are they like humans
where we all have flaws,

163
00:07:10,800 --> 00:07:13,110
we all have some kind of weakness,

164
00:07:13,110 --> 00:07:14,940
should we view AI

165
00:07:14,940 --> 00:07:19,320
and AI tools having similar weaknesses?

166
00:07:19,320 --> 00:07:22,320
- So I think it's depend
on the type of AI.

167
00:07:22,320 --> 00:07:23,910
It's evolving as we are building

168
00:07:23,910 --> 00:07:25,320
capability in those systems.

169
00:07:25,320 --> 00:07:28,440
That's to say, depending on the complexity

170
00:07:28,440 --> 00:07:31,680
of the AI technology, we
have quality assurance

171
00:07:31,680 --> 00:07:36,680
and safety tools that have
different level of maturity.

172
00:07:36,720 --> 00:07:38,640
And it's also very
important for the audience

173
00:07:38,640 --> 00:07:41,100
to be aware of this
because all this contribute

174
00:07:41,100 --> 00:07:42,060
to the level of trust

175
00:07:42,060 --> 00:07:44,700
that we should be putting
in those technology.

176
00:07:44,700 --> 00:07:47,550
- We had a conversation
prior to today's discussion

177
00:07:47,550 --> 00:07:52,530
and you mentioned that think
about AI as Swiss cheese.

178
00:07:52,530 --> 00:07:53,363
It's great.

179
00:07:53,363 --> 00:07:54,990
There are always some holes

180
00:07:54,990 --> 00:07:57,030
and it's trying to figure
out what those holes are.

181
00:07:57,030 --> 00:07:59,130
I thought that was really helpful analogy.

182
00:07:59,130 --> 00:08:00,870
- That's a big analogy behind my work,

183
00:08:00,870 --> 00:08:02,860
like, to me, my approach to trustworthy

184
00:08:02,860 --> 00:08:06,240
is to try to understand
as much as possible,

185
00:08:06,240 --> 00:08:08,730
where are the holes in those systems?

186
00:08:08,730 --> 00:08:11,400
Because even if the structure,

187
00:08:11,400 --> 00:08:15,150
it's fragile in the sense
that it has all these holes,

188
00:08:15,150 --> 00:08:16,620
if we know where the holes are,

189
00:08:16,620 --> 00:08:19,890
we can stack this cheese in a way

190
00:08:19,890 --> 00:08:22,740
that allow us to compensate
from these weaknesses

191
00:08:22,740 --> 00:08:23,970
and build a strong structure.

192
00:08:23,970 --> 00:08:27,120
So this allow us to build an
architecture that is robust,

193
00:08:27,120 --> 00:08:30,180
an architecture on which
we can actually provide

194
00:08:30,180 --> 00:08:32,190
some level of guarantees of behavior,

195
00:08:32,190 --> 00:08:33,810
an architecture on which we can have

196
00:08:33,810 --> 00:08:35,310
some level of observability

197
00:08:35,310 --> 00:08:36,900
because observability is very critical

198
00:08:36,900 --> 00:08:38,700
as when we're building complex systems.

199
00:08:38,700 --> 00:08:41,730
We need to understand how
the different pieces interact

200
00:08:41,730 --> 00:08:44,640
to contribute to the
behavior of the systems.

201
00:08:44,640 --> 00:08:45,660
And this understanding,

202
00:08:45,660 --> 00:08:48,540
it's what allow us to have
the right control in place

203
00:08:48,540 --> 00:08:49,890
and the right way in place.

204
00:08:49,890 --> 00:08:52,410
And this is what allow
us to build solution

205
00:08:52,410 --> 00:08:55,540
that are trustworthy on which users

206
00:08:57,780 --> 00:09:01,800
could reasonably put

207
00:09:01,800 --> 00:09:03,480
their trust on the system.

208
00:09:03,480 --> 00:09:07,890
- So Foutse, what is the
role of this testing in labs,

209
00:09:07,890 --> 00:09:12,360
maybe, like, the BrainBox AI
Lab where the testing occurs?

210
00:09:12,360 --> 00:09:14,490
I mean, what's the role
of collaboration there?

211
00:09:14,490 --> 00:09:19,080
Somebody like you who's a
scientific director, a researcher,

212
00:09:19,080 --> 00:09:20,670
what's the role of collaboration

213
00:09:20,670 --> 00:09:24,900
and bringing these tools to
the public that we can trust?

214
00:09:24,900 --> 00:09:28,293
- In the case, for example,
of BrainBox, the lab,

215
00:09:29,220 --> 00:09:32,400
it's now one of our
shining piece in Montreal.

216
00:09:32,400 --> 00:09:35,040
It's a very happy result
of a collaboration

217
00:09:35,040 --> 00:09:36,630
that happened within the ecosystem

218
00:09:36,630 --> 00:09:39,630
over a significant period of time

219
00:09:39,630 --> 00:09:42,330
where our academic prototypes,

220
00:09:42,330 --> 00:09:45,390
our academic technologies could interact

221
00:09:45,390 --> 00:09:47,580
with the reality of the use case

222
00:09:47,580 --> 00:09:50,640
of the industry in which BrainBox,

223
00:09:50,640 --> 00:09:52,770
it's actually deploying technology.

224
00:09:52,770 --> 00:09:55,680
And this information coming
from the industry field,

225
00:09:55,680 --> 00:10:00,360
it's what allow us to calibrate
testing evaluation framework

226
00:10:00,360 --> 00:10:02,550
to ensure that the tools

227
00:10:02,550 --> 00:10:05,880
and the new algorithm
that we are designing

228
00:10:05,880 --> 00:10:10,380
cope with the reality of that ecosystems.

229
00:10:10,380 --> 00:10:13,260
It's allowed to get use
cases, real-world use cases,

230
00:10:13,260 --> 00:10:15,720
for calibrating on
testing this technology.

231
00:10:15,720 --> 00:10:18,630
So to ensure trustworthy,
we need all these players

232
00:10:18,630 --> 00:10:21,780
to have these constant
interactions, right,

233
00:10:21,780 --> 00:10:24,540
because the challenge
of the real environment

234
00:10:24,540 --> 00:10:28,380
inform some of the
design academic challenge

235
00:10:28,380 --> 00:10:31,530
that we have on a scientific perspective.

236
00:10:31,530 --> 00:10:35,070
And it's allow us to expand the technology

237
00:10:35,070 --> 00:10:38,250
that we're building, which,
with the connections,

238
00:10:38,250 --> 00:10:40,620
it's being tested in real context.

239
00:10:40,620 --> 00:10:44,520
And we have this feedback that
is coming back to the labs.

240
00:10:44,520 --> 00:10:46,890
And all this, it's feeding this ecosystem

241
00:10:46,890 --> 00:10:48,870
and allow us to innovate constantly

242
00:10:48,870 --> 00:10:51,030
and to continue to pursue in this way.

243
00:10:51,030 --> 00:10:54,690
If you look at the landscape
of AI industry nowadays,

244
00:10:54,690 --> 00:10:57,960
the frontier labs that are contributing,

245
00:10:57,960 --> 00:11:00,930
the most advanced
recently are industry lab.

246
00:11:00,930 --> 00:11:01,763
Why?

247
00:11:01,763 --> 00:11:04,260
Because they have these use cases,

248
00:11:04,260 --> 00:11:08,430
because they have this
interaction with the user base.

249
00:11:08,430 --> 00:11:11,970
They have this third-party
feedback that are coming in

250
00:11:11,970 --> 00:11:14,550
to allow the scientific team to push,

251
00:11:14,550 --> 00:11:18,780
continue to push the boundary
of the research innovation.

252
00:11:18,780 --> 00:11:21,030
- So those feedback loops just continue.

253
00:11:21,030 --> 00:11:23,040
- That's continue and that is critical.

254
00:11:23,040 --> 00:11:26,310
So in our context, I think
the lab that have been opened

255
00:11:26,310 --> 00:11:27,900
by Brainbox AI in Montreal,

256
00:11:27,900 --> 00:11:30,810
it's a great addition to our ecosystem

257
00:11:30,810 --> 00:11:32,010
and it's what will allow us

258
00:11:32,010 --> 00:11:34,830
to continue innovating on this field,

259
00:11:34,830 --> 00:11:37,290
allow us to continue build AI systems

260
00:11:37,290 --> 00:11:40,410
that can pass the test of the real world,

261
00:11:40,410 --> 00:11:41,550
the real environment.

262
00:11:41,550 --> 00:11:42,870
And also for our students,

263
00:11:42,870 --> 00:11:45,000
and as we train students,

264
00:11:45,000 --> 00:11:47,800
they have the opportunity to
interact between these two.

265
00:11:47,800 --> 00:11:49,140
We're very happy and proud

266
00:11:49,140 --> 00:11:51,990
to have this addition in
our ecosystem, definitely.

267
00:11:51,990 --> 00:11:53,310
And also for our students,

268
00:11:53,310 --> 00:11:57,000
and as we train students,
they have the opportunity

269
00:11:57,000 --> 00:12:00,630
to interact between these two
faces, the lab on the campus

270
00:12:00,630 --> 00:12:04,650
and then the lab in grain,
in industrial use cases.

271
00:12:04,650 --> 00:12:07,350
This is only beneficial
to allow us to continue

272
00:12:07,350 --> 00:12:09,420
to contribute to this technology

273
00:12:09,420 --> 00:12:13,293
that we are inventing as
we are using in the field.

274
00:12:14,670 --> 00:12:15,503
- Okay.

275
00:12:15,503 --> 00:12:19,140
I really loved that Swiss cheese metaphor.

276
00:12:19,140 --> 00:12:20,460
- Mm. Me too.

277
00:12:20,460 --> 00:12:21,450
- I mean, he's right.

278
00:12:21,450 --> 00:12:23,400
Every system has holes, right?

279
00:12:23,400 --> 00:12:25,530
The skill is in finding them,

280
00:12:25,530 --> 00:12:29,100
and you can only really find
them if you actually test it

281
00:12:29,100 --> 00:12:31,500
in the real world and look
at where it's failing.

282
00:12:31,500 --> 00:12:34,050
- Yeah, I mean, but you don't
see that in the lab only.

283
00:12:34,050 --> 00:12:35,760
So research is important obviously,

284
00:12:35,760 --> 00:12:38,100
but the real world also teaches

285
00:12:38,100 --> 00:12:40,140
a lot of things about these applications.

286
00:12:40,140 --> 00:12:42,090
The lab and the field

287
00:12:42,090 --> 00:12:43,380
or the lab and the real world

288
00:12:43,380 --> 00:12:45,570
are two different platforms,
two different stages.

289
00:12:45,570 --> 00:12:48,570
And good news is there's a feedback loop.

290
00:12:48,570 --> 00:12:50,070
One learns from the other

291
00:12:50,070 --> 00:12:52,710
and improvements happen
over and over again.

292
00:12:52,710 --> 00:12:54,420
- Is that where we're heading, Scott,

293
00:12:54,420 --> 00:12:56,130
out into the real world?

294
00:12:56,130 --> 00:12:57,000
- Absolutely.

295
00:12:57,000 --> 00:13:00,090
And who better to take us
there than Riaz Raihan,

296
00:13:00,090 --> 00:13:03,153
who is the chief digital
officer at Trane Technologies?

297
00:13:04,020 --> 00:13:08,130
- So we've been planning
this for over 15 months

298
00:13:08,130 --> 00:13:12,870
and think of this lab as a
working laboratory, right?

299
00:13:12,870 --> 00:13:15,300
It's not just men and
women in white coats.

300
00:13:15,300 --> 00:13:18,810
It's a lot of very capable engineers

301
00:13:18,810 --> 00:13:21,120
and product managers and designers

302
00:13:21,120 --> 00:13:23,280
and people that really understand

303
00:13:23,280 --> 00:13:27,210
how AI is used in the HVAC industry.

304
00:13:27,210 --> 00:13:28,860
About two years ago,

305
00:13:28,860 --> 00:13:31,650
we acquired a wonderful
company called BrainBox

306
00:13:31,650 --> 00:13:36,650
because they were by far
the leader in this space.

307
00:13:36,870 --> 00:13:39,930
Some of the results
they were able to drive,

308
00:13:39,930 --> 00:13:42,930
tangible results, were very impressive.

309
00:13:42,930 --> 00:13:45,930
Customers were confirming
that they were saving

310
00:13:45,930 --> 00:13:47,610
15, 20, in some cases

311
00:13:47,610 --> 00:13:51,510
even 30 and 40%, and consistently.

312
00:13:51,510 --> 00:13:53,580
And then as we started integrating them

313
00:13:53,580 --> 00:13:56,587
into the fabric of Trane
Technologies, we said,

314
00:13:56,587 --> 00:14:00,090
"Why don't we truly make
this a showcase lab?

315
00:14:00,090 --> 00:14:04,260
Why don't we leverage the
location of BrainBox?"

316
00:14:04,260 --> 00:14:07,080
And BrainBox is based in Montreal,

317
00:14:07,080 --> 00:14:11,010
which is one of the world's top three hubs

318
00:14:11,010 --> 00:14:12,210
for AI development,

319
00:14:12,210 --> 00:14:15,660
the other two being
Silicon Valley and London

320
00:14:15,660 --> 00:14:17,190
and Cambridge in England.

321
00:14:17,190 --> 00:14:19,770
The lab itself is a working lab.

322
00:14:19,770 --> 00:14:22,620
We have real people doing real work.

323
00:14:22,620 --> 00:14:24,270
And when our customers

324
00:14:24,270 --> 00:14:28,380
or our prospects or
custodians of sustainability

325
00:14:28,380 --> 00:14:29,970
visit this lab,

326
00:14:29,970 --> 00:14:32,460
what they find is an opportunity

327
00:14:32,460 --> 00:14:35,430
to truly experience

328
00:14:35,430 --> 00:14:39,300
how this wonderful technology
works, to see it in action.

329
00:14:39,300 --> 00:14:41,790
They can walk through the lab itself

330
00:14:41,790 --> 00:14:44,280
and go meet the men and women

331
00:14:44,280 --> 00:14:45,960
who are building this technology

332
00:14:45,960 --> 00:14:48,990
and ask questions, interact with them,

333
00:14:48,990 --> 00:14:50,880
understand what's driving them,

334
00:14:50,880 --> 00:14:53,790
understand what are
the latest technologies

335
00:14:53,790 --> 00:14:54,690
that they're leveraging,

336
00:14:54,690 --> 00:14:58,230
understand is it predictive
AI, classical AI,

337
00:14:58,230 --> 00:15:00,603
generative AI, a combination of those.

338
00:15:01,631 --> 00:15:04,500
It's a lab that has a meaningful impact

339
00:15:04,500 --> 00:15:06,030
and it's a lab that allows us

340
00:15:06,030 --> 00:15:08,730
to truly engage with our stakeholders.

341
00:15:08,730 --> 00:15:09,960
- I'd love to also meet the people

342
00:15:09,960 --> 00:15:12,090
behind some of these AI tools

343
00:15:12,090 --> 00:15:13,770
and I think that's a unique moment.

344
00:15:13,770 --> 00:15:16,980
So Riaz, is this, like,
an academic exercise

345
00:15:16,980 --> 00:15:19,037
that we're conducting in
Montreal or is it something else?

346
00:15:19,037 --> 00:15:21,630
Are we actually producing
something for the world?

347
00:15:21,630 --> 00:15:24,210
- This is not just academic.
It's very commercial.

348
00:15:24,210 --> 00:15:27,150
In Montreal, we do a lot
of fundamental research,

349
00:15:27,150 --> 00:15:29,250
but then we create products,

350
00:15:29,250 --> 00:15:30,869
and these products are built

351
00:15:30,869 --> 00:15:32,850
for real customers in the real world.

352
00:15:32,850 --> 00:15:34,500
And we ship these products to them,

353
00:15:34,500 --> 00:15:38,190
we deploy these products,
we monetize these products,

354
00:15:38,190 --> 00:15:39,660
and then we operate these products

355
00:15:39,660 --> 00:15:41,580
24/7 for these customers.

356
00:15:41,580 --> 00:15:43,620
So it's a living lab.

357
00:15:43,620 --> 00:15:45,150
It's a very commercially focused lab,

358
00:15:45,150 --> 00:15:46,620
and it's solving real problems

359
00:15:46,620 --> 00:15:48,750
for real customers in the real world.

360
00:15:48,750 --> 00:15:51,090
- Yeah, Riaz, I hated that I had to miss

361
00:15:51,090 --> 00:15:54,093
the recent ribbon cutting
opening of the lab.

362
00:15:54,960 --> 00:15:57,180
I hope to get there at some
time in the near future.

363
00:15:57,180 --> 00:15:59,010
- Well, Scott, you're in luck

364
00:15:59,010 --> 00:16:01,650
because we put together
a little video montage.

365
00:16:01,650 --> 00:16:03,960
I know you missed it and we missed you,

366
00:16:03,960 --> 00:16:07,200
but I'd love to show you
what the lab is all about.

367
00:16:07,200 --> 00:16:10,200
But I thought I would
show you a few things

368
00:16:10,200 --> 00:16:14,310
that we're going to talk about
tomorrow at the big ceremony.

369
00:16:14,310 --> 00:16:15,420
First of all, I want you

370
00:16:15,420 --> 00:16:18,060
to take a look at this beautiful lab

371
00:16:18,060 --> 00:16:20,430
which has all of the cool tech

372
00:16:20,430 --> 00:16:22,920
that BrainBox has brought to market.

373
00:16:22,920 --> 00:16:25,470
The first station talks
a little bit about ARIA.

374
00:16:25,470 --> 00:16:26,850
This is the virtual engineer

375
00:16:26,850 --> 00:16:28,890
where you can actually converse with ARIA

376
00:16:28,890 --> 00:16:31,830
either in text-to-text or voice-to-text

377
00:16:31,830 --> 00:16:34,290
or use the pre-approved prompts

378
00:16:34,290 --> 00:16:36,780
and really get responses
from ARIA in real time.

379
00:16:36,780 --> 00:16:38,100
That was really cool.

380
00:16:38,100 --> 00:16:39,240
The second station

381
00:16:39,240 --> 00:16:42,300
is really about Trane Autonomous Controls

382
00:16:42,300 --> 00:16:44,070
or Trane AI Control.

383
00:16:44,070 --> 00:16:47,550
This is when you get to see a
lot of different algorithms.

384
00:16:47,550 --> 00:16:51,060
And what I loved about
the approach the team used

385
00:16:51,060 --> 00:16:52,930
was to have each cylinder

386
00:16:53,790 --> 00:16:56,340
on that table represent
a different algorithm.

387
00:16:56,340 --> 00:16:58,470
And then you could pick the
one you want to talk about,

388
00:16:58,470 --> 00:17:00,690
for example, Hydra or Hercules,

389
00:17:00,690 --> 00:17:02,400
and learn a little bit more about it,

390
00:17:02,400 --> 00:17:06,840
but also learn how these
algorithms work together in unison

391
00:17:06,840 --> 00:17:10,530
to solve problems for
complex HVAC systems.

392
00:17:10,530 --> 00:17:14,730
Now, the other station
was all about Cloud BMS.

393
00:17:14,730 --> 00:17:19,440
And Cloud BMS is this cloud-based
building management system

394
00:17:19,440 --> 00:17:22,620
that truly brings together
the power of ARIA,

395
00:17:22,620 --> 00:17:24,360
the power of AI control,

396
00:17:24,360 --> 00:17:27,510
and the power of cloud-based
BMS into one package.

397
00:17:27,510 --> 00:17:31,290
We saw some great demos of how
customers are using Cloud BMS

398
00:17:31,290 --> 00:17:33,210
to drive tremendous value.

399
00:17:33,210 --> 00:17:37,980
The next station was all
about M&V, measure and verify.

400
00:17:37,980 --> 00:17:40,260
As a matter of fact, there's
some people there right now,

401
00:17:40,260 --> 00:17:41,850
and this was all about understanding

402
00:17:41,850 --> 00:17:44,460
how customers are using BrainBox AI,

403
00:17:44,460 --> 00:17:46,260
the kind of benefits they get.

404
00:17:46,260 --> 00:17:48,420
And many of these customers
have actually allowed us

405
00:17:48,420 --> 00:17:50,970
to use their names and
use specific numbers

406
00:17:50,970 --> 00:17:54,000
about the types of savings
they've gotten on the cost side,

407
00:17:54,000 --> 00:17:57,150
but also how much carbon
emissions they have reduced.

408
00:17:57,150 --> 00:17:59,160
And I know you really care about that.

409
00:17:59,160 --> 00:18:04,160
And finally, there's a station
where we have a robotic arm

410
00:18:04,230 --> 00:18:09,060
of an actual robot from
one of Trane's factories.

411
00:18:09,060 --> 00:18:12,750
And this robot was adopted by BrainBox

412
00:18:12,750 --> 00:18:16,380
and brought into the lab here
and given a place of pride.

413
00:18:16,380 --> 00:18:18,270
And that's the station
where you learn a lot

414
00:18:18,270 --> 00:18:21,180
about all the different digital offerings

415
00:18:21,180 --> 00:18:23,340
that we have at Trane Technologies.

416
00:18:23,340 --> 00:18:27,300
- Last season, we had
Jean-Simon on the podcast

417
00:18:27,300 --> 00:18:28,133
where we talked

418
00:18:28,133 --> 00:18:31,500
about BrainBox's flagship
autonomous building controls,

419
00:18:31,500 --> 00:18:33,270
Riaz, which I think you know well,

420
00:18:33,270 --> 00:18:35,400
and I've heard you speak about this

421
00:18:35,400 --> 00:18:38,460
other places about the
potential and the opportunity

422
00:18:38,460 --> 00:18:41,340
for some of these tools
to truly make a difference

423
00:18:41,340 --> 00:18:43,950
in more sustainable operated buildings.

424
00:18:43,950 --> 00:18:46,530
So what's new in that space now?

425
00:18:46,530 --> 00:18:47,460
- A couple of years ago,

426
00:18:47,460 --> 00:18:50,010
a lot of our focus was on the air side.

427
00:18:50,010 --> 00:18:51,150
Customers really wanted

428
00:18:51,150 --> 00:18:53,280
to leverage this on
the water side as well.

429
00:18:53,280 --> 00:18:57,090
I'm happy to report
that we have algorithms

430
00:18:57,090 --> 00:18:59,160
that are now making a big impact.

431
00:18:59,160 --> 00:18:59,993
You know, we are looking

432
00:18:59,993 --> 00:19:01,517
at differential pressure optimization.

433
00:19:01,517 --> 00:19:02,400
We are looking

434
00:19:02,400 --> 00:19:05,100
at the condenser water
temperature optimization.

435
00:19:05,100 --> 00:19:07,560
We're looking at optimizing
the operation of the valves.

436
00:19:07,560 --> 00:19:08,430
And the combination

437
00:19:08,430 --> 00:19:11,880
of our traditional
chiller plant optimization

438
00:19:11,880 --> 00:19:14,460
through controls and BrainBox AI

439
00:19:14,460 --> 00:19:17,310
is proving to be extremely powerful.

440
00:19:17,310 --> 00:19:18,870
Let me give you an example.

441
00:19:18,870 --> 00:19:22,290
We have a major customer in Singapore.

442
00:19:22,290 --> 00:19:25,890
Singapore is a great place

443
00:19:25,890 --> 00:19:28,710
to test this kind of technology.

444
00:19:28,710 --> 00:19:32,790
They ran this pilot with
BrainBox for over three months.

445
00:19:32,790 --> 00:19:34,710
I think it's coming up to six months now.

446
00:19:34,710 --> 00:19:36,690
And during that period, they were able

447
00:19:36,690 --> 00:19:40,020
to show significant improvement

448
00:19:40,020 --> 00:19:42,990
in not only the runtime of the chillers,

449
00:19:42,990 --> 00:19:46,440
but also in the energy
consumed as a result of that.

450
00:19:46,440 --> 00:19:48,210
And they're in the process of figuring out

451
00:19:48,210 --> 00:19:49,950
the exact savings, but I can tell you

452
00:19:49,950 --> 00:19:53,070
it is well above what we
expected and very impressive,

453
00:19:53,070 --> 00:19:55,470
you know, we are already
in the 10, 20% range.

454
00:19:55,470 --> 00:19:56,303
It could be higher.

455
00:19:56,303 --> 00:19:58,920
And when customers asked for this,

456
00:19:58,920 --> 00:20:00,900
I think it was great for the BrainBox team

457
00:20:00,900 --> 00:20:02,490
to go off and work on it.

458
00:20:02,490 --> 00:20:05,550
We now have working algorithms.

459
00:20:05,550 --> 00:20:08,163
We are in the process
of a commercial rollout.

460
00:20:09,000 --> 00:20:11,340
We are starting with some of
the most important markets

461
00:20:11,340 --> 00:20:13,890
for us in the United States and Europe.

462
00:20:13,890 --> 00:20:15,420
So you're gonna hear a lot about this.

463
00:20:15,420 --> 00:20:19,080
- There's a lot of promise
that AI has for the world,

464
00:20:19,080 --> 00:20:21,090
and maybe the lab that you mentioned

465
00:20:21,090 --> 00:20:24,570
at the beginning of this
discussion is testing things

466
00:20:24,570 --> 00:20:26,580
and we see great things in a lab setting,

467
00:20:26,580 --> 00:20:27,960
but what happens in the real world?

468
00:20:27,960 --> 00:20:29,610
Can we trust the things that,

469
00:20:29,610 --> 00:20:31,980
I mean, you just gave some
figures about Singapore,

470
00:20:31,980 --> 00:20:34,650
a 10 to 20% improvement.

471
00:20:34,650 --> 00:20:37,500
Can customers, can building managers

472
00:20:37,500 --> 00:20:41,790
trust these tools that are AI related?

473
00:20:41,790 --> 00:20:46,200
- I think if done properly,
yes, they can, and they should.

474
00:20:46,200 --> 00:20:50,970
Our approach is founded on
three key principles, okay?

475
00:20:50,970 --> 00:20:54,150
The first principle is secure by design.

476
00:20:54,150 --> 00:20:55,530
This is very important.

477
00:20:55,530 --> 00:20:57,480
When we start designing algorithms,

478
00:20:57,480 --> 00:20:59,490
when we start designing solutions,

479
00:20:59,490 --> 00:21:03,060
security is the first thing
we look at, not the last.

480
00:21:03,060 --> 00:21:05,640
We start with security, data security,

481
00:21:05,640 --> 00:21:08,640
physical access security,
identity security, et cetera.

482
00:21:08,640 --> 00:21:09,630
So we start with that.

483
00:21:09,630 --> 00:21:13,320
The second is strong compliance

484
00:21:13,320 --> 00:21:15,030
and going beyond compliance

485
00:21:15,030 --> 00:21:19,200
with every regulation that
exists across the globe.

486
00:21:19,200 --> 00:21:21,750
The third is partnerships.

487
00:21:21,750 --> 00:21:25,260
We are part with some of the
world's leading organizations.

488
00:21:25,260 --> 00:21:28,320
We are working with some
world-class organizations,

489
00:21:28,320 --> 00:21:30,360
and these folks are creating

490
00:21:30,360 --> 00:21:34,710
amazing capability around
AI trust and AI security.

491
00:21:34,710 --> 00:21:37,941
So if you put those three things
together, secure by design,

492
00:21:37,941 --> 00:21:42,360
which is a choice, I call
that the triangle of trust,

493
00:21:42,360 --> 00:21:45,690
and it allows us to truly put
our customers in the middle

494
00:21:45,690 --> 00:21:47,280
and put the triangle around them,

495
00:21:47,280 --> 00:21:49,740
make sure we are building
trust from day one.

496
00:21:49,740 --> 00:21:52,320
- There's a open secret, I think, with AI.

497
00:21:52,320 --> 00:21:53,490
I think a lot of people have heard

498
00:21:53,490 --> 00:21:56,220
that hard to make any
money with it though,

499
00:21:56,220 --> 00:21:58,680
not even sure if companies
are making money with AI.

500
00:21:58,680 --> 00:22:00,300
So what's your take on that?

501
00:22:00,300 --> 00:22:03,390
- If you look across the
AI ecosystem globally

502
00:22:03,390 --> 00:22:07,110
of the hundreds of
thousands of AI companies

503
00:22:07,110 --> 00:22:10,950
that have mushroomed over
the past four to five years,

504
00:22:10,950 --> 00:22:15,950
less than 5% have managed to
monetize their technology.

505
00:22:16,470 --> 00:22:17,820
When we acquired BrainBox,

506
00:22:17,820 --> 00:22:19,740
they were one of the few companies

507
00:22:19,740 --> 00:22:23,520
that were actually monetizing
their software, their AI.

508
00:22:23,520 --> 00:22:24,990
So they are in that 5%.

509
00:22:24,990 --> 00:22:27,450
What makes this company successful

510
00:22:27,450 --> 00:22:29,220
and what puts it in that 5%

511
00:22:29,220 --> 00:22:32,550
when everybody else is
not able to even monetize?

512
00:22:32,550 --> 00:22:33,840
It turns out, Scott,

513
00:22:33,840 --> 00:22:35,880
I think there are two important factors.

514
00:22:35,880 --> 00:22:37,140
Factor number one,

515
00:22:37,140 --> 00:22:42,120
you need to be able to
solve a specific problem

516
00:22:42,120 --> 00:22:45,600
that actually has a monetary payback.

517
00:22:45,600 --> 00:22:49,020
And if you can do that, you
have a much better chance

518
00:22:49,020 --> 00:22:51,420
of customers wanting to
pay for that solution.

519
00:22:51,420 --> 00:22:54,210
Number two, your solution itself

520
00:22:54,210 --> 00:22:58,140
has to be very focused
on a specific domain.

521
00:22:58,140 --> 00:23:02,400
Just leveraging general purpose
AI that can do many things

522
00:23:02,400 --> 00:23:04,860
doesn't tend to lead to
successful AI companies.

523
00:23:04,860 --> 00:23:07,680
Having an AI company that
does something very well

524
00:23:07,680 --> 00:23:09,780
and, you know, very specific

525
00:23:09,780 --> 00:23:12,060
and solves it with domain expertise,

526
00:23:12,060 --> 00:23:15,510
that seems to be the holy
grail of successful companies.

527
00:23:15,510 --> 00:23:17,190
And we see that as a trend.

528
00:23:17,190 --> 00:23:20,280
So BrainBox obviously is
optimizing HVAC operations,

529
00:23:20,280 --> 00:23:21,540
and that's a very specific domain.

530
00:23:21,540 --> 00:23:23,970
They bring a lot of domain expertise.

531
00:23:23,970 --> 00:23:25,170
They're doing it really well.

532
00:23:25,170 --> 00:23:27,120
So they picked a problem worth solving

533
00:23:27,120 --> 00:23:28,560
and they're solving it really well, right?

534
00:23:28,560 --> 00:23:30,990
So those two things come
together, they're monetizing.

535
00:23:30,990 --> 00:23:33,420
The other piece is companies

536
00:23:33,420 --> 00:23:36,810
that solve problems at scale

537
00:23:36,810 --> 00:23:38,910
tend to be far more successful

538
00:23:38,910 --> 00:23:41,760
than companies that are trying
to solve very small problems.

539
00:23:41,760 --> 00:23:44,040
So to come back to your original question,

540
00:23:44,040 --> 00:23:46,230
I think those are the
reasons why monetizing

541
00:23:46,230 --> 00:23:49,350
is very difficult and then
break even is the next bar,

542
00:23:49,350 --> 00:23:51,210
and then after that is
actual profitability.

543
00:23:51,210 --> 00:23:54,030
So those are kind of the
stops along the journey,

544
00:23:54,030 --> 00:23:56,970
but getting to monetization
is not trivial.

545
00:23:56,970 --> 00:23:58,200
- You're sort of speaking my language

546
00:23:58,200 --> 00:24:00,750
as a sustainability leader,
helping solve something,

547
00:24:00,750 --> 00:24:02,220
helping something be more productive,

548
00:24:02,220 --> 00:24:04,320
helping something
operate more efficiently.

549
00:24:04,320 --> 00:24:07,200
Can we talk about a
sustainability-related concern

550
00:24:07,200 --> 00:24:09,660
that a lot of our
listeners and viewers have?

551
00:24:09,660 --> 00:24:13,170
And that is around the
amount of energy and water

552
00:24:13,170 --> 00:24:15,630
that it takes to run these data centers.

553
00:24:15,630 --> 00:24:16,890
What are your views on that,

554
00:24:16,890 --> 00:24:18,720
and where do you think this is going?

555
00:24:18,720 --> 00:24:21,930
- The data center discussion is real.

556
00:24:21,930 --> 00:24:25,560
I mean, data centers
are a massive consumer

557
00:24:25,560 --> 00:24:27,270
of electric power, no question,

558
00:24:27,270 --> 00:24:28,350
and of water.
- That's right.

559
00:24:28,350 --> 00:24:29,250
- And of land.

560
00:24:29,250 --> 00:24:32,730
I've been inside and on top
of and around data centers

561
00:24:32,730 --> 00:24:33,563
for a long time.

562
00:24:33,563 --> 00:24:36,090
And you learn a lot when you spend

563
00:24:36,090 --> 00:24:39,330
that much time in data
centers and working with them.

564
00:24:39,330 --> 00:24:41,430
So let's start with the positive.

565
00:24:41,430 --> 00:24:43,290
The first thing one has to
remember is data centers

566
00:24:43,290 --> 00:24:46,080
have solved a real problem for humanity.

567
00:24:46,080 --> 00:24:49,290
They have created an
infrastructure that has allowed us

568
00:24:49,290 --> 00:24:51,540
to bring computing to the masses

569
00:24:51,540 --> 00:24:54,300
at a cost which is acceptable.

570
00:24:54,300 --> 00:24:57,360
As we get into the AI age, the good news

571
00:24:57,360 --> 00:25:00,240
is that data centers
are creating capability

572
00:25:00,240 --> 00:25:03,090
that can solve the
problem that they create.

573
00:25:03,090 --> 00:25:04,590
So if you look at the total amount

574
00:25:04,590 --> 00:25:06,663
of energy the world uses globally,

575
00:25:07,620 --> 00:25:09,540
the EPA publishes these numbers

576
00:25:09,540 --> 00:25:11,550
and you can actually go and see

577
00:25:11,550 --> 00:25:12,450
and you kind of figure out

578
00:25:12,450 --> 00:25:15,030
how much of that energy is used for HVAC.

579
00:25:15,030 --> 00:25:16,980
It's a large chunk, about 10% globally.

580
00:25:16,980 --> 00:25:18,720
Now, you have a number x.

581
00:25:18,720 --> 00:25:21,960
Question, with technologies like BrainBox,

582
00:25:21,960 --> 00:25:23,910
if we were to apply them globally,

583
00:25:23,910 --> 00:25:25,620
which is an AI technology,

584
00:25:25,620 --> 00:25:27,630
how much of that x could you save?

585
00:25:27,630 --> 00:25:30,360
Well, you could save at least 20 to 30%.

586
00:25:30,360 --> 00:25:31,230
Let's take 20%.

587
00:25:31,230 --> 00:25:33,030
So you're saving 20% of x.

588
00:25:33,030 --> 00:25:33,863
What is that number?

589
00:25:33,863 --> 00:25:35,100
Well, that's a big number.

590
00:25:35,100 --> 00:25:38,310
That number is more than all the power

591
00:25:38,310 --> 00:25:42,090
used by every data center
on the planet times four.

592
00:25:42,090 --> 00:25:43,740
So you pause and think about that.

593
00:25:43,740 --> 00:25:48,360
This is an AI solution born
and resident in a data center,

594
00:25:48,360 --> 00:25:49,950
or in data centers,

595
00:25:49,950 --> 00:25:53,190
that can help humanity
save four times more energy

596
00:25:53,190 --> 00:25:56,400
than all data centers combined

597
00:25:56,400 --> 00:25:58,890
if applied at scale consistently.

598
00:25:58,890 --> 00:26:00,210
That's a big if.

599
00:26:00,210 --> 00:26:02,280
But the point of this thought experiment

600
00:26:02,280 --> 00:26:05,250
is to look at the
opportunity and to say, "Wow,

601
00:26:05,250 --> 00:26:06,990
that is the power of AI."

602
00:26:06,990 --> 00:26:10,140
So while AI, you know, is vilified

603
00:26:10,140 --> 00:26:12,420
often and for all the right reasons,

604
00:26:12,420 --> 00:26:15,300
I do believe that there's
also massive opportunity

605
00:26:15,300 --> 00:26:17,370
in looking at this space as a solution

606
00:26:17,370 --> 00:26:19,230
to solve the problem it creates.

607
00:26:19,230 --> 00:26:20,910
Because if you do solve
the problem, Scott,

608
00:26:20,910 --> 00:26:22,530
if you kind of step back for a minute

609
00:26:22,530 --> 00:26:24,570
and you look at 5, 10 years from today,

610
00:26:24,570 --> 00:26:27,210
let's say we apply this
kind of technology,

611
00:26:27,210 --> 00:26:29,160
the world needs more HVAC.

612
00:26:29,160 --> 00:26:31,170
And as HVAC becomes more prevalent,

613
00:26:31,170 --> 00:26:33,120
especially in the Eastern Hemisphere,

614
00:26:33,120 --> 00:26:36,090
the world will consume
more energy around HVAC,

615
00:26:36,090 --> 00:26:38,550
and that represents a massive opportunity

616
00:26:38,550 --> 00:26:42,600
for us as custodians and
stewards of this industry

617
00:26:42,600 --> 00:26:45,900
to leverage AI solutions to
reduce their consumption.

618
00:26:45,900 --> 00:26:48,210
And eventually we could get to a point

619
00:26:48,210 --> 00:26:50,340
where the reduced power

620
00:26:50,340 --> 00:26:53,400
actually equals or is better
than the consumed power.

621
00:26:53,400 --> 00:26:57,510
And that is truly the
promise of such technologies.

622
00:26:57,510 --> 00:26:58,950
- Well, look at that.

623
00:26:58,950 --> 00:27:02,160
We have just come full circle.

624
00:27:02,160 --> 00:27:04,500
This has been the journey
on this episode, right?

625
00:27:04,500 --> 00:27:07,800
We started with Foutse talking
us through the research

626
00:27:07,800 --> 00:27:10,890
and the testing and the
Swiss cheese analogy,

627
00:27:10,890 --> 00:27:13,410
and here's Riaz coming and talking to us

628
00:27:13,410 --> 00:27:15,630
about real-world deployment.

629
00:27:15,630 --> 00:27:18,630
I've walked away
realizing that the AI lab,

630
00:27:18,630 --> 00:27:20,400
it isn't just about theory.

631
00:27:20,400 --> 00:27:23,250
It's about building and
testing real solutions

632
00:27:23,250 --> 00:27:25,710
that can be deployed immediately

633
00:27:25,710 --> 00:27:28,830
and really help solve some big challenges.

634
00:27:28,830 --> 00:27:32,280
And this is the proof of the
power of partnership, right?

635
00:27:32,280 --> 00:27:34,620
- Yeah, and that is the
bigger picture here.

636
00:27:34,620 --> 00:27:36,600
I mean, connecting sustainability goals

637
00:27:36,600 --> 00:27:39,300
with operational outcomes,

638
00:27:39,300 --> 00:27:42,270
this is where we're going to
see the big impacts happening,

639
00:27:42,270 --> 00:27:44,010
and that's where things
get really interesting.

640
00:27:44,010 --> 00:27:46,200
I mean, this is linked to
Riaz's thought experiment

641
00:27:46,200 --> 00:27:51,153
where energy that's not
used is the best path.

642
00:27:52,170 --> 00:27:56,010
HVAC is roughly 10% of the
world's energy consumption.

643
00:27:56,010 --> 00:27:58,863
And if we apply AI tools to that issue,

644
00:27:59,970 --> 00:28:02,910
then we're going to save
four times the energy

645
00:28:02,910 --> 00:28:04,410
that data centers consume globally.

646
00:28:04,410 --> 00:28:07,170
I mean, that is the amazing moment.

647
00:28:07,170 --> 00:28:09,870
This has been "Healthy
Spaces" with me, Scott Tew,

648
00:28:09,870 --> 00:28:11,910
and my co-host Dominique Silva.

649
00:28:11,910 --> 00:28:14,220
We're back in two weeks
with another episode,

650
00:28:14,220 --> 00:28:17,583
so be sure to like and
subscribe so you don't miss out.