Popular in Professional

Architecture

Connecting Cities and Nature

University of Virginia Timothy Beatley

Human Rights Law

Human Rights and Corruption: Problems and Potential of Individualising a Systemic Problem

Max Planck Institute Anne Peters

Finance

Chinese Exchange Rate Policy: Lessons for Global Investors

University of California San Diego Michael Melvin

Law

A statistical approach to law school citation rankings

University of Virginia Joshua Fischman

Education

Firm Performance Pay as Insurance against Promotion Risk

Stockholm School of Economics Alvin Chen

Constitutional Law

What is constitutional ideology?

University of Oxford Nick Barber

Machine Learning: Architecture in the age of Artificial Intelligence

View more insights in Professional - Architecture

View more figures from Yale University

View the page of Phillip Bernstein

Read the Study

Next in Architecture

Connecting Cities and Nature

Timothy Beatley | University of Virginia

Cities and Rivers

Iñaki N Alday Sanz | Tulane University

Phillip Bernstein outlines strategies for architects to adapt to the transformative impacts of AI and machine learning, emphasizing the need for the profession to embrace new technologies that will reshape design processes and roles, ensuring future relevance and success

Image courtesy of interviewee. October 10, 2024

Log-in or Sign-up to Faculti
Currently viewing this subject insight as a guest. You have insight(s) remaining for this month. Login to view 8000+ figures on the platform.

Generating Workspace...

Export as PDF

Language:

Export as PDF

Export as PDF
Here are five key points from the video:

1. **Integration of Technology in Architecture**: The author emphasizes the need for architects to embrace technology, particularly AI, not just for productivity but to enhance the value of their work. He argues that architects can leverage technology to automate routine tasks, enabling them to focus on complex design processes.

2. **Concerns About Job Displacement**: There is a fear among architects that AI could replace them. The author counters this notion by suggesting that instead of fearing displacement, architects should see AI as a tool that can help them provide more valuable outputs, such as predicting building performance.

3. **Ethics and Data in the Architecture Industry**: The video discusses the ethical implications of material sourcing and the lack of data transparency in building materials. The author advocates for a cooperative data commons to enable better insights into the ethical sourcing of materials, which is currently obscured in the existing data infrastructures.

4. **Challenges of Data Utilization**: The building industry faces the challenge of insufficient data to train AI models effectively. The author proposes focusing on discrete tasks where data is more readily available, such as sustainability metrics, to create meaningful insights rather than attempting to amass vast amounts of architectural data.

5. **Future of Architectural Practice and Business Models**: The author argues for a shift in the business model of architecture from fixed-fee transactional work to a focus on creating value through insights and predictions. He believes that architects should evolve into the “making things happen” business, where they charge for their predictive capabilities and the value they deliver rather than merely for design services.

Export as PDF
Here are some key quotes from the video , highlighting the main themes and ideas discussed:

1. **Interdisciplinary Roots**: "I was studying architecture, but also computer science. And so I realized this was an opportunity for me to sort of connect the dots."

2. **Technology's Potential**: "I argued in my book that a, that's not true, and b, worrying about that is to miss the big opportunity of the use of the technology, which is not just around productivity, but around increasing the value of what architects actually do."

3. **Role of Architects**: "By having the machines do things that they're good at, like organize and understand data and evaluate it, and have architects use that as an input to what we are good at."

4. **Automation in Architecture**: "A lot of that stuff will get automated."

5. **Business Model Implications**: "Architecture is largely transacted on fixed fees, which are highly commoditized and highly competitive."

6. **Value Proposition**: "If I say to you, I'm an AI-enabled architect... those are valuable outputs that we should get paid for."

7. **Ethics in Material Supply Chains**: "One of the things... I've been working on is trying to understand the ethics of the material supply chain and... modern slavery."

8. **Data Challenges**: "We can't use the systems to reify existing data infrastructures 'cause they don't give us enough insight into what's going on in practice."

9. **Need for Cooperative Data Commons**: "I think this idea of picking much more limited discrete tasks and training on those discrete tasks is the right... way to go."

10. **AI's Role in Sustainability**: "Sustainability happens to be one realm where we have a lot of data... and I think a combination of those things could provide some terrific platforms for providing sustainability insight."

11. **Complexity of Building Data**: "The IP questions... you're gonna have to immunize me from its illicit use."

12. **Skepticism towards Technology**: "I'm a skeptical thinker about technology implementation... I predict what's gonna happen is we're gonna have three or four years of anecdotal, episodic..."

13. **Future of Architecture's Engagement with Technology**: "I still think that part is right. My examples are outta date... but the central argument I think still obtains."

These quotes encapsulate the author's reflections on the intersection of architecture and technology, the ethical considerations inherent in modern practices, and the evolving role of architects in a technologically advancing world.

X-TIMESTAMP-MAP=LOCAL:00:00:00.000,MPEGTS:90000

1
00:00:00.000 --> 00:00:02.525
When I was an undergraduate in the seventies,

2
00:00:03.165 --> 00:00:05.965
I had taken several courses about what we now know

3
00:00:06.345 --> 00:00:07.885
as natural language processing.

4
00:00:07.965 --> 00:00:10.325
I was studying architecture, but also computer science.

5
00:00:11.065 --> 00:00:13.685
And so I realized this was an opportunity for me to sort

6
00:00:13.685 --> 00:00:15.045
of connect the dots.

7
00:00:15.385 --> 00:00:18.645
I'm also very interested in the role technology plays in

8
00:00:19.425 --> 00:00:23.605
how architects operate in systems of design,

9
00:00:24.205 --> 00:00:25.645
construction and asset operation.

10
00:00:26.265 --> 00:00:30.605
And so my previous, both my previous academic work

11
00:00:30.605 --> 00:00:33.685
and my commercial work, when I was a, an employee

12
00:00:33.685 --> 00:00:36.965
of a large software company, was to work on that problem.

13
00:00:37.105 --> 00:00:40.445
To work on that question, I was asked to write it in 2020

14
00:00:40.545 --> 00:00:43.445
and worked on it in 2021, thanks to the pandemic.

15
00:00:44.145 --> 00:00:47.885
Um, the, the, the efficacy

16
00:00:47.885 --> 00:00:50.245
that technology wasn't very well understood by architects.

17
00:00:50.245 --> 00:00:54.935
And of course there was an immediate concern that it faced

18
00:00:55.085 --> 00:00:56.895
that we faced an existential threat

19
00:00:57.555 --> 00:00:59.255
to be replaced by algorithms.

20
00:01:00.075 --> 00:01:04.775
And I argued in my book that a, that's not true

21
00:01:04.775 --> 00:01:06.735
because it's really hard to be an architect.

22
00:01:07.605 --> 00:01:11.545
And b, worrying about that is to miss the big opportunity

23
00:01:12.085 --> 00:01:14.825
of the use of the technology, which is not just

24
00:01:14.825 --> 00:01:16.105
around productivity,

25
00:01:16.205 --> 00:01:18.465
but around increasing the value of

26
00:01:18.465 --> 00:01:19.665
what architects actually do.

27
00:01:20.325 --> 00:01:22.905
By having the machines do things that they're good at,

28
00:01:23.215 --> 00:01:26.185
like organize and understand data and evaluate it,

29
00:01:26.445 --> 00:01:29.265
and have architects use that as an input to

30
00:01:29.265 --> 00:01:33.945
what we are good at, which is taking large complex contexts

31
00:01:33.965 --> 00:01:38.665
and synthesizing them into answers using a design process

32
00:01:39.485 --> 00:01:42.505
and making trade-offs and understanding

33
00:01:42.605 --> 00:01:43.825
and projecting into the future.

34
00:01:43.975 --> 00:01:46.825
I've been in the architecture world for four decades now.

35
00:01:47.295 --> 00:01:48.865
There's a tendency for architects

36
00:01:48.865 --> 00:01:49.985
to take every new technology

37
00:01:50.005 --> 00:01:53.145
and try to make shapes for images with it.

38
00:01:53.445 --> 00:01:54.545
It doesn't matter what it is.

39
00:01:55.005 --> 00:01:57.305
The first instinct is to make, is

40
00:01:57.305 --> 00:01:59.705
to make new interesting shapes and make images.

41
00:01:59.765 --> 00:02:01.225
And there's nothing wrong with that.

42
00:02:01.855 --> 00:02:04.265
It's just the problem is it, with ai,

43
00:02:04.295 --> 00:02:05.545
it's kind of a distraction.

44
00:02:05.975 --> 00:02:08.105
It's a, it's a, it's a place

45
00:02:08.105 --> 00:02:09.985
where an opportunity might get missed.

46
00:02:10.085 --> 00:02:13.185
And while we're missing that opportunity, I do think

47
00:02:13.185 --> 00:02:16.585
that some swath of the kinds of day-to-day work

48
00:02:16.585 --> 00:02:20.545
that architects do, transactional correspondence,

49
00:02:21.465 --> 00:02:24.905
checking documents for graphic conformance,

50
00:02:24.905 --> 00:02:26.185
three dimensional coordination

51
00:02:26.185 --> 00:02:27.905
of things in space, which is a lot of work.

52
00:02:28.435 --> 00:02:31.425
Let's make sure that this doc doesn't run into this beam

53
00:02:32.005 --> 00:02:35.345
or this, there's enough space for this guy

54
00:02:35.345 --> 00:02:36.625
to get a paintbrush in here.

55
00:02:37.345 --> 00:02:38.985
A lot of that stuff will get automated.

56
00:02:39.885 --> 00:02:43.505
So this is why part of my work is around business models.

57
00:02:43.865 --> 00:02:47.425
'cause I feel like if these tools make, make us

58
00:02:47.425 --> 00:02:49.185
as architects, a lot more productive

59
00:02:49.775 --> 00:02:53.735
architecture is largely transacted on fixed fees,

60
00:02:54.625 --> 00:02:57.215
which are highly commoditized and highly competitive.

61
00:02:57.555 --> 00:02:58.775
And so when everybody starts to

62
00:02:58.775 --> 00:03:00.965
Get more, more productive, the fees are gonna go down

63
00:03:01.465 --> 00:03:03.365
unless there's a different value proposition.

64
00:03:03.985 --> 00:03:07.045
But if I say to you, I'm an AI enabled architect

65
00:03:07.065 --> 00:03:10.965
and I'm using these algorithms to predict the behavior

66
00:03:11.065 --> 00:03:13.725
of your project in the future, I'm going to,

67
00:03:13.905 --> 00:03:15.125
the algorithm's gonna tell me

68
00:03:15.515 --> 00:03:17.885
what your energy consumption is

69
00:03:17.905 --> 00:03:21.325
or how fast people can get out in a fire or whether

70
00:03:21.325 --> 00:03:23.605
or not the infection rate in your operating

71
00:03:23.605 --> 00:03:25.245
theater is gonna be reduced.

72
00:03:26.105 --> 00:03:30.405
Um, those are valuable outputs that we should get paid for.

73
00:03:31.295 --> 00:03:33.315
So my thesis has always been we need

74
00:03:33.315 --> 00:03:35.915
to get into the making things happen business instead

75
00:03:35.915 --> 00:03:37.995
of the producing drawings business.

76
00:03:38.615 --> 00:03:41.995
And that's the, that's the root out of this sort

77
00:03:41.995 --> 00:03:45.355
of productivity race to the bottom, that technology

78
00:03:45.865 --> 00:03:47.955
creates the built environment

79
00:03:48.175 --> 00:03:51.075
and the data about the built environment represents the

80
00:03:51.075 --> 00:03:53.835
current built environment, which has a lot

81
00:03:53.895 --> 00:03:57.595
of ethical challenges to it that we're all familiar with.

82
00:03:58.135 --> 00:04:00.195
One of the things, for example, that I've been working on

83
00:04:00.415 --> 00:04:03.395
is, trying to understand the ethics

84
00:04:03.455 --> 00:04:05.795
of the material supply chain and,

85
00:04:06.295 --> 00:04:09.235
and slavery, modern slavery, as you call it in the uk

86
00:04:09.255 --> 00:04:11.555
or we call it supply chain equity

87
00:04:11.705 --> 00:04:12.995
here in our research group.

88
00:04:13.815 --> 00:04:16.235
And that's the kind of stuff

89
00:04:16.255 --> 00:04:18.355
that's endemic to the system right now.

90
00:04:19.225 --> 00:04:22.125
So right now, if I use an AI to search for materials,

91
00:04:22.125 --> 00:04:24.725
let's say that were in intelligently possible

92
00:04:25.065 --> 00:04:26.845
and I use AI to search for a material.

93
00:04:27.525 --> 00:04:29.645
I get all the normal things about that material,

94
00:04:29.705 --> 00:04:32.645
who sells it, how much it costs, how much it weighs,

95
00:04:33.105 --> 00:04:35.405
how much energy consumed, I might even get

96
00:04:35.405 --> 00:04:38.565
how much carbon's in it, but I don't get anything about its

97
00:04:39.155 --> 00:04:40.205
ethical sourcing.

98
00:04:40.605 --> 00:04:42.765
I may have chosen to put zinc on my building,

99
00:04:43.305 --> 00:04:46.165
but I have no visibility into the fact that

100
00:04:46.475 --> 00:04:50.165
that zinc started out as zinc oxide mine

101
00:04:50.265 --> 00:04:53.685
by a child in a Peruvian zinc oxide mine.

102
00:04:54.425 --> 00:04:58.525
So there are, we, we can't use the systems

103
00:04:58.665 --> 00:05:01.925
to reify existing data infrastructures

104
00:05:02.285 --> 00:05:04.365
'cause they don't give us enough insight into

105
00:05:04.365 --> 00:05:05.685
what's going on in practice.

106
00:05:05.685 --> 00:05:07.365
It's a lot more challenging. Right.

107
00:05:07.905 --> 00:05:09.885
And a lot, a lot of, and,

108
00:05:09.905 --> 00:05:12.405
and that raises a whole other set of info of of

109
00:05:13.125 --> 00:05:16.945
questions about where architecture, engineering,

110
00:05:17.225 --> 00:05:19.625
construction and asset operation data even comes from.

111
00:05:20.865 --> 00:05:23.525
My design is my intellectual property.

112
00:05:24.355 --> 00:05:25.655
And when I was practicing

113
00:05:25.675 --> 00:05:28.655
and I practiced for almost 20 years, most

114
00:05:28.655 --> 00:05:30.335
of my clients had confidentiality

115
00:05:30.345 --> 00:05:31.815
provisions in their contracts.

116
00:05:32.625 --> 00:05:35.245
So I couldn't, I can't even use that data

117
00:05:36.155 --> 00:05:37.775
to train my own AI

118
00:05:37.775 --> 00:05:40.695
because the client doesn't want anything about

119
00:05:40.695 --> 00:05:43.535
that information being distributed in any kind of way.

120
00:05:44.155 --> 00:05:48.455
So unlike architecture's not like architectural data,

121
00:05:48.515 --> 00:05:51.865
really good architectural data like building models

122
00:05:51.965 --> 00:05:56.705
and plans and specifications, that stuff's not scrapable

123
00:05:56.705 --> 00:05:58.565
Off the internet the way the

124
00:05:58.565 --> 00:05:59.965
large language models have been trained.

125
00:06:00.145 --> 00:06:01.285
So we need a different strategy.

126
00:06:01.685 --> 00:06:04.605
I wrote generally about an idea in my book,

127
00:06:04.625 --> 00:06:08.125
and I still think this idea has some legs that the building

128
00:06:08.685 --> 00:06:13.045
industry needs to do something around a sort of cooperative

129
00:06:13.595 --> 00:06:15.125
data commons.

130
00:06:16.275 --> 00:06:19.575
And if a third party were empowered to

131
00:06:20.365 --> 00:06:23.895
take the data from various sources like designers

132
00:06:23.915 --> 00:06:27.135
and builders and building operators curate it

133
00:06:27.135 --> 00:06:30.595
so it was usable as training data, anonymize it

134
00:06:31.335 --> 00:06:33.875
so it wasn't traceable back to the original source

135
00:06:34.575 --> 00:06:36.955
and create some kind of transactional business model,

136
00:06:37.345 --> 00:06:40.075
then we could, as an industry, excuse me,

137
00:06:40.095 --> 00:06:42.635
we could collectively put enough stuff together

138
00:06:42.635 --> 00:06:44.195
that we could start training these models.

139
00:06:45.315 --> 00:06:48.025
Right now, I think you, you, you need something

140
00:06:48.335 --> 00:06:50.705
that allows you to collect enough data to do this.

141
00:06:51.855 --> 00:06:56.275
Unless a new theory of machine learning emerges

142
00:06:56.275 --> 00:06:57.995
that requires a lot less data than

143
00:06:58.145 --> 00:06:59.515
what is necessary right now.

144
00:07:00.365 --> 00:07:04.935
Because you know, the, if, if there are trillions

145
00:07:05.035 --> 00:07:09.455
of tokens that are training algorithms like Che GPT,

146
00:07:10.225 --> 00:07:12.635
there's not, I mean there's, I,

147
00:07:12.635 --> 00:07:15.235
there's probably 20 architecture firms within a one mile

148
00:07:15.255 --> 00:07:16.595
radius of where I'm sitting right now.

149
00:07:17.115 --> 00:07:18.875
I can take every bit of data in all their offices.

150
00:07:19.415 --> 00:07:23.615
And it's not enough data to reasonably train an algorithm

151
00:07:24.845 --> 00:07:26.175
because it's just not enough.

152
00:07:26.465 --> 00:07:27.735
There are many, many challenges,

153
00:07:27.735 --> 00:07:29.495
which is why I think this idea

154
00:07:29.515 --> 00:07:32.295
of picking much more limited discrete tasks

155
00:07:33.655 --> 00:07:36.515
and training on those discrete tasks is the right,

156
00:07:36.575 --> 00:07:37.675
is really the right way to go.

157
00:07:37.695 --> 00:07:40.555
Sustainability happens to be one realm where we have a lot

158
00:07:40.555 --> 00:07:43.115
of data, we have a lot of energy data, we have a lot

159
00:07:43.115 --> 00:07:44.275
of building performance data.

160
00:07:44.535 --> 00:07:46.315
We have a lot of material performance data.

161
00:07:46.775 --> 00:07:49.435
And I think a co a combination of those things

162
00:07:50.205 --> 00:07:52.675
could provide some terrific platforms

163
00:07:52.695 --> 00:07:54.515
for providing sustainability insight.

164
00:07:55.135 --> 00:07:57.315
But if I don't think it's gonna be designed me

165
00:07:57.315 --> 00:07:59.835
as sustainable building, I think it's gonna be

166
00:08:00.675 --> 00:08:03.435
evaluate the embodied carbon of my design proposal,

167
00:08:03.565 --> 00:08:05.235
which turns out to be a really hard problem

168
00:08:06.035 --> 00:08:10.445
because the carbon footprint of a building is comprised of

169
00:08:11.025 --> 00:08:13.965
not just the carbon it produces when it's operating,

170
00:08:14.265 --> 00:08:17.245
but all the embodied carbon streams of all the materials

171
00:08:17.435 --> 00:08:19.925
that went into the building and all the embodied carbon

172
00:08:19.955 --> 00:08:20.965
that, streams

173
00:08:20.965 --> 00:08:22.965
of all the labor that went into the building.

174
00:08:23.425 --> 00:08:26.405
It turns out a front end loader that digs the foundation

175
00:08:27.015 --> 00:08:29.445
dumps a huge amount of carbon into the atmosphere.

176
00:08:30.185 --> 00:08:33.085
So how are you gonna, how are you gonna coalesce all

177
00:08:33.085 --> 00:08:34.245
that into a coherent view?

178
00:08:34.665 --> 00:08:36.085
AI and AI is perfect for that

179
00:08:36.425 --> 00:08:37.885
and humans are not so good at that.

180
00:08:39.105 --> 00:08:42.845
So great. It's a, that's a really great place to go.

181
00:08:43.005 --> 00:08:45.645
I think the business models in the building industry are not

182
00:08:45.645 --> 00:08:47.285
very complex.

183
00:08:47.995 --> 00:08:50.565
It's basically, you want me to design something

184
00:08:50.565 --> 00:08:53.605
and I promise you I can design it for X number of pounds.

185
00:08:55.245 --> 00:08:56.465
Uh, you want me to build something

186
00:08:56.465 --> 00:08:59.625
and I promise you I can build it for X number of pounds.

187
00:09:00.535 --> 00:09:02.355
And then we argue about the fringes

188
00:09:02.495 --> 00:09:04.235
and we hope we're still in business when we're done.

189
00:09:04.575 --> 00:09:06.835
And that's, that's the model essentially.

190
00:09:07.785 --> 00:09:12.545
Um, the, there isn't a giant pool of data

191
00:09:13.175 --> 00:09:16.705
that is available to do the kind of training

192
00:09:17.055 --> 00:09:20.465
that has created the first, the first generation of these

193
00:09:21.145 --> 00:09:24.395
language and image models because there are lots of pictures

194
00:09:24.395 --> 00:09:25.715
of buildings on the internet,

195
00:09:26.215 --> 00:09:27.995
but a building is a much more complicated

196
00:09:27.995 --> 00:09:29.075
thing and a than a cat.

197
00:09:29.655 --> 00:09:32.885
So the IP questions ultimately

198
00:09:34.065 --> 00:09:36.085
as they relate to business model are gonna,

199
00:09:36.205 --> 00:09:37.805
I think you're gonna have to be very transactional.

200
00:09:38.245 --> 00:09:39.485
I have a lot of really great data.

201
00:09:39.785 --> 00:09:40.965
You can buy it from me,

202
00:09:41.465 --> 00:09:42.685
but you're not just gonna be able to,

203
00:09:42.685 --> 00:09:43.965
you're not gonna just buy it from me.

204
00:09:44.105 --> 00:09:47.085
You're gonna have to immunize me from its illicit use.

205
00:09:47.825 --> 00:09:51.125
And because I am an architect with a license

206
00:09:51.235 --> 00:09:54.205
that says I'm responsible for the public's health, safety,

207
00:09:54.225 --> 00:09:57.445
and welfare, this isn't a small poor problem.

208
00:09:57.905 --> 00:09:59.405
It is a large poor problem.

209
00:10:00.145 --> 00:10:02.125
And you know, in, in the UK

210
00:10:02.135 --> 00:10:05.325
after the GRENVILLE report, there's this responsibility

211
00:10:05.325 --> 00:10:07.165
for somebody called the principal designer, right?

212
00:10:07.665 --> 00:10:09.485
And that person has

213
00:10:09.485 --> 00:10:12.845
to take responsibility in a very fundamental way

214
00:10:13.145 --> 00:10:16.845
for the entire process, which is gonna generate a lot of ip.

215
00:10:17.525 --> 00:10:19.265
And that IP is extremely valuable.

216
00:10:19.525 --> 00:10:21.905
And if it's gonna get used for it, it,

217
00:10:21.925 --> 00:10:23.905
if it's gonna get used outside that person's office

218
00:10:23.925 --> 00:10:26.385
for training purposes, there has to be a quick,

219
00:10:26.385 --> 00:10:27.745
there has to be a quick pro quo.

220
00:10:27.845 --> 00:10:29.905
In my opinion, I'm a skeptical

221
00:10:30.895 --> 00:10:35.265
thinker about technology implementation because I've seen

222
00:10:35.685 --> 00:10:36.865
and was right in the middle

223
00:10:36.925 --> 00:10:39.105
of one major technology transition.

224
00:10:40.105 --> 00:10:43.345
Actually, two, I was a customer in the CAD transition

225
00:10:43.685 --> 00:10:46.145
and I was a vendor in the BIM transition.

226
00:10:46.525 --> 00:10:49.025
So I've seen the psychodynamics of how

227
00:10:49.645 --> 00:10:51.225
and the financial dynamics

228
00:10:51.225 --> 00:10:52.745
and the technical dynamics of

229
00:10:52.745 --> 00:10:54.865
how the building industry adopts new technology.

230
00:10:55.605 --> 00:10:56.625
And so what's gonna happen,

231
00:10:57.545 --> 00:11:00.305
I predict what's gonna happen is we're gonna have three

232
00:11:00.925 --> 00:11:05.725
or four years of anecdotal, episodic,

233
00:11:05.815 --> 00:11:09.405
small bore task related AI implementation

234
00:11:10.245 --> 00:11:14.185
until the technology jumps past the,

235
00:11:15.115 --> 00:11:18.385
giant model owned by five software companies,

236
00:11:19.295 --> 00:11:21.745
uses all the extra energy we can create on the planet

237
00:11:21.925 --> 00:11:23.185
to run it model.

238
00:11:23.805 --> 00:11:25.665
And if the technology can get to the point

239
00:11:25.665 --> 00:11:27.625
where it is accessible

240
00:11:27.765 --> 00:11:31.945
and can scale to participants in the building industry,

241
00:11:32.255 --> 00:11:33.945
then we'll start to see some real progress.

242
00:11:34.965 --> 00:11:38.745
And so, I, I'd like to leave behind the whole argument

243
00:11:38.745 --> 00:11:42.745
that architects are gonna lose their jobs because of ai.

244
00:11:42.825 --> 00:11:45.365
'cause I, I just don't buy it. I don't buy it.

245
00:11:45.925 --> 00:11:48.405
I think the, I think the technology's very use useful.

246
00:11:48.985 --> 00:11:50.445
We are using it in this building.

247
00:11:50.935 --> 00:11:54.605
We're using it on this campus. We're teaching courses on it.

248
00:11:54.785 --> 00:11:56.365
Our students are using it every day,

249
00:11:57.025 --> 00:12:01.205
but they're not doing profound things.

250
00:12:01.345 --> 00:12:03.665
But the, and it's really,

251
00:12:03.765 --> 00:12:05.825
my skepticism is about prof is really about

252
00:12:05.825 --> 00:12:06.905
profundity at this point.

253
00:12:07.825 --> 00:12:10.845
And I, I remain concerned that while we are

254
00:12:12.105 --> 00:12:16.135
so emotionally involved in making images, making pictures

255
00:12:16.135 --> 00:12:19.255
of buildings, we could be doing something more important.

256
00:12:19.895 --> 00:12:21.835
That's really my, my my point.

257
00:12:22.225 --> 00:12:24.595
It's a great question because, you know, an AI book

258
00:12:24.595 --> 00:12:27.835
that was written in 2021 and published in 2022,

259
00:12:27.895 --> 00:12:29.195
and now it's 2024.

260
00:12:30.335 --> 00:12:33.435
Um, it, it, it needs a refresh,

261
00:12:33.435 --> 00:12:34.915
which I'm working on right now.

262
00:12:35.055 --> 00:12:38.395
But the basic thesis of the book,

263
00:12:38.415 --> 00:12:42.035
or the basic message of the book is architecture

264
00:12:42.135 --> 00:12:43.875
as a profession needs a strategy

265
00:12:44.255 --> 00:12:45.555
to deal with this technology.

266
00:12:46.925 --> 00:12:50.055
That strategy has certain characteristics that,

267
00:12:50.205 --> 00:12:52.535
that one strategy has certain characteristics

268
00:12:52.535 --> 00:12:55.935
that I put forward about deploying the technology in the

269
00:12:55.935 --> 00:12:58.095
service of value and performance.

270
00:12:58.445 --> 00:13:01.495
Both deploying the technology in the service of

271
00:13:02.045 --> 00:13:04.695
extending the architect's relationship to construction

272
00:13:04.755 --> 00:13:06.895
and asset operation,

273
00:13:07.235 --> 00:13:11.215
and being clear with the technology vendors about the things

274
00:13:11.245 --> 00:13:12.895
that we're really interested in doing

275
00:13:13.315 --> 00:13:15.255
and not being carried away by the stream

276
00:13:15.675 --> 00:13:17.135
of everything else that's going on.

277
00:13:17.395 --> 00:13:19.335
That's, that message remains.

278
00:13:19.995 --> 00:13:24.015
And if I re Reba wants to re-release the book,

279
00:13:24.575 --> 00:13:28.055
I guess early next year, if I can finish the revisions

280
00:13:28.355 --> 00:13:29.735
and I'm not changing that part,

281
00:13:30.335 --> 00:13:31.495
I still think that part is right.

282
00:13:32.245 --> 00:13:33.585
My examples are outta date.

283
00:13:34.245 --> 00:13:36.105
Um, and there are lots of details,

284
00:13:36.205 --> 00:13:39.465
but the, the central argument I think still obtains

285
00:13:39.575 --> 00:13:41.025
this argument is projective.

286
00:13:41.055 --> 00:13:44.385
It's like, I recommend that our profession do these things.

287
00:13:46.335 --> 00:13:48.455
I might be wrong. I've been wrong in the past

288
00:13:49.425 --> 00:13:51.695
about these things and I'm gonna be wrong in the future,

289
00:13:52.515 --> 00:13:53.575
but I this,

290
00:13:54.095 --> 00:13:56.575
I still think the recommended approach is correct.

291
00:13:57.455 --> 00:13:59.435
And the question is, where is the innovation

292
00:13:59.655 --> 00:14:03.355
and the, the willingness to take advantage

293
00:14:03.355 --> 00:14:05.725
of the technology gonna exist?

294
00:14:05.825 --> 00:14:08.645
One thing to remember is, in the,

295
00:14:08.645 --> 00:14:12.005
of the major players in the building design continuum,

296
00:14:12.705 --> 00:14:17.045
we are the smallest, we're teeny tiny compared to say

297
00:14:17.815 --> 00:14:19.515
the contractors or the owners

298
00:14:20.175 --> 00:14:22.755
or even the sub, even subcontractors as a whole.

299
00:14:23.415 --> 00:14:25.755
So we live at the headwater of the process,

300
00:14:26.095 --> 00:14:28.995
but we don't, that's where our leverage is.

301
00:14:29.025 --> 00:14:31.555
It's our positioning in the process, not by virtue

302
00:14:31.575 --> 00:14:33.835
of our scale or our economic

303
00:14:33.855 --> 00:14:36.595
or political influence, which is another argument

304
00:14:36.595 --> 00:14:40.235
for being super strategic, but it's hard to do.

Bridging Architecture and Technology: Embracing AI's Potential

00:00:00 - When I was an undergraduate in the seventies, I had taken several courses about what we now know as natural language processing. I was studying architecture, but also computer science. And so I realized this was an opportunity for me to sort of connect the dots. I'm also very interested in the role technology plays in how architects operate in systems of design, construction and asset operation. And so my previous, both my previous academic work and my commercial work, when I was a, an employee

00:00:37 - To work on that question, I was asked to write it in 2020 and worked on it in 2021, thanks to the pandemic. the, the, the efficacy that technology wasn't very well understood by architects. And of course there was an immediate concern that it faced that we faced an existential threat to be replaced by algorithms. And I argued in my book that a, that's not true

00:01:07 - And b, worrying about that is to miss the big opportunity of the use of the technology, which is not just around productivity, but around increasing the value of what architects actually do. By having the machines do things that they're good at, like organize and understand data and evaluate it, and have architects use that as an input to what we are good at, which is taking large complex contexts and synthesizing them into answers using a design process

Navigating AI Distraction in Architecture

00:01:42 - and projecting into the future. I've been in the architecture world for four decades now. There's a tendency for architects to take every new technology and try to make shapes for images with it. It doesn't matter what it is. The first instinct is to make, is to make new interesting shapes and make images. And there's nothing wrong with that. It's just the problem is it, with ai, it's kind of a distraction. It's a, it's a, it's a place where an opportunity might get missed.

00:02:13 - that some swath of the kinds of day-to-day work that architects do, transactional correspondence, checking documents for graphic conformance, three dimensional coordination of things in space, which is a lot of work. Let's make sure that this doc doesn't run into this beam or this, there's enough space for this guy to get a paintbrush in here. A lot of that stuff will get automated. So this is why part of my work is around business models.

00:02:47 - as architects, a lot more productive architecture is largely transacted on fixed fees, which are highly commoditized and highly competitive. And so when everybody starts to Get more, more productive, the fees are gonna go down unless there's a different value proposition. But if I say to you, I'm an AI enabled architect and I'm using these algorithms to predict the behavior of your project in the future, I'm going to,

Shifting Focus: From Drawings to Ethical Impact in Architecture

00:03:15 - what your energy consumption is or how fast people can get out in a fire or whether or not the infection rate in your operating theater is gonna be reduced. those are valuable outputs that we should get paid for. So my thesis has always been we need to get into the making things happen business instead of the producing drawings business. And that's the, that's the root out of this sort of productivity race to the bottom, that technology

00:03:48 - and the data about the built environment represents the current built environment, which has a lot of ethical challenges to it that we're all familiar with. One of the things, for example, that I've been working on is, trying to understand the ethics of the material supply chain and, and slavery, modern slavery, as you call it in the uk or we call it supply chain equity here in our research group. And that's the kind of stuff

00:04:19 - So right now, if I use an AI to search for materials, let's say that were in intelligently possible and I use AI to search for a material. I get all the normal things about that material, who sells it, how much it costs, how much it weighs, how much energy consumed, I might even get how much carbon's in it, but I don't get anything about its ethical sourcing. I may have chosen to put zinc on my building, but I have no visibility into the fact that

Challenges of Data Utilization in Architecture

00:04:50 - by a child in a Peruvian zinc oxide mine. So there are, we, we can't use the systems to reify existing data infrastructures 'cause they don't give us enough insight into what's going on in practice. It's a lot more challenging. Right. And a lot, a lot of, and, and that raises a whole other set of info of of questions about where architecture, engineering,

00:05:20 - My design is my intellectual property. And when I was practicing and I practiced for almost 20 years, most of my clients had confidentiality provisions in their contracts. So I couldn't, I can't even use that data to train my own AI because the client doesn't want anything about that information being distributed in any kind of way. So unlike architecture's not like architectural data,

00:05:51 - and plans and specifications, that stuff's not scrapable Off the internet the way the large language models have been trained. So we need a different strategy. I wrote generally about an idea in my book, and I still think this idea has some legs that the building industry needs to do something around a sort of cooperative data commons. And if a third party were empowered to

Leveraging Data for Sustainable Architecture

00:06:23 - and builders and building operators curate it so it was usable as training data, anonymize it so it wasn't traceable back to the original source and create some kind of transactional business model, then we could, as an industry, excuse me, we could collectively put enough stuff together that we could start training these models. Right now, I think you, you, you need something that allows you to collect enough data to do this.

00:06:56 - that requires a lot less data than what is necessary right now. Because you know, the, if, if there are trillions of tokens that are training algorithms like Che GPT, there's not, I mean there's, I, there's probably 20 architecture firms within a one mile radius of where I'm sitting right now. I can take every bit of data in all their offices. And it's not enough data to reasonably train an algorithm

00:07:26 - There are many, many challenges, which is why I think this idea of picking much more limited discrete tasks and training on those discrete tasks is the right, is really the right way to go. Sustainability happens to be one realm where we have a lot of data, we have a lot of energy data, we have a lot of building performance data. We have a lot of material performance data. And I think a co a combination of those things could provide some terrific platforms for providing sustainability insight.

Evaluating Embodied Carbon in Sustainable Building Design

00:07:57 - as sustainable building, I think it's gonna be evaluate the embodied carbon of my design proposal, which turns out to be a really hard problem because the carbon footprint of a building is comprised of not just the carbon it produces when it's operating, but all the embodied carbon streams of all the materials that went into the building and all the embodied carbon that, streams of all the labor that went into the building. It turns out a front end loader that digs the foundation

00:08:30 - So how are you gonna, how are you gonna coalesce all that into a coherent view? AI and AI is perfect for that and humans are not so good at that. So great. It's a, that's a really great place to go. I think the business models in the building industry are not very complex. It's basically, you want me to design something and I promise you I can design it for X number of pounds. you want me to build something and I promise you I can build it for X number of pounds.

00:09:02 - and we hope we're still in business when we're done. And that's, that's the model essentially. the, there isn't a giant pool of data that is available to do the kind of training that has created the first, the first generation of these language and image models because there are lots of pictures of buildings on the internet, but a building is a much more complicated thing and a than a cat. So the IP questions ultimately

Navigating Data Responsibility in Architecture

00:09:36 - I think you're gonna have to be very transactional. I have a lot of really great data. You can buy it from me, but you're not just gonna be able to, you're not gonna just buy it from me. You're gonna have to immunize me from its illicit use. And because I am an architect with a license that says I'm responsible for the public's health, safety, and welfare, this isn't a small poor problem. It is a large poor problem. And you know, in, in the UK after the GRENVILLE report, there's this responsibility

00:10:07 - And that person has to take responsibility in a very fundamental way for the entire process, which is gonna generate a lot of ip. And that IP is extremely valuable. And if it's gonna get used for it, it, if it's gonna get used outside that person's office for training purposes, there has to be a quick, there has to be a quick pro quo. In my opinion, I'm a skeptical thinker about technology implementation because I've seen

00:10:36 - of one major technology transition. Actually, two, I was a customer in the CAD transition and I was a vendor in the BIM transition. So I've seen the psychodynamics of how and the financial dynamics and the technical dynamics of how the building industry adopts new technology. And so what's gonna happen, I predict what's gonna happen is we're gonna have three or four years of anecdotal, episodic,

Bridging Technology and Architecture: Progress and Skepticism

00:11:10 - until the technology jumps past the, giant model owned by five software companies, uses all the extra energy we can create on the planet to run it model. And if the technology can get to the point where it is accessible and can scale to participants in the building industry, then we'll start to see some real progress. And so, I, I'd like to leave behind the whole argument

00:11:42 - 'cause I, I just don't buy it. I don't buy it. I think the, I think the technology's very use useful. We are using it in this building. We're using it on this campus. We're teaching courses on it. Our students are using it every day, but they're not doing profound things. But the, and it's really, my skepticism is about prof is really about profundity at this point. And I, I remain concerned that while we are

00:12:16 - of buildings, we could be doing something more important. That's really my, my my point. It's a great question because, you know, an AI book that was written in 2021 and published in 2022, and now it's 2024. it, it, it needs a refresh, which I'm working on right now. But the basic thesis of the book, or the basic message of the book is architecture

Strategic Deployment of Technology in Architecture

00:12:44 - to deal with this technology. That strategy has certain characteristics that, that one strategy has certain characteristics that I put forward about deploying the technology in the service of value and performance. Both deploying the technology in the service of extending the architect's relationship to construction and asset operation, and being clear with the technology vendors about the things that we're really interested in doing

00:13:15 - of everything else that's going on. That's, that message remains. And if I re Reba wants to re-release the book, I guess early next year, if I can finish the revisions and I'm not changing that part, I still think that part is right. My examples are outta date. and there are lots of details, but the, the central argument I think still obtains this argument is projective. It's like, I recommend that our profession do these things.

00:13:49 - about these things and I'm gonna be wrong in the future, but I this, I still think the recommended approach is correct. And the question is, where is the innovation and the, the willingness to take advantage of the technology gonna exist? One thing to remember is, in the, of the major players in the building design continuum, we are the smallest, we're teeny tiny compared to say

Leveraging Positioning in Architectural Influence

00:14:36 - or even the sub, even subcontractors as a whole. So we live at the headwater of the process, but we don't, that's where our leverage is. It's our positioning in the process, not by virtue of our scale or our economic or political influence, which is another argument for being super strategic, but it's hard to do.

Faculti AI can make mistakes

Research

Policy

Analysis