Cutting the carbon footprint of supercomputing in scientific research

1
00:00:05,120 --> 00:00:07,519
Physics World. Hello, and welcome to the Physics

2
00:00:07,519 --> 00:00:10,480
World Stories podcast. I'm Andrew Glester. And in

3
00:00:10,480 --> 00:00:13,599
the Physics World magazine and on physicsworld.com,

4
00:00:13,599 --> 00:00:16,524
we've recently been covering the carbon footprint

5
00:00:16,824 --> 00:00:19,964
of physics as a discipline. A feature article

6
00:00:20,024 --> 00:00:21,225
from March

7
00:00:21,225 --> 00:00:24,664
by Michael Allen considers the huge footprint of

8
00:00:24,664 --> 00:00:25,964
large scale computing.

9
00:00:26,505 --> 00:00:28,505
In this episode of the podcast, we'll talk

10
00:00:28,505 --> 00:00:29,404
to 2 scientists

11
00:00:29,809 --> 00:00:31,969
who, alongside the work they're doing, are trying

12
00:00:31,969 --> 00:00:33,489
their best to make sure that the science

13
00:00:33,489 --> 00:00:36,469
we do is also looking after the planet

14
00:00:36,609 --> 00:00:37,750
as best as possible.

15
00:00:38,129 --> 00:00:40,549
Unusually for the podcast, we'll hear from somebody

16
00:00:40,609 --> 00:00:43,329
from another field of science. But first up,

17
00:00:43,329 --> 00:00:44,309
here's an astrophysicist.

18
00:00:44,689 --> 00:00:46,844
My name is Simon Poteghi Swart. I am

19
00:00:46,844 --> 00:00:49,984
a professor on computational astrophysics at Leiden Observatory

20
00:00:50,125 --> 00:00:53,664
at University of Leiden. I'm doing research on,

21
00:00:54,524 --> 00:00:55,744
astrophysical phenomena

22
00:00:56,125 --> 00:00:57,505
using computers mostly.

23
00:00:57,859 --> 00:01:00,179
What what are you looking at? Well, I

24
00:01:00,179 --> 00:01:02,340
I I guess that with the computer, the,

25
00:01:02,659 --> 00:01:04,359
the entire universe is my playground.

26
00:01:04,819 --> 00:01:06,200
And I I like

27
00:01:07,540 --> 00:01:08,680
to think freely

28
00:01:09,060 --> 00:01:11,319
as as like a free range thinking

29
00:01:12,134 --> 00:01:12,795
about universe

30
00:01:13,734 --> 00:01:16,155
and computing in, in both aspects.

31
00:01:17,414 --> 00:01:19,334
So I also like other things in computing,

32
00:01:19,334 --> 00:01:20,795
like the hardware, the software,

33
00:01:21,494 --> 00:01:24,394
the mathematics which goes behind it, the algorithms

34
00:01:24,614 --> 00:01:26,854
that, that you need, the writing, the software

35
00:01:26,854 --> 00:01:27,354
itself,

36
00:01:28,829 --> 00:01:31,890
and the astrophysics. So all these things combined

37
00:01:32,109 --> 00:01:34,770
makes my job in a computational astrophysicist.

38
00:01:35,709 --> 00:01:37,710
Okay. But what sort of thing might you

39
00:01:37,710 --> 00:01:39,915
be looking at from an astrophysical point of

40
00:01:39,915 --> 00:01:41,435
view? I I would I would say basically

41
00:01:41,435 --> 00:01:42,415
everything. So,

42
00:01:43,114 --> 00:01:44,895
I like the small scales,

43
00:01:45,355 --> 00:01:48,474
like the formation of planets or the evolution

44
00:01:48,474 --> 00:01:49,215
of asteroids

45
00:01:50,155 --> 00:01:50,655
to,

46
00:01:51,114 --> 00:01:54,075
the evolution of multiple stars or individual stars

47
00:01:54,075 --> 00:01:54,575
even,

48
00:01:55,340 --> 00:01:57,420
how stars interact with each other, how they

49
00:01:57,420 --> 00:01:59,680
dynamically interact with the the role of gravity,

50
00:02:01,019 --> 00:02:03,599
the role of of gas interacting with stars,

51
00:02:03,659 --> 00:02:06,079
so in star clusters, in star cluster formation,

52
00:02:06,619 --> 00:02:08,539
how that works together in the galaxy if

53
00:02:08,539 --> 00:02:09,405
you have the interaction

54
00:02:09,884 --> 00:02:10,784
with dark matter,

55
00:02:11,965 --> 00:02:13,185
or something else,

56
00:02:14,125 --> 00:02:15,965
how the stars interact with each other in

57
00:02:15,965 --> 00:02:18,064
that environment, how you get something like Oumuamua,

58
00:02:18,284 --> 00:02:20,125
the the object that was discovered a few

59
00:02:20,125 --> 00:02:20,865
years ago.

60
00:02:21,324 --> 00:02:24,064
That was a complete surprise to many astronomers

61
00:02:24,125 --> 00:02:27,000
at least. And how galaxies form and interact

62
00:02:27,139 --> 00:02:29,460
together. I guess 1 of the few things

63
00:02:29,460 --> 00:02:31,240
that is not really part of my

64
00:02:31,540 --> 00:02:32,520
academic palette

65
00:02:33,219 --> 00:02:33,699
is,

66
00:02:34,260 --> 00:02:37,300
chemical evolution and radiation and is interaction the

67
00:02:37,300 --> 00:02:39,780
chemical evolution the the interaction between radiation and

68
00:02:39,780 --> 00:02:40,280
chemistry.

69
00:02:41,194 --> 00:02:43,275
I'm not a chemist. I don't know much

70
00:02:43,275 --> 00:02:45,034
about it, and I think it's a fascinating

71
00:02:45,034 --> 00:02:46,635
field, but it is just a little bit

72
00:02:46,635 --> 00:02:48,974
too far from my, academic palette.

73
00:02:49,435 --> 00:02:51,275
Now we'll return to the main topic of

74
00:02:51,275 --> 00:02:52,715
this podcast very soon, but I hope that

75
00:02:52,715 --> 00:02:54,574
you'll indulge me a small diversion

76
00:02:55,034 --> 00:02:57,719
because I'm endlessly fascinated by Oumuamua

77
00:02:58,099 --> 00:02:59,780
having written a feature about it for Physics

78
00:02:59,780 --> 00:03:01,939
World several years ago. And I wondered what

79
00:03:01,939 --> 00:03:03,479
the role is of a supercomputer

80
00:03:04,019 --> 00:03:05,959
in trying to work out the origin

81
00:03:06,340 --> 00:03:07,560
of something like Oumuamua,

82
00:03:07,939 --> 00:03:08,680
this surprising

83
00:03:09,060 --> 00:03:11,985
visitor from another star that entered our solar

84
00:03:11,985 --> 00:03:13,125
system? Well, actually,

85
00:03:13,504 --> 00:03:15,685
a few years ago, when Oumuamua was observed,

86
00:03:15,745 --> 00:03:16,485
I I,

87
00:03:17,025 --> 00:03:19,125
wrote a paper exactly about this topic,

88
00:03:20,625 --> 00:03:22,784
being a surprise to me at least that

89
00:03:22,784 --> 00:03:25,105
this object existed enough. If you think about

90
00:03:25,105 --> 00:03:27,120
it And if you do simulations

91
00:03:27,500 --> 00:03:30,300
of its existence and where you expect these

92
00:03:30,300 --> 00:03:32,479
objects to move around in the galaxy,

93
00:03:32,860 --> 00:03:34,780
it turns out as in many things that

94
00:03:34,780 --> 00:03:36,800
it in the end, it isn't that surprising

95
00:03:37,020 --> 00:03:38,960
that the object should have been discovered

96
00:03:39,340 --> 00:03:41,120
if you look carefully at the sky.

97
00:03:42,435 --> 00:03:44,514
And in in in this case, use of

98
00:03:44,514 --> 00:03:46,534
supercomputers or the use of any computer

99
00:03:46,995 --> 00:03:47,735
would be,

100
00:03:48,034 --> 00:03:49,795
in my case, to to try to make

101
00:03:49,795 --> 00:03:52,295
estimates of where does it come from,

102
00:03:53,395 --> 00:03:55,349
what does it do, how does it appear

103
00:03:55,349 --> 00:03:56,949
to be the way it is, and where

104
00:03:56,949 --> 00:03:57,769
does it go.

105
00:03:58,150 --> 00:03:59,349
And to be to be a bit more

106
00:03:59,349 --> 00:04:00,489
explicit in that

107
00:04:00,870 --> 00:04:01,269
is,

108
00:04:01,669 --> 00:04:03,689
what we figured out is that

109
00:04:04,229 --> 00:04:07,049
a is a type of object that naturally

110
00:04:07,110 --> 00:04:07,610
follows

111
00:04:08,150 --> 00:04:10,090
out of the planet formation process.

112
00:04:10,895 --> 00:04:12,415
And if you make planets, you make a

113
00:04:12,415 --> 00:04:13,235
lot of these,

114
00:04:13,935 --> 00:04:14,995
pieces of junk,

115
00:04:15,694 --> 00:04:16,435
like Oumuamua,

116
00:04:17,134 --> 00:04:20,095
which are spreading around by the star making

117
00:04:20,095 --> 00:04:21,634
the planets into the galactic,

118
00:04:22,175 --> 00:04:24,889
potential. And therefore, any other star that moves

119
00:04:24,889 --> 00:04:26,810
around in the galactic potential may need a

120
00:04:26,810 --> 00:04:28,650
few of these objects. Now if you're anything

121
00:04:28,650 --> 00:04:30,889
like me, that begs a question. And don't

122
00:04:30,889 --> 00:04:32,750
worry. I do ask that question

123
00:04:33,050 --> 00:04:34,350
later in the podcast.

124
00:04:35,129 --> 00:04:35,629
Astrophysicists

125
00:04:36,090 --> 00:04:38,904
use super computers to process the enormous amounts

126
00:04:38,904 --> 00:04:40,604
of data collected by telescopes

127
00:04:41,144 --> 00:04:42,604
and to carry out simulations

128
00:04:43,144 --> 00:04:44,444
to understand cosmological

129
00:04:44,824 --> 00:04:45,324
processes.

130
00:04:45,865 --> 00:04:47,944
But should we be concerned about the amount

131
00:04:47,944 --> 00:04:49,704
of energy that they use? No. I I

132
00:04:49,865 --> 00:04:51,240
yeah. I think I think we should be

133
00:04:51,240 --> 00:04:53,019
concerned. I don't think we should worry.

134
00:04:53,879 --> 00:04:55,079
I mean, we should of course, we should

135
00:04:55,079 --> 00:04:56,220
worry about the climate,

136
00:04:56,919 --> 00:04:58,860
because we are using too many resources,

137
00:04:59,399 --> 00:05:01,720
or natural resources in order to to keep

138
00:05:01,720 --> 00:05:03,579
our, convenient healthy life.

139
00:05:04,134 --> 00:05:07,194
But, computers take an awful lot of energy

140
00:05:07,975 --> 00:05:09,514
to work. And,

141
00:05:10,295 --> 00:05:12,615
the I think I think the underlying problem

142
00:05:12,694 --> 00:05:14,375
so I think this is the fundamental problem.

143
00:05:14,375 --> 00:05:16,154
Right? Supercomputers, in particular,

144
00:05:16,979 --> 00:05:18,519
are extremely power hungry,

145
00:05:19,220 --> 00:05:20,680
and, some people,

146
00:05:21,779 --> 00:05:24,259
use them maybe in in not the ideal

147
00:05:24,259 --> 00:05:26,120
way, which I would say something like,

148
00:05:27,139 --> 00:05:28,120
mining Bitcoins,

149
00:05:29,105 --> 00:05:32,384
which is extremely energy unfriendly. But also high

150
00:05:32,384 --> 00:05:35,685
performance computing in science is extremely environmentally unfriendly

151
00:05:36,384 --> 00:05:38,625
as long as these computers are not powered

152
00:05:38,625 --> 00:05:40,004
by renewable resources.

153
00:05:40,545 --> 00:05:42,324
So 1 solution, of course, is

154
00:05:42,709 --> 00:05:45,370
to power these machines by renew renewable resources.

155
00:05:45,910 --> 00:05:47,350
But the problem there is, of course, there

156
00:05:47,350 --> 00:05:50,149
is a limited supply of renewable resources. And

157
00:05:50,149 --> 00:05:50,649
therefore,

158
00:05:50,949 --> 00:05:53,750
if supercomputers all use renewable resources, then the

159
00:05:53,750 --> 00:05:56,069
people at their homes are doomed to use

160
00:05:56,069 --> 00:05:57,830
oil and gas in order to heat their,

161
00:05:58,069 --> 00:05:59,645
environment. I mean, it's it's it's always a

162
00:05:59,645 --> 00:06:01,665
sort of a a 2 sides.

163
00:06:02,045 --> 00:06:04,125
You you can be say, I'm I'm using

164
00:06:04,125 --> 00:06:06,285
only green supercomputers, but then, of course, they

165
00:06:06,285 --> 00:06:08,545
take away the the greeniness from other people.

166
00:06:09,165 --> 00:06:10,685
So I think this is this is the

167
00:06:10,685 --> 00:06:11,839
main issue.

168
00:06:13,980 --> 00:06:15,040
Having said that,

169
00:06:17,579 --> 00:06:18,079
supercomputing

170
00:06:18,459 --> 00:06:21,100
in itself is not the worst thing, at

171
00:06:21,100 --> 00:06:23,279
least for astronomers, is not the worst thing

172
00:06:23,339 --> 00:06:25,019
or the most damaging for the climate. I

173
00:06:25,019 --> 00:06:26,399
think the traveling around

174
00:06:27,045 --> 00:06:28,024
is way more,

175
00:06:28,964 --> 00:06:30,964
demanding on the climate, on on at least

176
00:06:30,964 --> 00:06:32,185
the renewable resources.

177
00:06:32,884 --> 00:06:35,444
And, of course, astronomers like, to have their

178
00:06:35,444 --> 00:06:37,845
telescopes on the very high spots in the

179
00:06:37,845 --> 00:06:39,545
mountains, which are very fragile

180
00:06:40,165 --> 00:06:41,384
environmentally systems,

181
00:06:42,689 --> 00:06:43,189
And

182
00:06:43,569 --> 00:06:44,389
and, you know,

183
00:06:45,170 --> 00:06:45,990
using a lot

184
00:06:46,610 --> 00:06:47,830
of debris there,

185
00:06:48,770 --> 00:06:50,610
is not good for the environment. We have

186
00:06:50,610 --> 00:06:53,810
fairly recently done episodes on the impact of

187
00:06:53,810 --> 00:06:56,194
the science that we do on our climate.

188
00:06:56,574 --> 00:06:58,334
But I wanted to look specifically for this

189
00:06:58,334 --> 00:06:59,634
1 at supercomputers.

190
00:07:00,574 --> 00:07:02,415
How many supercomputers are there? Are you sort

191
00:07:02,415 --> 00:07:04,894
of, vying for time on them the way

192
00:07:04,894 --> 00:07:06,915
the way you would the Hubble Space Telescope

193
00:07:06,975 --> 00:07:09,520
or James Webb? Yeah. Basically. So there there

194
00:07:09,520 --> 00:07:11,759
are let's say there are 500 supercomputers. I

195
00:07:11,759 --> 00:07:13,279
mean, there is there is something what we

196
00:07:13,279 --> 00:07:15,600
call the top 500, which is the 500

197
00:07:15,600 --> 00:07:16,100
fastest,

198
00:07:16,800 --> 00:07:17,939
computers on the planet.

199
00:07:18,400 --> 00:07:20,500
And they're all supercomputers. And there are more,

200
00:07:21,024 --> 00:07:22,785
than this, but this is the the the

201
00:07:22,785 --> 00:07:24,704
the most powerful machines. And after that, there

202
00:07:24,704 --> 00:07:26,545
are, of course, the the the less powerful

203
00:07:26,545 --> 00:07:27,845
the machines become,

204
00:07:28,545 --> 00:07:29,764
the more there are.

205
00:07:30,305 --> 00:07:32,545
There's only 1 fastest computer, of course, and

206
00:07:32,545 --> 00:07:34,785
there are, you know, a lot of which

207
00:07:34,785 --> 00:07:36,300
are hundred times slower than that.

208
00:07:36,939 --> 00:07:39,419
And these computers, the the the fastest computers,

209
00:07:39,419 --> 00:07:40,879
of course, they take the most energy.

210
00:07:41,339 --> 00:07:43,180
And the amount of energy, if you use

211
00:07:43,180 --> 00:07:45,759
the biggest supercomputer on the planet,

212
00:07:46,300 --> 00:07:47,039
full force

213
00:07:47,659 --> 00:07:50,319
would be comparable to, to launching a spacecraft

214
00:07:50,705 --> 00:07:52,564
in the amount of energy it requires.

215
00:07:53,904 --> 00:07:55,764
Or to put it in other terms,

216
00:07:56,145 --> 00:07:58,305
in the city I'm I'm living in, in

217
00:07:58,305 --> 00:07:58,805
Haarlem,

218
00:07:59,904 --> 00:08:02,064
the the biggest supercomputers on the planet take

219
00:08:02,064 --> 00:08:04,410
as much energy as basically a small city

220
00:08:04,410 --> 00:08:05,470
like like Harlem.

221
00:08:06,329 --> 00:08:08,410
So there's a considerable amount of energy going

222
00:08:08,410 --> 00:08:10,089
into these machines, and, of course, they run

223
00:08:10,089 --> 00:08:11,149
twenty four seven.

224
00:08:11,529 --> 00:08:13,050
Do you have a concept of how much

225
00:08:13,050 --> 00:08:14,669
of that time is taken up with astronomy?

226
00:08:15,209 --> 00:08:17,595
Actually, it is only a small fraction. So

227
00:08:17,595 --> 00:08:20,235
the astronomers, they they think themselves also always

228
00:08:20,235 --> 00:08:21,134
as being very,

229
00:08:21,995 --> 00:08:23,754
looking using a lot of computer power, which

230
00:08:23,754 --> 00:08:25,615
is sort of a pride for some people.

231
00:08:25,995 --> 00:08:27,835
Yeah. That counts in the fast computer on

232
00:08:27,835 --> 00:08:29,514
the planet. But, actually, I mean, I think

233
00:08:29,514 --> 00:08:31,740
that astronomy is maybe 5% to 4% of

234
00:08:31,740 --> 00:08:32,399
the total,

235
00:08:32,940 --> 00:08:33,440
supercomputer,

236
00:08:34,779 --> 00:08:37,340
usage on the planet. And what would be

237
00:08:37,340 --> 00:08:38,720
the other 95%?

238
00:08:39,340 --> 00:08:41,419
Well, a lot goes up in chemistry, and

239
00:08:41,419 --> 00:08:43,600
a lot goes up in in particle physics

240
00:08:43,820 --> 00:08:46,139
and, and and and regular other types of

241
00:08:46,139 --> 00:08:46,605
physics.

242
00:08:47,004 --> 00:08:49,245
Is is there anything we can do? Like,

243
00:08:49,245 --> 00:08:50,784
you have to use supercomputers

244
00:08:51,164 --> 00:08:53,105
to do the kind of science that you're

245
00:08:53,324 --> 00:08:55,584
wanting to do. Yes and no.

246
00:08:56,044 --> 00:08:57,964
I think there is a tendency of using

247
00:08:57,964 --> 00:08:59,824
bigger machines because they are available.

248
00:09:00,365 --> 00:09:02,410
And, of course, they are faster, but you

249
00:09:02,410 --> 00:09:04,169
have to invest a lot of time to

250
00:09:04,169 --> 00:09:06,669
program them and program them efficiently.

251
00:09:07,290 --> 00:09:09,210
And that's the other part. So there there

252
00:09:09,210 --> 00:09:10,730
are 2 parts to this story. 1 is

253
00:09:10,730 --> 00:09:12,750
there is a tendency to use bigger machines,

254
00:09:13,370 --> 00:09:14,985
and the other thing is there is a

255
00:09:15,225 --> 00:09:16,524
tendency to,

256
00:09:17,464 --> 00:09:19,304
to to be a little bit sloppy in

257
00:09:19,304 --> 00:09:21,164
optimization if you have a faster computer.

258
00:09:22,105 --> 00:09:25,144
And these 2 things, I think, are not

259
00:09:25,144 --> 00:09:27,485
really necessary in in many cases.

260
00:09:28,129 --> 00:09:29,830
You can do very nice,

261
00:09:30,289 --> 00:09:32,870
scientific calculation on smaller computers,

262
00:09:33,410 --> 00:09:34,769
in which case you have to program a

263
00:09:34,769 --> 00:09:36,289
little bit more efficiently. So you have to

264
00:09:36,289 --> 00:09:38,370
be more efficient in or spend more time

265
00:09:38,370 --> 00:09:39,990
in how you optimize your codes.

266
00:09:41,274 --> 00:09:43,695
And, of course, there are problems which just

267
00:09:43,835 --> 00:09:44,335
are

268
00:09:44,715 --> 00:09:46,654
get better if you have the bigger machines.

269
00:09:47,514 --> 00:09:49,115
So I think you should use the biggest

270
00:09:49,115 --> 00:09:51,274
machines really for the biggest problems for where

271
00:09:51,274 --> 00:09:53,595
you really need those machines, but those codes

272
00:09:53,595 --> 00:09:55,774
should be, you know, as optimized as possible

273
00:09:56,190 --> 00:09:58,929
in order to be as environmentally friendly as

274
00:09:59,309 --> 00:10:02,269
possible. But I like, for example, to, to

275
00:10:02,269 --> 00:10:03,629
to use a lot of,

276
00:10:04,429 --> 00:10:06,910
computer time on, you know, cheap laptops or

277
00:10:07,070 --> 00:10:10,210
well, not cheap laptops, but but, laptop computers,

278
00:10:10,750 --> 00:10:11,490
which are

279
00:10:12,055 --> 00:10:14,295
about the most environmentally friendly you can have

280
00:10:14,295 --> 00:10:16,215
because these machines are built to run on

281
00:10:16,215 --> 00:10:18,695
a battery rather than on, being plugged into

282
00:10:18,695 --> 00:10:19,835
the the the wall.

283
00:10:20,455 --> 00:10:23,735
But modern laptops are are so amazingly fast

284
00:10:23,735 --> 00:10:24,235
already,

285
00:10:25,095 --> 00:10:26,075
that many calculations

286
00:10:27,049 --> 00:10:29,610
done actually on bigger machines could be done

287
00:10:29,610 --> 00:10:32,029
on, more efficiently on on on laptops.

288
00:10:32,809 --> 00:10:34,169
On the other side, I mean, at the

289
00:10:34,169 --> 00:10:37,529
same time, using a supercomputer, of course, everything

290
00:10:37,529 --> 00:10:39,049
goes fast. You don't have to run it,

291
00:10:39,690 --> 00:10:41,825
for a week or so. But that's surely

292
00:10:41,965 --> 00:10:44,205
not gonna make up for the for the

293
00:10:44,205 --> 00:10:46,205
amount of power you're using compared to a

294
00:10:46,205 --> 00:10:48,445
laptop. That's right. That's right. 1 of the

295
00:10:48,445 --> 00:10:50,045
ways we might be able to reduce the

296
00:10:50,045 --> 00:10:52,925
time on computers is more efficient coding. Of

297
00:10:52,925 --> 00:10:54,845
course, many people have been talking about it

298
00:10:54,845 --> 00:10:56,950
for very long times, and there there are

299
00:10:56,950 --> 00:10:59,690
many aspects on efficient coding. First of all,

300
00:11:00,149 --> 00:11:02,330
the choice of computer language is important.

301
00:11:03,509 --> 00:11:05,750
The the choice of algorithm is important, and

302
00:11:05,750 --> 00:11:08,230
the detail of of, what you need in

303
00:11:08,230 --> 00:11:09,049
your calculation,

304
00:11:09,545 --> 00:11:11,465
of course, is important too. Right? We cannot

305
00:11:11,465 --> 00:11:12,365
simulate nature

306
00:11:12,745 --> 00:11:14,425
the way nature is. We don't want that

307
00:11:14,425 --> 00:11:16,504
because then we wouldn't be able to interpret

308
00:11:16,504 --> 00:11:17,465
nature as,

309
00:11:18,105 --> 00:11:19,884
as we do now. Then we have to

310
00:11:20,184 --> 00:11:22,605
the the the simulations would be as complicated

311
00:11:22,665 --> 00:11:24,264
as the observations, and we have the same

312
00:11:24,264 --> 00:11:24,764
problem.

313
00:11:25,169 --> 00:11:27,089
So the whole strength of simulation is to

314
00:11:27,089 --> 00:11:29,269
leave everything out what is not necessary

315
00:11:29,570 --> 00:11:30,470
in your opinion,

316
00:11:31,409 --> 00:11:34,049
and that makes the calculations doable, faster, and

317
00:11:34,049 --> 00:11:34,549
interpretable.

318
00:11:35,250 --> 00:11:36,929
So the question is, can you leave out

319
00:11:36,929 --> 00:11:39,524
more and still get the answer you got?

320
00:11:39,764 --> 00:11:43,205
And the tendency in science is to include

321
00:11:43,205 --> 00:11:45,445
as much as possible because, you know, everything

322
00:11:45,445 --> 00:11:46,504
seems to be important.

323
00:11:46,884 --> 00:11:48,404
But sometimes, you know, it depends on the

324
00:11:48,404 --> 00:11:48,904
question.

325
00:11:49,524 --> 00:11:52,184
Now in terms of of more efficient coding,

326
00:11:53,220 --> 00:11:55,139
so you have the algorithms. You have what

327
00:11:55,139 --> 00:11:57,379
can you leave out? But you also have

328
00:11:57,379 --> 00:12:00,179
the optimization of the hardware like, using GPUs

329
00:12:00,179 --> 00:12:01,159
instead of CPU.

330
00:12:02,019 --> 00:12:03,779
There are and, you know, you you wouldn't

331
00:12:03,779 --> 00:12:05,620
believe it almost, but there are still people

332
00:12:05,620 --> 00:12:07,639
using CPUs for their daily calculations

333
00:12:08,125 --> 00:12:10,845
where GPUs, the graphical processing units are a

334
00:12:10,845 --> 00:12:12,705
thousand times more efficient

335
00:12:13,485 --> 00:12:15,965
than CPU. It is like almost as I

336
00:12:16,045 --> 00:12:17,965
if I see people using, you know, high

337
00:12:17,965 --> 00:12:20,945
performance computing completely based on on CPU,

338
00:12:21,769 --> 00:12:22,750
that they're taking,

339
00:12:23,529 --> 00:12:25,529
a big truck, you know, a 30 ton

340
00:12:25,529 --> 00:12:27,450
truck to go to the supermarket in order

341
00:12:27,450 --> 00:12:28,429
to to buy

342
00:12:28,809 --> 00:12:29,710
a dozen x.

343
00:12:30,090 --> 00:12:31,769
They think, well, this is this is not

344
00:12:31,769 --> 00:12:32,269
efficient.

345
00:12:33,129 --> 00:12:35,149
So that is definitely part of

346
00:12:35,915 --> 00:12:37,934
of a change in the community, I think.

347
00:12:38,795 --> 00:12:40,735
More efficient programming, parallelization,

348
00:12:42,315 --> 00:12:43,855
optimum use of the hardware,

349
00:12:44,235 --> 00:12:45,215
the better algorithms,

350
00:12:47,035 --> 00:12:49,434
and and and, and and the proper language,

351
00:12:49,675 --> 00:12:52,259
which is associated with that. And is is

352
00:12:52,259 --> 00:12:53,080
there particular

353
00:12:53,700 --> 00:12:55,620
languages that we people should be using? Is

354
00:12:55,620 --> 00:12:57,379
it about that? Well, that is this is

355
00:12:57,379 --> 00:13:00,019
an extremely difficult discussion because if you really

356
00:13:00,019 --> 00:13:01,779
want to make it as optimum as possible,

357
00:13:01,779 --> 00:13:03,799
you should program in assembler probably.

358
00:13:04,954 --> 00:13:06,794
And with all with all respect, but I

359
00:13:06,794 --> 00:13:08,794
I I do not recommend most people to

360
00:13:08,794 --> 00:13:11,534
use assembler, and I think everybody knows why.

361
00:13:11,834 --> 00:13:13,195
I don't know why. Can you tell me

362
00:13:13,195 --> 00:13:15,034
why? Okay. So there are several reasons why.

363
00:13:15,034 --> 00:13:16,970
First of all, it's it's machine dependent. To

364
00:13:16,970 --> 00:13:18,250
those of you who are listening who are

365
00:13:18,250 --> 00:13:20,110
not so familiar with computer programming,

366
00:13:20,490 --> 00:13:23,049
and as you heard, that included me, an

367
00:13:23,049 --> 00:13:25,370
assembler is a program used to produce the

368
00:13:25,370 --> 00:13:28,990
intermediate machine language that takes your computer code

369
00:13:29,129 --> 00:13:32,575
and assembles it into binary machine instructions that

370
00:13:32,575 --> 00:13:34,754
your computer processor understands.

371
00:13:35,375 --> 00:13:37,215
That was a nice description that I heard

372
00:13:37,215 --> 00:13:39,855
on Jacob Sorber's YouTube channel. It is a

373
00:13:39,855 --> 00:13:42,495
tremendous pain to to write an assembler and

374
00:13:42,495 --> 00:13:44,174
to change language every time you have a

375
00:13:44,174 --> 00:13:45,554
new computer or a new chipset.

376
00:13:47,110 --> 00:13:49,829
And the the languages are not built to,

377
00:13:50,149 --> 00:13:52,309
communicate with a human. They're built to communicate

378
00:13:52,309 --> 00:13:54,549
with a computer. So you really have a

379
00:13:54,549 --> 00:13:57,110
a long baseline of learning in order to

380
00:13:57,110 --> 00:13:58,870
to learn these languages. There are languages which

381
00:13:58,870 --> 00:14:00,250
are much easier to program

382
00:14:00,605 --> 00:14:02,605
a computer with, like Fortran or c or

383
00:14:02,605 --> 00:14:04,524
c plus plus. And, you know, there are,

384
00:14:04,524 --> 00:14:06,545
I don't know, 200, 2 hundred 50 languages,

385
00:14:06,684 --> 00:14:08,144
to program a computer with.

386
00:14:08,684 --> 00:14:10,925
So the best way probably for a human

387
00:14:11,004 --> 00:14:13,644
and it's it's a balance between investing time

388
00:14:13,644 --> 00:14:16,460
and optimization and the possibility to optimize certain

389
00:14:16,460 --> 00:14:17,519
parts of your code.

390
00:14:18,379 --> 00:14:20,240
So the best problem is having

391
00:14:20,860 --> 00:14:24,000
a relatively quick language where you can rapidly

392
00:14:24,059 --> 00:14:25,360
prototype your problem

393
00:14:25,820 --> 00:14:27,200
like you can do in Python,

394
00:14:28,139 --> 00:14:30,559
then optimize the parts which are,

395
00:14:31,825 --> 00:14:33,365
take a lot of time on the computer

396
00:14:33,904 --> 00:14:36,625
and paralyze them using other c or maybe

397
00:14:36,625 --> 00:14:37,365
even assembler.

398
00:14:37,904 --> 00:14:39,985
So having a sort of hybridization of of

399
00:14:39,985 --> 00:14:40,485
languages.

400
00:14:41,184 --> 00:14:43,200
The the difficulty here in is that that,

401
00:14:43,200 --> 00:14:45,120
for example, for an astronomer, I mean, I'm

402
00:14:45,120 --> 00:14:46,500
I'm trained as an astrophysicist.

403
00:14:47,920 --> 00:14:50,660
So what astrophysics should I not learn

404
00:14:51,040 --> 00:14:52,420
in order to learn to program?

405
00:14:53,040 --> 00:14:54,559
Right? I mean, I would love to to

406
00:14:54,559 --> 00:14:55,540
teach my students,

407
00:14:55,945 --> 00:14:57,865
assembler and and the most optimum way of

408
00:14:57,865 --> 00:14:59,464
using a computer, but, you know, I I

409
00:14:59,464 --> 00:15:01,384
want them to know orbital mechanics too, and

410
00:15:01,384 --> 00:15:03,544
I want them to know about, stellar evolution

411
00:15:03,544 --> 00:15:04,204
and hydrodynamics.

412
00:15:04,664 --> 00:15:07,065
So I have to somewhere, you have to

413
00:15:07,065 --> 00:15:07,565
to,

414
00:15:08,024 --> 00:15:10,539
to dilute the the knowledge. That's not the

415
00:15:10,539 --> 00:15:13,019
only area where balancing of time comes into

416
00:15:13,019 --> 00:15:14,480
play. The problem is,

417
00:15:15,259 --> 00:15:16,079
human time

418
00:15:16,539 --> 00:15:19,740
versus computer time versus environment. Right? So it

419
00:15:19,740 --> 00:15:20,959
is a sort of a continuous

420
00:15:21,500 --> 00:15:24,115
tension between these these 3 fields. I would

421
00:15:24,115 --> 00:15:26,115
like to spend as least time as possible

422
00:15:26,115 --> 00:15:28,434
programming in order to do my physics, and

423
00:15:28,434 --> 00:15:29,875
I wanted to do as fast as possible.

424
00:15:29,875 --> 00:15:30,934
So in some sense,

425
00:15:31,555 --> 00:15:34,055
several of these aspects go together.

426
00:15:34,595 --> 00:15:36,355
And I think 1 of the problems we

427
00:15:36,355 --> 00:15:37,875
have at the moment is that we don't

428
00:15:37,875 --> 00:15:39,860
really know very well

429
00:15:40,399 --> 00:15:42,720
how efficient our coding is in terms of

430
00:15:42,720 --> 00:15:43,220
energy.

431
00:15:44,160 --> 00:15:45,679
And, of course, we can look up the

432
00:15:45,679 --> 00:15:46,740
specs of a computer,

433
00:15:47,759 --> 00:15:49,539
but it is very hard to determine,

434
00:15:51,345 --> 00:15:53,024
how how bad you're doing if you're just

435
00:15:53,024 --> 00:15:55,365
running a full computer, at full force.

436
00:15:56,464 --> 00:15:57,985
And I think it will be good if

437
00:15:57,985 --> 00:15:59,264
there is a little bit more research on

438
00:15:59,264 --> 00:16:01,605
this. And and I was a bit surprised

439
00:16:01,664 --> 00:16:02,164
by,

440
00:16:02,945 --> 00:16:04,964
1 of the papers I I wrote recently

441
00:16:05,024 --> 00:16:06,164
on on this aspect,

442
00:16:06,699 --> 00:16:09,820
how little information was available on actually these,

443
00:16:10,779 --> 00:16:13,419
actual measurements. We'll return to Simon later in

444
00:16:13,419 --> 00:16:15,980
the podcast. But some scientists, this time from

445
00:16:15,980 --> 00:16:18,379
outside the world of physics, have developed a

446
00:16:18,379 --> 00:16:20,079
tool called green algorithms

447
00:16:20,644 --> 00:16:23,065
that can help us to understand the impact

448
00:16:23,284 --> 00:16:24,184
of the computations

449
00:16:24,725 --> 00:16:26,745
of our work. My name is Loic Landelong.

450
00:16:26,804 --> 00:16:29,845
I'm a research associate in biomedical data science

451
00:16:29,845 --> 00:16:31,144
at the University of Cambridge.

452
00:16:32,004 --> 00:16:34,325
And for the past two years, I've been,

453
00:16:34,644 --> 00:16:35,669
studying the

454
00:16:36,149 --> 00:16:39,589
carbon footprint of computational science in general, with

455
00:16:39,589 --> 00:16:41,610
a slight focus on computational biology.

456
00:16:42,070 --> 00:16:43,610
It all started actually with

457
00:16:44,149 --> 00:16:46,789
the bushfires in Australia, so in at the

458
00:16:46,789 --> 00:16:49,475
January 2020 because part of our lab is

459
00:16:49,475 --> 00:16:50,914
based in Melbourne, so we have a lot

460
00:16:50,914 --> 00:16:52,294
of collaborators there.

461
00:16:52,754 --> 00:16:54,615
So, it was kind of like a striking,

462
00:16:54,995 --> 00:16:57,254
you know, manifestation of climate change.

463
00:16:57,875 --> 00:16:59,715
And at the same time, we came across

464
00:16:59,715 --> 00:17:02,200
an article by MS Truble, which was 1

465
00:17:02,200 --> 00:17:04,279
of the first article about the carbon footprint

466
00:17:04,279 --> 00:17:06,619
of AI and natural language processing.

467
00:17:07,319 --> 00:17:09,160
So you may have come across the headline

468
00:17:09,160 --> 00:17:10,920
saying that, you know, 1 AI is as

469
00:17:10,920 --> 00:17:11,900
bad as 5,

470
00:17:12,519 --> 00:17:14,440
cars during their lifetime, and it got picked

471
00:17:14,440 --> 00:17:16,144
up quite a lot. And so we just

472
00:17:16,144 --> 00:17:18,544
wanted to know what's the carbon footprint of

473
00:17:18,544 --> 00:17:20,065
what we're doing. And that was the main

474
00:17:20,065 --> 00:17:21,904
question. We you know, very small question. We're

475
00:17:21,904 --> 00:17:23,825
like, okay. Well, AI can have a large

476
00:17:23,825 --> 00:17:24,724
carbon footprint,

477
00:17:25,105 --> 00:17:26,944
but we you know, algorithms we use in,

478
00:17:27,424 --> 00:17:27,924
computational

479
00:17:28,224 --> 00:17:28,724
biology,

480
00:17:29,029 --> 00:17:30,389
also, you know, they run for a long

481
00:17:30,389 --> 00:17:31,909
time. They can they they we need a

482
00:17:31,909 --> 00:17:33,750
lot of resources, a lot of memory, so

483
00:17:33,750 --> 00:17:35,909
maybe there's a large carbon footprint there too.

484
00:17:35,909 --> 00:17:37,269
And I thought it would be a two

485
00:17:37,269 --> 00:17:39,029
days project. You know, we just look it

486
00:17:39,029 --> 00:17:40,950
up, find a calculator somewhere. It would be

487
00:17:40,950 --> 00:17:41,450
brilliant.

488
00:17:41,965 --> 00:17:43,725
It turns out there was no way to

489
00:17:43,725 --> 00:17:45,805
find it. So I poked around a little

490
00:17:45,805 --> 00:17:47,244
bit and there was just no tool at

491
00:17:47,244 --> 00:17:49,085
all and no one was looking at it.

492
00:17:49,325 --> 00:17:50,845
So there was a little bit of work

493
00:17:50,845 --> 00:17:54,125
done for deep learning in particular, but very

494
00:17:54,125 --> 00:17:56,065
specific to the tools and the

495
00:17:56,445 --> 00:17:58,720
hardware used by deep learning. And that was

496
00:17:58,720 --> 00:18:01,279
it. So suddenly we're like, okay. Well, there's

497
00:18:01,279 --> 00:18:02,960
clearly a need for that and it seems

498
00:18:02,960 --> 00:18:05,299
a bit, you know, we especially in biology,

499
00:18:06,480 --> 00:18:08,000
the big part of our focus is to

500
00:18:08,000 --> 00:18:09,220
improve human health,

501
00:18:09,535 --> 00:18:11,694
but also what we're doing also contribute to

502
00:18:11,694 --> 00:18:13,934
climate change, which has a massive impact on,

503
00:18:14,654 --> 00:18:16,734
well, on human health. So it sounds like

504
00:18:16,734 --> 00:18:18,275
we should be acknowledging it,

505
00:18:18,734 --> 00:18:19,634
to some extent.

506
00:18:20,255 --> 00:18:22,150
And so that's really what started it, and

507
00:18:22,150 --> 00:18:23,910
there was no tool to do it. So

508
00:18:23,910 --> 00:18:27,509
I but, yeah, in collaboration with, Jason Greeley

509
00:18:27,509 --> 00:18:29,829
from, the Bacon Institute and Michael Linway from

510
00:18:29,829 --> 00:18:31,450
Cambridge, we we tried to

511
00:18:31,829 --> 00:18:34,069
build 1. So that's what created the green

512
00:18:34,069 --> 00:18:36,230
algorithms project. We have no calculator. And I

513
00:18:36,230 --> 00:18:38,464
thought what was more important than just another

514
00:18:38,464 --> 00:18:40,704
theoretical paper that, you know, not many people

515
00:18:40,704 --> 00:18:43,664
would read rather rather was rather an online

516
00:18:43,664 --> 00:18:45,904
tool so that scientists, in my case, who'd

517
00:18:45,904 --> 00:18:47,424
wanted to do exactly what I wanted to

518
00:18:47,424 --> 00:18:47,924
do,

519
00:18:48,384 --> 00:18:50,144
could have a tool without spending, you know,

520
00:18:50,144 --> 00:18:52,980
months trying to understand how each component works

521
00:18:52,980 --> 00:18:54,819
and things like that. So you've been using

522
00:18:54,819 --> 00:18:57,220
it. What have you found? Not a big

523
00:18:57,220 --> 00:18:58,899
surprise, but we found that, yes, when an

524
00:18:58,899 --> 00:19:02,419
algorithm runs for days or, like, or even

525
00:19:02,419 --> 00:19:05,000
just hours by using a lot of processing

526
00:19:05,059 --> 00:19:07,195
calls and a lot of memory, it has

527
00:19:07,195 --> 00:19:08,654
a significant carbon footprint.

528
00:19:09,035 --> 00:19:11,295
And by significance, it can,

529
00:19:11,835 --> 00:19:13,674
be like, you know, depending on the task,

530
00:19:13,674 --> 00:19:16,095
it can be equivalent to, many flights

531
00:19:16,555 --> 00:19:17,055
between,

532
00:19:17,595 --> 00:19:19,295
Europe and The US, for example.

533
00:19:19,595 --> 00:19:21,195
So that's the orders of magnitudes we can

534
00:19:21,195 --> 00:19:23,829
be talking about. Doesn't mean all tasks have

535
00:19:23,829 --> 00:19:25,990
large carbon footprints. Lots of analysis, you know,

536
00:19:25,990 --> 00:19:27,210
they run fairly quickly.

537
00:19:27,509 --> 00:19:29,429
The kind of thing, you know, you you

538
00:19:29,429 --> 00:19:31,829
you would, you take an image, you export

539
00:19:31,829 --> 00:19:33,589
it in PDF, for example, it's a blink

540
00:19:33,589 --> 00:19:35,315
of an eye. That doesn't have a huge

541
00:19:35,315 --> 00:19:37,955
carbon footprint. Not at all. But, yeah, many

542
00:19:37,955 --> 00:19:40,615
tasks do have large carbon footprints and especially

543
00:19:40,994 --> 00:19:43,015
it like it's across all fields of science.

544
00:19:43,234 --> 00:19:45,634
So, it's, the there's a lot of that

545
00:19:45,634 --> 00:19:47,875
in physics, obviously, because lots of simulations, for

546
00:19:47,875 --> 00:19:50,360
example. So in our initial work, we started

547
00:19:50,360 --> 00:19:51,880
by trying to look a bit at everything.

548
00:19:51,880 --> 00:19:54,460
So we looked at physics simulation, weather forecast,

549
00:19:54,840 --> 00:19:57,180
this kind of thing. And, yeah, all these

550
00:19:57,240 --> 00:19:59,740
all these tasks have carbon footprint.

551
00:20:00,200 --> 00:20:01,640
1 thing that does seem to make a

552
00:20:01,640 --> 00:20:04,184
difference is where the science is taking place.

553
00:20:04,424 --> 00:20:07,325
To find the carbon the carbon footprint of

554
00:20:07,625 --> 00:20:09,965
running a task, you know, running an algorithm,

555
00:20:10,505 --> 00:20:12,664
so it's just based on the energy needed

556
00:20:12,664 --> 00:20:15,085
to power the computer during that time.

557
00:20:15,465 --> 00:20:16,845
And to do that,

558
00:20:17,369 --> 00:20:19,210
it depends on 2 things, how much energy

559
00:20:19,210 --> 00:20:19,869
you need

560
00:20:20,250 --> 00:20:23,289
and what's the carbon footprint of producing this

561
00:20:23,289 --> 00:20:25,309
energy. That's a very simple formula.

562
00:20:26,329 --> 00:20:28,589
So obviously because it's such a, like, linear

563
00:20:28,730 --> 00:20:31,130
relationship, if you multiply the carbon footprint of

564
00:20:31,130 --> 00:20:33,734
producing energy by 2, you multiply the total

565
00:20:33,734 --> 00:20:34,394
thing by

566
00:20:34,775 --> 00:20:37,275
2. And what we found is the discrepancies

567
00:20:37,335 --> 00:20:40,535
between count countries is is like huge. For

568
00:20:40,535 --> 00:20:42,615
a simple example is if you run the

569
00:20:42,615 --> 00:20:44,775
exact same tool, exact same analysis on the

570
00:20:44,775 --> 00:20:46,555
exact same hardware in Switzerland

571
00:20:47,230 --> 00:20:48,609
compared to Australia,

572
00:20:49,390 --> 00:20:52,369
the task will emit 74 times more

573
00:20:52,750 --> 00:20:53,730
greenhouse gases.

574
00:20:54,269 --> 00:20:55,569
Wow. Because Australia,

575
00:20:55,869 --> 00:20:58,109
the energy mix is massively based on gas.

576
00:20:58,109 --> 00:21:00,909
Well, like Switzerland is based on hydro and

577
00:21:00,909 --> 00:21:03,044
apart nuclear from France and things like that.

578
00:21:03,044 --> 00:21:05,005
So, yeah, so it's just based on how

579
00:21:05,005 --> 00:21:05,984
energy is produced.

580
00:21:06,605 --> 00:21:08,784
We have to do science. Right? Because

581
00:21:09,244 --> 00:21:11,244
it's the most important thing that we do

582
00:21:11,244 --> 00:21:13,244
as humans. I mean, I'm biased, but we

583
00:21:13,244 --> 00:21:15,085
have to do science. It it seems to

584
00:21:15,085 --> 00:21:17,505
me that we're going to use energy.

585
00:21:17,960 --> 00:21:18,859
So if we

586
00:21:19,160 --> 00:21:21,900
find ways of making the energy more efficiently,

587
00:21:22,200 --> 00:21:22,940
more sustainably,

588
00:21:23,640 --> 00:21:26,700
and use those, use solar, use hydro, use

589
00:21:26,759 --> 00:21:29,640
use wind, then then that's the answer, isn't

590
00:21:29,640 --> 00:21:31,315
it? Or or can we look at what

591
00:21:31,315 --> 00:21:33,255
we're doing as well? I think it's both.

592
00:21:33,875 --> 00:21:34,694
I I think

593
00:21:35,474 --> 00:21:36,835
well, I mean, we've seen with the, like,

594
00:21:36,835 --> 00:21:37,974
the recent, like, IPCC

595
00:21:38,434 --> 00:21:40,595
reports, really. It's just we can't afford to

596
00:21:40,595 --> 00:21:41,335
just say

597
00:21:41,634 --> 00:21:43,394
this part is the biggest chunk of the

598
00:21:43,394 --> 00:21:45,819
emission so we can afford to not look

599
00:21:45,819 --> 00:21:48,220
at everything else. That's the argument we've seen

600
00:21:48,220 --> 00:21:51,019
for, like, not, you know, climate change deniers,

601
00:21:51,259 --> 00:21:52,779
in all over the world saying, oh, my

602
00:21:52,779 --> 00:21:54,700
country and and for example, I've heard that

603
00:21:54,700 --> 00:21:56,460
a lot in France. My country is only

604
00:21:56,460 --> 00:21:58,380
responsible for, like, 1.2%

605
00:21:58,380 --> 00:22:00,220
of emissions. So, you know, there's no point

606
00:22:00,220 --> 00:22:02,595
doing it. The US and China should be

607
00:22:02,595 --> 00:22:04,515
doing all the work. But, actually, that's not

608
00:22:04,515 --> 00:22:06,434
true. What we need is, like, across the

609
00:22:06,434 --> 00:22:07,735
board, every single,

610
00:22:09,154 --> 00:22:11,894
like, aspect needs to be tackled and reduced.

611
00:22:12,195 --> 00:22:13,174
So, yes,

612
00:22:13,795 --> 00:22:16,535
reduce like, having more efficient energy is

613
00:22:16,890 --> 00:22:17,390
undoubtedly

614
00:22:17,849 --> 00:22:19,630
like 1 thing to be done,

615
00:22:20,089 --> 00:22:21,630
and it's it will have

616
00:22:21,930 --> 00:22:24,009
orders of magnitude more impact than what we

617
00:22:24,009 --> 00:22:26,170
can do by addressing the carbon footprint of

618
00:22:26,170 --> 00:22:26,670
science,

619
00:22:27,049 --> 00:22:29,230
because, you know, obviously, it will benefit housing,

620
00:22:29,684 --> 00:22:31,605
you know, all aspects of energy usage. So

621
00:22:31,605 --> 00:22:32,904
that's that's without a doubt.

622
00:22:33,365 --> 00:22:36,245
But also as scientists, we have limited impact

623
00:22:36,245 --> 00:22:36,904
on that,

624
00:22:37,684 --> 00:22:39,765
you know, through, like, what we're doing. You

625
00:22:39,765 --> 00:22:42,005
know, with there can be grassroot movements or

626
00:22:42,005 --> 00:22:43,769
things like that. But, you know, through our

627
00:22:43,769 --> 00:22:46,009
science, there's limited impact we can do about

628
00:22:46,009 --> 00:22:48,110
the energy policy of a country.

629
00:22:49,130 --> 00:22:51,370
So that's why it is definitely what would

630
00:22:51,370 --> 00:22:52,670
be the more efficient way,

631
00:22:53,450 --> 00:22:55,309
but that's not what we can do personally.

632
00:22:55,684 --> 00:22:57,605
So that's why we looked at, like, what

633
00:22:57,605 --> 00:22:59,845
can we do as scientists instead of, you

634
00:22:59,845 --> 00:23:02,244
know, tackling that. But to answer to answer

635
00:23:02,244 --> 00:23:03,924
your point about we need to do science,

636
00:23:03,924 --> 00:23:05,525
I totally agree. I mean, I would be

637
00:23:05,525 --> 00:23:07,125
out of a job, especially my background is

638
00:23:07,125 --> 00:23:08,505
machine learning. So,

639
00:23:09,080 --> 00:23:10,680
you know, if I say we can't do

640
00:23:10,680 --> 00:23:12,680
heavy AI anymore, I I need to find

641
00:23:12,680 --> 00:23:13,820
something else to do.

642
00:23:14,200 --> 00:23:14,700
But

643
00:23:15,480 --> 00:23:17,080
the the whole point of this project is

644
00:23:17,080 --> 00:23:18,680
definitely not to do to say we shouldn't

645
00:23:18,680 --> 00:23:20,279
do science and not even to say we

646
00:23:20,279 --> 00:23:22,599
shouldn't do big analysis. Like, some of the

647
00:23:22,599 --> 00:23:26,025
large models have, like, brilliant outcomes and, you

648
00:23:26,025 --> 00:23:28,365
know, it it's just a cost benefit analysis.

649
00:23:28,984 --> 00:23:31,484
It's just saying when if we don't acknowledge

650
00:23:31,544 --> 00:23:33,865
at all the carbon cost, we can't decide

651
00:23:33,865 --> 00:23:35,325
whether or not it's worth it.

652
00:23:35,704 --> 00:23:37,299
So we're just saying if we if in

653
00:23:37,299 --> 00:23:39,220
the first place we consider the carbon cost

654
00:23:39,220 --> 00:23:41,400
the same way we consider the financial cost,

655
00:23:41,940 --> 00:23:43,619
you know, it's not because it costs money

656
00:23:43,619 --> 00:23:45,960
that we don't do things anymore, but instead

657
00:23:46,099 --> 00:23:47,799
before doing it, we say, okay.

658
00:23:48,180 --> 00:23:50,144
Is it still worth doing? Say I'm a

659
00:23:50,144 --> 00:23:52,384
scientist and I go to your website. What

660
00:23:52,384 --> 00:23:54,545
what what happens? What what can it what

661
00:23:54,625 --> 00:23:56,545
how can it help? So the website is

662
00:23:56,545 --> 00:23:58,725
very much is very much a tool.

663
00:23:59,105 --> 00:24:01,505
So I'd say the the important things to

664
00:24:01,505 --> 00:24:04,109
do as a scientist is before running a

665
00:24:04,109 --> 00:24:06,029
large scale project again, I'm not I'm not

666
00:24:06,029 --> 00:24:07,950
saying we should, like, burden all of us

667
00:24:07,950 --> 00:24:09,330
with, like, all the possible

668
00:24:09,869 --> 00:24:12,430
details and every and before right clicking every

669
00:24:12,430 --> 00:24:14,430
button, you should estimate the carbon footprint. Not

670
00:24:14,430 --> 00:24:16,349
at all. But, you know, before running something

671
00:24:16,349 --> 00:24:17,650
significant that will

672
00:24:18,174 --> 00:24:21,295
require hours or days of of computation and,

673
00:24:21,295 --> 00:24:22,994
like, loads of memory, lots of,

674
00:24:23,295 --> 00:24:23,795
cores,

675
00:24:24,654 --> 00:24:26,494
just plug in what you think the time

676
00:24:26,494 --> 00:24:29,394
will be. Just estimate the carbon footprint beforehand

677
00:24:29,775 --> 00:24:31,615
and and look at what the carbon footprint

678
00:24:31,615 --> 00:24:33,634
will be and say, okay. Do I think

679
00:24:33,769 --> 00:24:35,369
it's worth it or am I just running

680
00:24:35,369 --> 00:24:36,809
it for the sake of it and I

681
00:24:36,809 --> 00:24:38,649
don't, you know, actually, I don't really need

682
00:24:38,649 --> 00:24:41,069
it. It's not worth that waste of carbon.

683
00:24:42,889 --> 00:24:44,750
So that's the first thing. Then

684
00:24:45,289 --> 00:24:47,789
mitigate the carbon footprint as much as possible

685
00:24:48,034 --> 00:24:49,554
and I'll I'll just come back to that

686
00:24:49,554 --> 00:24:50,214
in a moment.

687
00:24:51,154 --> 00:24:51,894
And then

688
00:24:52,595 --> 00:24:54,994
afterwards, once, you know, the results are here,

689
00:24:54,994 --> 00:24:57,154
I think it's important we acknowledge what the

690
00:24:57,154 --> 00:24:58,615
carbon cost of that was.

691
00:24:58,994 --> 00:25:01,015
The same way you acknowledge, you know, ethical

692
00:25:01,075 --> 00:25:03,519
concerns after you've done a study with animals

693
00:25:03,519 --> 00:25:06,099
or humans or, you know, just saying, okay,

694
00:25:07,039 --> 00:25:08,259
this was the analysis,

695
00:25:08,559 --> 00:25:10,480
brilliant results, we're really happy with it. And

696
00:25:10,480 --> 00:25:12,320
that's what we've been trying to do and

697
00:25:12,320 --> 00:25:14,000
encourage people to do is to just include

698
00:25:14,000 --> 00:25:16,420
acknowledgements at the end of the publication saying,

699
00:25:16,664 --> 00:25:18,985
running all this work or or this new

700
00:25:18,985 --> 00:25:19,485
tool

701
00:25:20,105 --> 00:25:22,285
cost, like, emitted that much carbon.

702
00:25:22,825 --> 00:25:24,265
And it's not a bad thing. It's just

703
00:25:24,265 --> 00:25:26,184
something so people are aware that this is

704
00:25:26,184 --> 00:25:27,725
the cost of running such analysis.

705
00:25:28,184 --> 00:25:29,884
And it means people who want to replicate

706
00:25:29,945 --> 00:25:31,580
it or people who want to use the

707
00:25:31,580 --> 00:25:33,339
tool, scientists who want to use the tool

708
00:25:33,339 --> 00:25:35,099
later on, well, they can, you know, they

709
00:25:35,099 --> 00:25:36,400
know what they're getting into.

710
00:25:37,180 --> 00:25:39,339
So I think that's that's the the important

711
00:25:39,339 --> 00:25:39,740
thing.

712
00:25:40,299 --> 00:25:42,559
Mitigating obviously is is really important,

713
00:25:43,259 --> 00:25:44,700
and there are many ways to do that.

714
00:25:44,859 --> 00:25:46,539
I would suggest people can check out with

715
00:25:46,539 --> 00:25:46,994
published

716
00:25:47,555 --> 00:25:49,795
paper called paper called 10 simple rules to

717
00:25:49,795 --> 00:25:51,974
make your research more environmentally sustainable.

718
00:25:52,755 --> 00:25:53,255
And,

719
00:25:54,115 --> 00:25:56,515
that's basically, yeah, a very short 10 simple

720
00:25:56,515 --> 00:25:58,115
rules article. So very easy to read, a

721
00:25:58,115 --> 00:25:59,714
bit more easy like, a bit easier than

722
00:25:59,714 --> 00:26:01,315
a long method paper or anything. And and

723
00:26:01,315 --> 00:26:03,389
that just list how what you can do.

724
00:26:03,389 --> 00:26:04,769
But basically, it means,

725
00:26:06,589 --> 00:26:08,750
making the code efficient when you can. I

726
00:26:08,750 --> 00:26:10,909
mean, we've all been there saying I could

727
00:26:10,909 --> 00:26:13,230
either spend half an hour being smart about

728
00:26:13,230 --> 00:26:14,909
the code and reduce, you know, how much

729
00:26:14,909 --> 00:26:15,970
memory I need

730
00:26:16,275 --> 00:26:18,434
for, like, for example, to merge 2 tables

731
00:26:18,434 --> 00:26:20,275
or things like that or say, oh, I

732
00:26:20,275 --> 00:26:22,835
can't be bothered instead I'll just like request

733
00:26:22,835 --> 00:26:25,414
a hundred gigabytes on the HPC server

734
00:26:25,715 --> 00:26:27,955
and I'll just do it in five minute.

735
00:26:27,955 --> 00:26:29,474
And that's, I mean, I've done it, I

736
00:26:29,474 --> 00:26:31,019
I'm many of us have done it, I'm

737
00:26:31,019 --> 00:26:32,859
sure. And and that's a completely valid point,

738
00:26:32,859 --> 00:26:34,720
man. Sometimes it's not worth. But

739
00:26:35,580 --> 00:26:37,820
sometimes it's worth just like plugging in thinking,

740
00:26:37,820 --> 00:26:40,619
okay, well, maybe it's worth being efficient especially

741
00:26:40,619 --> 00:26:42,299
if many people will be using my code

742
00:26:42,299 --> 00:26:44,220
in the future and all things like that.

743
00:26:44,220 --> 00:26:45,759
So just having that in mind,

744
00:26:46,325 --> 00:26:47,865
using updated softwares,

745
00:26:48,565 --> 00:26:50,244
is a painless way to do it. I

746
00:26:50,244 --> 00:26:52,404
know I know everyone hates updating softwares because

747
00:26:52,404 --> 00:26:54,565
it usually breaks the entire pipeline so so

748
00:26:54,565 --> 00:26:56,345
you don't want to touch it. But,

749
00:26:56,644 --> 00:26:59,470
sometimes we we did the example with genome

750
00:26:59,470 --> 00:27:02,130
wide association studies that try to find association

751
00:27:02,349 --> 00:27:02,849
between,

752
00:27:03,950 --> 00:27:06,929
genomic variance and traits. And so they require,

753
00:27:06,990 --> 00:27:08,909
like, large volume of data because lots of

754
00:27:08,909 --> 00:27:10,929
participants and so they are they're quite computationally

755
00:27:11,069 --> 00:27:13,069
intensive which try to find association between a

756
00:27:13,069 --> 00:27:14,130
lot of things together.

757
00:27:15,325 --> 00:27:17,644
And these, we find by just updating from

758
00:27:17,644 --> 00:27:19,244
1 version of a tool to the next

759
00:27:19,244 --> 00:27:19,744
1,

760
00:27:20,444 --> 00:27:23,265
you you would like reduce the carbon footprint

761
00:27:23,325 --> 00:27:24,865
by 50 or 60%,

762
00:27:26,044 --> 00:27:27,644
just because and and that's entire you know,

763
00:27:27,644 --> 00:27:29,590
just because the authors made the the package

764
00:27:29,590 --> 00:27:31,870
a lot more efficient. So in many cases,

765
00:27:31,870 --> 00:27:33,870
actually, you you would reduce run time so

766
00:27:33,870 --> 00:27:36,690
it would make life easier life easier anyway.

767
00:27:37,070 --> 00:27:40,130
It's not just about these huge computer processes,

768
00:27:40,269 --> 00:27:42,350
though. There's a message in here for all

769
00:27:42,350 --> 00:27:45,924
of us. The total life cycle environmental impact

770
00:27:45,924 --> 00:27:46,664
of your laptop,

771
00:27:47,285 --> 00:27:48,644
between 7080%

772
00:27:48,644 --> 00:27:50,424
of it is only due to manufacturing.

773
00:27:51,044 --> 00:27:52,005
So, actually, you know,

774
00:27:52,484 --> 00:27:55,444
if you only charge your laptop every other

775
00:27:55,444 --> 00:27:57,809
day instead of charging it every day, Okay.

776
00:27:57,809 --> 00:28:00,130
But you're only acting on 20% of the

777
00:28:00,130 --> 00:28:02,289
total share because the bulk of it is

778
00:28:02,289 --> 00:28:05,750
extracting the raw material fabric like, making it,

779
00:28:06,049 --> 00:28:07,649
shipping it to you, and things like that.

780
00:28:07,649 --> 00:28:08,149
So

781
00:28:08,529 --> 00:28:09,029
typically,

782
00:28:09,329 --> 00:28:11,909
keeping your device, phone, tablet, laptop

783
00:28:12,234 --> 00:28:14,654
for longer is much better

784
00:28:14,954 --> 00:28:16,014
than, you know,

785
00:28:16,634 --> 00:28:18,794
trying to not charge it as much or

786
00:28:18,794 --> 00:28:20,794
things like that. So that's why they're, like,

787
00:28:20,794 --> 00:28:23,434
keep, repair, reuse, moto is is quite a

788
00:28:23,434 --> 00:28:25,674
good 1. So sort of updating your phone

789
00:28:25,674 --> 00:28:27,855
every year is not an ideal scenario?

790
00:28:28,210 --> 00:28:30,450
No. It's it it really it really isn't.

791
00:28:30,450 --> 00:28:32,850
No. Regular listeners to the Physics World Stories

792
00:28:32,850 --> 00:28:35,330
podcast will remember our Sept. 0 episode where

793
00:28:35,330 --> 00:28:36,470
we explored the importance

794
00:28:36,930 --> 00:28:39,410
of free and open source software. The code

795
00:28:39,410 --> 00:28:40,470
for green algorithms

796
00:28:41,105 --> 00:28:44,065
is freely available on GitHub. Well, the idea

797
00:28:44,065 --> 00:28:45,285
from the start was

798
00:28:45,825 --> 00:28:47,424
we can't have a black box and, you

799
00:28:47,424 --> 00:28:49,265
know, scientists put on numbers and then they

800
00:28:49,265 --> 00:28:51,424
have like an a carbon footprint popping out

801
00:28:51,424 --> 00:28:53,105
without being able to check where it comes

802
00:28:53,105 --> 00:28:53,549
from.

803
00:28:54,110 --> 00:28:55,710
We thought it was really important people could

804
00:28:55,710 --> 00:28:57,070
trust it, and it would be so that's

805
00:28:57,070 --> 00:28:58,610
why everything is open source.

806
00:28:59,230 --> 00:29:00,529
All the data is available

807
00:29:01,070 --> 00:29:03,490
there. All the papers on the topic are,

808
00:29:04,190 --> 00:29:05,330
are also open access,

809
00:29:05,710 --> 00:29:07,309
because we thought, you know, if someone wants

810
00:29:07,309 --> 00:29:08,610
to dive in and check

811
00:29:08,984 --> 00:29:11,325
what, you know, how we made the calculation

812
00:29:11,384 --> 00:29:14,105
exactly or what the code does or what

813
00:29:14,105 --> 00:29:16,825
data we use, it's completely out there. It's

814
00:29:16,825 --> 00:29:18,105
been I mean, it's been great. I know

815
00:29:18,105 --> 00:29:19,325
some people, for example,

816
00:29:19,785 --> 00:29:21,244
got in touch because they

817
00:29:21,720 --> 00:29:23,400
they were like, oh, the calculator is great,

818
00:29:23,400 --> 00:29:24,759
but we would like to, like, make a

819
00:29:24,759 --> 00:29:27,000
local version for our own institution to have,

820
00:29:27,000 --> 00:29:29,480
like, our own hardware on it. Can and

821
00:29:29,480 --> 00:29:31,160
and on GitHub, it's really easy. You can

822
00:29:31,160 --> 00:29:32,679
just fork it. You know? You just download

823
00:29:32,679 --> 00:29:34,039
it and you have your own copy and

824
00:29:34,039 --> 00:29:35,400
you can edit the code and it's your

825
00:29:35,400 --> 00:29:37,414
own version. And that was brilliant. That's the

826
00:29:37,414 --> 00:29:39,335
goal, you know. It doesn't necessarily has to

827
00:29:39,335 --> 00:29:41,414
be our thing or I, you know, I

828
00:29:41,494 --> 00:29:43,255
I'm really happy when people do these kind

829
00:29:43,255 --> 00:29:44,775
of things. Been great to be able to

830
00:29:44,775 --> 00:29:45,974
share all the code. I think it's quite

831
00:29:45,974 --> 00:29:47,575
an important part of of this kind of

832
00:29:47,575 --> 00:29:50,309
project. Simon Porsche Wiesbad worked with some other

833
00:29:50,309 --> 00:29:53,349
scientists to calculate the impact of astronomy and

834
00:29:53,349 --> 00:29:53,849
astrophysics,

835
00:29:54,470 --> 00:29:57,109
not just looking at the computational side of

836
00:29:57,109 --> 00:29:58,890
things. It was a sort of consortium,

837
00:29:59,349 --> 00:29:59,849
effort,

838
00:30:00,390 --> 00:30:00,890
by,

839
00:30:01,484 --> 00:30:04,684
the environmentally committee of the Dutch, astronomical society

840
00:30:04,684 --> 00:30:06,384
or the Dutch astronomical community,

841
00:30:07,805 --> 00:30:09,724
led by NOVA and and what we call

842
00:30:09,724 --> 00:30:10,944
the RAFA and the Astronomy

843
00:30:11,724 --> 00:30:12,384
in Dutch.

844
00:30:13,164 --> 00:30:14,410
And what what we try to do is

845
00:30:14,410 --> 00:30:16,330
make an inventory of how much energy actually

846
00:30:16,330 --> 00:30:18,349
are we using. What is actually our footprint,

847
00:30:18,970 --> 00:30:20,509
if I can say it that way,

848
00:30:21,049 --> 00:30:21,930
in terms of,

849
00:30:22,410 --> 00:30:24,570
of of carbon production or in terms of

850
00:30:24,570 --> 00:30:26,670
renewable or nonrenewable energy sources?

851
00:30:27,565 --> 00:30:29,244
And it was a very interesting thing to

852
00:30:29,244 --> 00:30:31,085
do because it it showed that,

853
00:30:31,404 --> 00:30:33,244
computing is not a big part of that,

854
00:30:33,244 --> 00:30:35,105
at least as far as we could determine.

855
00:30:36,445 --> 00:30:38,705
The travel, which is not a complete surprise,

856
00:30:39,805 --> 00:30:40,865
takes the biggest,

857
00:30:41,244 --> 00:30:43,500
hits. And and after travel,

858
00:30:44,039 --> 00:30:46,440
it is probably just the the the buildings

859
00:30:46,440 --> 00:30:47,740
itself and the commuting

860
00:30:48,119 --> 00:30:48,940
we are doing.

861
00:30:49,880 --> 00:30:51,960
And I think the interesting thing for me

862
00:30:51,960 --> 00:30:52,759
was that,

863
00:30:53,240 --> 00:30:54,944
so there was a second study of,

864
00:30:55,424 --> 00:30:58,164
the big European conference, the the European,

865
00:30:58,625 --> 00:30:59,684
Astronomical Conference

866
00:31:00,384 --> 00:31:02,865
held in 02/18

867
00:31:02,865 --> 00:31:04,005
in Lyon

868
00:31:04,544 --> 00:31:07,024
in France and in 02/20

869
00:31:07,024 --> 00:31:07,505
in,

870
00:31:07,825 --> 00:31:09,664
Leiden in in the city where my university

871
00:31:09,664 --> 00:31:10,164
is

872
00:31:10,909 --> 00:31:13,730
online because of, of of the COVID pandemic.

873
00:31:14,509 --> 00:31:17,390
And the difference in energy of 1 conference

874
00:31:17,390 --> 00:31:19,409
to the other was a factor of 30

875
00:31:20,269 --> 00:31:21,409
just because of,

876
00:31:22,269 --> 00:31:23,809
mostly of due to the travel.

877
00:31:25,184 --> 00:31:27,984
So travel is way more environmentally unfriendly as

878
00:31:27,984 --> 00:31:30,224
we know. Right? I mean, this is not

879
00:31:30,224 --> 00:31:32,544
a complete surprise, of course. But what I

880
00:31:32,544 --> 00:31:34,164
think what is a little bit of surprise

881
00:31:34,464 --> 00:31:35,204
is that,

882
00:31:36,304 --> 00:31:38,005
more than half of the,

883
00:31:38,830 --> 00:31:40,029
total amount of,

884
00:31:40,509 --> 00:31:42,609
c o 2 produced in travel

885
00:31:43,230 --> 00:31:45,150
is from less than 10% of the people

886
00:31:45,150 --> 00:31:46,990
attending the conference, which are the people who

887
00:31:46,990 --> 00:31:48,930
travel from all the way around the world,

888
00:31:49,549 --> 00:31:51,970
take a flight from Australia or from Chile,

889
00:31:52,595 --> 00:31:54,375
in order to get to The Netherlands.

890
00:31:55,154 --> 00:31:58,035
And the short trips are not that bad

891
00:31:58,035 --> 00:31:59,815
even though they are much more frequent.

892
00:32:00,275 --> 00:32:01,955
So, but is the solution to that that

893
00:32:01,955 --> 00:32:04,450
you just have these conferences online instead? Well,

894
00:32:04,450 --> 00:32:06,930
clearly not completely because if the the biggest

895
00:32:06,930 --> 00:32:08,070
traveling people,

896
00:32:09,890 --> 00:32:12,049
they are they are indeed contributing a lot,

897
00:32:12,049 --> 00:32:14,070
but they they may still want to be

898
00:32:14,289 --> 00:32:14,789
present.

899
00:32:15,970 --> 00:32:17,410
So so I think I think the the

900
00:32:17,410 --> 00:32:20,055
story is is twofold. Right? Yes. You you

901
00:32:20,134 --> 00:32:21,434
maybe you should go

902
00:32:21,974 --> 00:32:23,595
online for certain people

903
00:32:24,055 --> 00:32:25,974
or for certain conferences or for part of

904
00:32:25,974 --> 00:32:27,734
the conference, but you also like the sort

905
00:32:27,734 --> 00:32:29,755
of life feel of a conference. I mean,

906
00:32:30,390 --> 00:32:32,950
during the pandemic, I had a discussion with

907
00:32:32,950 --> 00:32:34,570
a with a mathematician friend

908
00:32:35,269 --> 00:32:37,990
once a week discussing a problem, and we

909
00:32:37,990 --> 00:32:39,450
couldn't get any further

910
00:32:39,830 --> 00:32:41,910
with the problem over the two years of

911
00:32:41,910 --> 00:32:43,289
having Zoom conversations.

912
00:32:43,845 --> 00:32:45,765
And then in this Jan. 0, we came

913
00:32:45,765 --> 00:32:47,865
together for half an hour in a room

914
00:32:48,244 --> 00:32:50,565
sitting together. We solved the problem, basically. And

915
00:32:50,565 --> 00:32:52,325
maybe we needed that two years of of

916
00:32:52,325 --> 00:32:54,644
discussions online in order to solve it. On

917
00:32:54,644 --> 00:32:56,164
the other side, I have the feeling that

918
00:32:56,164 --> 00:32:57,944
the effectiveness of doing science

919
00:32:58,359 --> 00:33:00,440
where you just bang your hats together is,

920
00:33:00,679 --> 00:33:01,740
is is amazingly,

921
00:33:02,119 --> 00:33:04,380
more effective than than having Zoom conversations.

922
00:33:05,319 --> 00:33:06,919
The other thing is that we we still

923
00:33:06,919 --> 00:33:09,339
like this live interaction, this live conversations.

924
00:33:10,200 --> 00:33:12,595
So I can imagine that instead of going

925
00:33:12,595 --> 00:33:14,035
for a conference for three days or a

926
00:33:14,035 --> 00:33:15,795
week or so, you you spent a day

927
00:33:15,875 --> 00:33:17,555
a week before visiting a few people in

928
00:33:17,555 --> 00:33:19,234
the neighborhood and maybe a week after that

929
00:33:19,234 --> 00:33:21,474
visiting another few people in the neighborhood. So

930
00:33:21,474 --> 00:33:22,375
you just do

931
00:33:22,755 --> 00:33:24,595
working visits like you do you did a

932
00:33:24,595 --> 00:33:26,355
hundred years ago. Right? When the moment you

933
00:33:26,355 --> 00:33:29,210
went traveling to another continent by ship and

934
00:33:29,210 --> 00:33:31,950
by by and maybe horse drawn carts,

935
00:33:32,730 --> 00:33:34,809
you didn't go there for five hours to

936
00:33:34,809 --> 00:33:36,890
give a talk and then, have hop on

937
00:33:36,890 --> 00:33:38,669
on on the the the

938
00:33:39,049 --> 00:33:41,305
the the ship again. Right? So you spend

939
00:33:41,305 --> 00:33:43,144
much more time and and travel was a

940
00:33:43,144 --> 00:33:43,884
part of

941
00:33:44,265 --> 00:33:44,924
the endeavor.

942
00:33:45,384 --> 00:33:46,985
And I think we should we we probably

943
00:33:46,985 --> 00:33:48,825
should go back to that. There's certainly an

944
00:33:48,825 --> 00:33:51,305
argument for that. Stepping away from physics and

945
00:33:51,305 --> 00:33:53,519
thinking about the world as a whole, it's

946
00:33:53,519 --> 00:33:56,320
tempting to think that perhaps it's other uses

947
00:33:56,320 --> 00:33:59,519
of supercomputers, things like mining Bitcoins that should

948
00:33:59,519 --> 00:34:02,080
be the focus of people trying to reduce

949
00:34:02,080 --> 00:34:04,400
the impact. But there's probably a whole series

950
00:34:04,400 --> 00:34:07,279
of podcasts we could do discussing the various

951
00:34:07,279 --> 00:34:10,765
benefits of science versus finance. Let's leave that

952
00:34:10,824 --> 00:34:12,824
for another time perhaps. But when it comes

953
00:34:12,824 --> 00:34:15,864
to blockchain, the technology underpinning Bitcoin and other

954
00:34:15,864 --> 00:34:16,364
cryptocurrencies,

955
00:34:16,905 --> 00:34:19,785
its uses vary massively in their ethics and

956
00:34:19,785 --> 00:34:20,285
environmental

957
00:34:20,664 --> 00:34:21,164
impacts.

958
00:34:21,699 --> 00:34:23,219
And you can hear a lot more about

959
00:34:23,219 --> 00:34:26,199
that in an interview with computer scientist Suzanne

960
00:34:26,260 --> 00:34:26,760
Koehler.

961
00:34:27,219 --> 00:34:29,780
Suzanne Koehler in an episode of the Physics

962
00:34:29,780 --> 00:34:31,000
World Weekly podcast

963
00:34:31,300 --> 00:34:32,840
from April.

964
00:34:32,900 --> 00:34:34,980
And there's no doubt from the research that's

965
00:34:34,980 --> 00:34:37,315
been done and as highlighted in the recent

966
00:34:37,315 --> 00:34:38,535
IPCC report

967
00:34:38,835 --> 00:34:41,394
that every bit that we can do in

968
00:34:41,394 --> 00:34:42,375
our own endeavors

969
00:34:42,914 --> 00:34:45,635
to help fight climate change is something that

970
00:34:45,635 --> 00:34:47,655
we should very, very seriously

971
00:34:48,114 --> 00:34:48,614
consider.

972
00:34:49,460 --> 00:34:51,539
After all, without a habitable planet to live

973
00:34:51,539 --> 00:34:54,579
on, there's no point and nobody to point

974
00:34:54,579 --> 00:34:56,599
telescopes up to the skies.

975
00:34:56,900 --> 00:34:59,139
I promised you we'd return to the subject

976
00:34:59,139 --> 00:35:01,380
of Oumuamua, and the question that I couldn't

977
00:35:01,380 --> 00:35:03,674
help but ask was if Oumuamua

978
00:35:04,054 --> 00:35:05,734
is the type of object that we should

979
00:35:05,734 --> 00:35:07,594
be seeing fairly regularly,

980
00:35:08,295 --> 00:35:10,714
why haven't we seen them more regularly?

981
00:35:11,094 --> 00:35:13,034
Yeah. I think that's a very good question.

982
00:35:13,734 --> 00:35:16,059
And and I I I wondered about that,

983
00:35:16,059 --> 00:35:17,980
and I'm I'm not thinking too deeply about

984
00:35:17,980 --> 00:35:19,199
it at the moment, but,

985
00:35:19,500 --> 00:35:21,119
I definitely would think that

986
00:35:21,900 --> 00:35:23,659
more would be seen. 1 other 1 was

987
00:35:23,659 --> 00:35:26,400
was seen. Borisov was discovered a year later.

988
00:35:27,494 --> 00:35:28,855
But I would expect that,

989
00:35:29,335 --> 00:35:30,775
that a few more would have been seen

990
00:35:30,775 --> 00:35:32,534
by now. I I I don't know the

991
00:35:32,534 --> 00:35:33,914
reason. It could be statistics,

992
00:35:34,534 --> 00:35:36,635
but I my hopes are on,

993
00:35:37,014 --> 00:35:39,034
the Vera Rubin Telescope, so LSST,

994
00:35:40,219 --> 00:35:42,619
which will, I hope, see, a daily,

995
00:35:42,940 --> 00:35:43,440
Oumuamua

996
00:35:43,980 --> 00:35:44,719
type object.

997
00:35:45,579 --> 00:35:46,079
And,

998
00:35:46,780 --> 00:35:48,400
and, of course, James Webb,

999
00:35:49,260 --> 00:35:51,099
which doesn't have a big field of view,

1000
00:35:51,099 --> 00:35:52,559
but, you know, it's it's so

1001
00:35:52,934 --> 00:35:54,155
sharp and and,

1002
00:35:54,775 --> 00:35:55,434
you know,

1003
00:35:55,735 --> 00:35:57,735
effective that it will discover a lot of

1004
00:35:57,735 --> 00:35:59,815
these objects, I think. So I I think

1005
00:35:59,815 --> 00:36:01,114
it's just a matter of waiting.

1006
00:36:02,055 --> 00:36:04,375
And the typical thing that astronomers do, now

1007
00:36:04,375 --> 00:36:06,135
we have 2 objects, yeah, Oumu Oumu and

1008
00:36:06,135 --> 00:36:07,994
bodies of. They seem to be 2 classifications.

1009
00:36:08,400 --> 00:36:10,559
Right? 1 with gas and 1 without gas.

1010
00:36:10,559 --> 00:36:12,339
The third one will be a third class.

1011
00:36:12,719 --> 00:36:14,639
Of course, it will. And thank you so

1012
00:36:14,639 --> 00:36:16,960
much to Simon and Loic for talking to

1013
00:36:16,960 --> 00:36:18,719
me. If you'd like to know more about

1014
00:36:18,719 --> 00:36:20,799
the main topic of this podcast, then I

1015
00:36:20,799 --> 00:36:23,839
can highly recommend the feature by Michael Allen

1016
00:36:23,839 --> 00:36:26,775
on physics world dot com, the huge carbon

1017
00:36:26,775 --> 00:36:28,155
footprint of supercomputing.

1018
00:36:28,855 --> 00:36:30,295
And if you'd like to know more about

1019
00:36:30,295 --> 00:36:30,795
Oumuamua,

1020
00:36:31,175 --> 00:36:33,114
I highly recommend several articles

1021
00:36:33,574 --> 00:36:35,335
on physicsworld.com.

1022
00:36:35,335 --> 00:36:37,414
And we'll be back next month with something

1023
00:36:37,414 --> 00:36:39,994
else from this wonderful world of physics.

1024
00:36:40,400 --> 00:36:42,099
Thank you very much for listening.

1025
00:36:47,280 --> 00:36:48,500
Physics world.