talk_script_YY_fosdem2021.ai.srt

1
00:00:00,780 --> 00:00:05,310
Hi, my name is Yo Yehudi. I'm
the open source technology lead

2
00:00:05,310 --> 00:00:08,700
for the data science and health
team at the Wellcome Trust. And

3
00:00:08,700 --> 00:00:11,190
I'm here today to talk a little
bit about what my team does to

4
00:00:11,190 --> 00:00:14,310
actually fund foundational
tools, trust and talent for open

5
00:00:14,310 --> 00:00:18,030
research in life science. And I
also have my twitter and

6
00:00:18,030 --> 00:00:21,390
fediverse handles here. So if
you want to tweet or toot about

7
00:00:21,390 --> 00:00:25,710
the talk, then you can do so.
And also some contact details at

8
00:00:25,710 --> 00:00:29,190
the end of the slides as well,
including my email address. So

9
00:00:29,460 --> 00:00:31,590
I'm going to start by setting a
scene.

10
00:00:32,640 --> 00:00:38,220
First of all, let's talk about
Sam. So Sam is the scientist who

11
00:00:38,220 --> 00:00:41,250
does science in the lab, wearing
a lab coat and perhaps doing

12
00:00:41,250 --> 00:00:44,160
interesting things with
pippettes or microscopes. This

13
00:00:44,160 --> 00:00:47,160
is possibly what the traditional
images that pops into people's

14
00:00:47,160 --> 00:00:50,280
head when they think about the
word scientist, but it's really

15
00:00:50,280 --> 00:00:51,450
only half the picture.

16
00:00:53,010 --> 00:00:56,610
So this case, I will actually
also be talking about Joy. And

17
00:00:56,880 --> 00:01:00,480
primarily, I'll be talking about
Joy. They work in the same area,

18
00:01:00,480 --> 00:01:03,660
Sam, but they focus more on
computational research. And they

19
00:01:03,660 --> 00:01:07,800
tend to write their code, do
their work, to tidy data, maybe

20
00:01:07,800 --> 00:01:11,310
to analyse or visualise their
results. And it might even be

21
00:01:11,310 --> 00:01:14,130
that they couldn't do their work
at all without code, because

22
00:01:14,130 --> 00:01:17,430
maybe they have gigabytes and
gigabytes worth of image data or

23
00:01:17,430 --> 00:01:21,060
some other type of big data that
really does need to be analysed

24
00:01:21,060 --> 00:01:22,170
using computation.

25
00:01:23,350 --> 00:01:26,410
Sometimes these types of data,
there's no way a single human or

26
00:01:26,410 --> 00:01:29,770
even a team could properly
analyse it. And so we do have a

27
00:01:29,770 --> 00:01:32,440
lot of computers in research.
And I suspect that since I'm

28
00:01:32,440 --> 00:01:35,710
speaking to the open research
room at fosdem, many of you may

29
00:01:35,710 --> 00:01:38,530
already be computational
researchers and be familiar with

30
00:01:38,530 --> 00:01:40,660
this. But I think it's important
to set that baseline for

31
00:01:40,660 --> 00:01:43,870
everyone else since this isn't
only an academic conference.

32
00:01:45,610 --> 00:01:48,280
Anyway, code is a non
traditional research and

33
00:01:48,280 --> 00:01:51,220
academic output, which means
that it doesn't really fit into

34
00:01:51,220 --> 00:01:53,890
the same incentive systems that
may already exist within

35
00:01:53,890 --> 00:01:58,810
research. So actually, all of it
both Sam and Joy's work is still

36
00:01:58,810 --> 00:02:02,020
really important, and it's
around a stain definitely.

37
00:02:02,320 --> 00:02:05,740
Today, we'll focus on joy. And
if you identify with Joy, then

38
00:02:05,740 --> 00:02:09,070
hopefully this is a talk for
you. So I'll dig a little bit

39
00:02:09,070 --> 00:02:14,110
into the title of my talk, which
was talking about the tools, the

40
00:02:14,110 --> 00:02:18,400
talent and the incentive. So
what my some of the challenges

41
00:02:18,400 --> 00:02:21,880
that actually people have today
is that we have missing

42
00:02:21,880 --> 00:02:25,600
computational tools. For the
research, should we have

43
00:02:25,600 --> 00:02:29,470
insufficient incentive to draw,
retain and reward talent, and

44
00:02:29,470 --> 00:02:31,990
insufficient trust in
computational work.

45
00:02:33,730 --> 00:02:36,760
Right. So first of all, looking
at tools, there's a couple of

46
00:02:36,760 --> 00:02:39,250
interesting and perhaps almost
opposite cases here that may

47
00:02:39,250 --> 00:02:42,820
occur. On the one hand, research
by its nature is often cutting

48
00:02:42,820 --> 00:02:45,670
edge, or it's just such a very
narrow domain that actually the

49
00:02:45,670 --> 00:02:49,240
software tools don't yet exist
for people need to do. On the

50
00:02:49,240 --> 00:02:51,790
other hand, in some domains,
there's the opposite problem,

51
00:02:51,790 --> 00:02:54,520
where actually the tools are
fragmented. And so there might

52
00:02:54,520 --> 00:02:56,800
be lots and lots of different
standards and lots and lots of

53
00:02:56,800 --> 00:02:59,440
different ways to do something.
Of course, here I have the

54
00:02:59,440 --> 00:03:02,650
obligatory XKCD, talking about
how easy it is to actually

55
00:03:02,650 --> 00:03:06,610
result in fragmented standards.
Anyway, these are two challenges

56
00:03:06,610 --> 00:03:10,120
with regarding to tools I'm sure
there are many, many more. So

57
00:03:10,120 --> 00:03:14,650
I'll also try and talk about
trust. So unfortunately, I think

58
00:03:14,650 --> 00:03:17,200
a lot of us are familiar with
different ways that tech has

59
00:03:17,200 --> 00:03:20,470
gained a bad name, and has
gained in many cases a bad name

60
00:03:20,470 --> 00:03:24,190
for a good reason. For example,
anonymized data is very often

61
00:03:24,400 --> 00:03:26,830
re-identifiable. It's not that
long ago, we heard about the

62
00:03:26,830 --> 00:03:30,400
huge Cambridge Analytica
Facebook data scandal. And

63
00:03:30,400 --> 00:03:33,220
perhaps most recently, people
have been flocking away from

64
00:03:33,220 --> 00:03:36,670
WhatsApp due to privacy
concerns. I think the problem

65
00:03:36,670 --> 00:03:39,610
around trust is that like I
mentioned earlier, computational

66
00:03:39,610 --> 00:03:42,130
research for managing people's
data is really, really

67
00:03:42,130 --> 00:03:46,030
essential. There's so many
scenarios where we can't do it

68
00:03:46,030 --> 00:03:48,580
any way except computationally
do the work that we need to do,

69
00:03:48,580 --> 00:03:51,580
because there's too much data or
it's too is too complex for us

70
00:03:51,580 --> 00:03:54,670
to do it any other way. But at
the same time, if we haven't

71
00:03:54,670 --> 00:03:58,120
acted in trustworthy ways, and
we haven't actually proven that

72
00:03:58,120 --> 00:04:01,240
we deserve the trust of people,
then that people are going to be

73
00:04:01,240 --> 00:04:04,840
understandably wary of sharing
data and of working with

74
00:04:04,840 --> 00:04:08,230
computational tools, which means
it can be really hard to make

75
00:04:08,230 --> 00:04:11,050
the necessary advances that we
actually want to help people

76
00:04:11,050 --> 00:04:12,580
when we do have good intent.

77
00:04:14,070 --> 00:04:18,060
And so I'll talk a little bit
more perhaps about talent as

78
00:04:18,060 --> 00:04:22,950
well. So to set the scene or
perhaps elaborate a little bit

79
00:04:22,950 --> 00:04:27,960
on how research and academia
works. So if you remember, Sam,

80
00:04:28,080 --> 00:04:31,080
the traditional researcher, when
they finish up their lab work,

81
00:04:31,080 --> 00:04:33,690
they'll write up what their
results are. And they'll publish

82
00:04:33,690 --> 00:04:37,410
this paper that they've written
in a journal. And so since this

83
00:04:37,410 --> 00:04:40,680
is this is how things evolved.
The academic credit model

84
00:04:40,680 --> 00:04:43,710
typically revolves around how
many papers you've written, and

85
00:04:43,710 --> 00:04:47,100
what's in those papers. But Joy
does certainly write papers,

86
00:04:47,100 --> 00:04:49,740
Joy's the computational
researcher, so they spend time

87
00:04:49,740 --> 00:04:54,090
writing code. And code isn't yet
built into the system in the

88
00:04:54,090 --> 00:04:58,110
same way. So it's not measured
or incentivized. And ultimately,

89
00:04:58,110 --> 00:05:01,020
if you're writing code, you're
probably not writing papers, and

90
00:05:01,020 --> 00:05:04,320
you may not be earning the types
of research that might advance

91
00:05:04,320 --> 00:05:07,290
your career prospects,
unfortunately.

92
00:05:08,310 --> 00:05:11,880
And so while code produces
scientific outputs, it also

93
00:05:11,880 --> 00:05:14,010
doesn't mean that people
actually appreciate good

94
00:05:14,010 --> 00:05:19,440
practice. So I know as a
research software engineer, I

95
00:05:19,470 --> 00:05:22,230
would appreciate it if I came
back to my own code, and I'd

96
00:05:22,230 --> 00:05:25,410
done good documentation or you
know, written tests or done

97
00:05:25,410 --> 00:05:28,230
whatever other good bit of
practice I should be doing. And

98
00:05:28,230 --> 00:05:30,300
I know that any other
maintainers, who maybe stepped

99
00:05:30,300 --> 00:05:33,960
up after me would also
appreciate that, but I'm not

100
00:05:33,960 --> 00:05:37,260
sure if anyone else would. Right
now, unfortunately, this is

101
00:05:37,260 --> 00:05:40,500
another thing that we really
don't incentivize, which kind of

102
00:05:40,500 --> 00:05:43,170
actually nicely moves on to my
next point.

103
00:05:44,770 --> 00:05:49,150
So another problem with code
related outputs is that they

104
00:05:49,150 --> 00:05:52,240
might not be shared. And this
could be because there wasn't

105
00:05:52,240 --> 00:05:56,140
good practice, like I mentioned
earlier, or it could be related

106
00:05:56,170 --> 00:06:01,570
to other reasons. For example,
people may want to capitalise on

107
00:06:01,750 --> 00:06:06,160
intellectual property or
something else. And, and perhaps

108
00:06:06,160 --> 00:06:08,950
another interesting area is that
researchers may not always wish

109
00:06:08,950 --> 00:06:13,510
to share, that they actually are
writing code. So this could be a

110
00:06:13,510 --> 00:06:17,290
recall when I was at fosdem last
year, someone said that it was

111
00:06:17,290 --> 00:06:20,950
incredibly hard to actually get
a grant to get someone to pay

112
00:06:20,950 --> 00:06:22,450
you to write software.

113
00:06:25,240 --> 00:06:27,970
And one of the consequences of
that is that you might not write

114
00:06:27,970 --> 00:06:30,610
in your grant application that
you're even writing code,

115
00:06:30,640 --> 00:06:32,830
because you may feel like you
need to obscure it just in order

116
00:06:32,830 --> 00:06:35,800
to get funded. But we believe
that you should be able to say

117
00:06:35,800 --> 00:06:38,620
that you want to write code,
because this is how this type of

118
00:06:38,620 --> 00:06:42,280
work is done. And I'm going to
do this code well. So another

119
00:06:42,280 --> 00:06:45,100
interesting thing is that some
people also don't even realise

120
00:06:45,100 --> 00:06:47,890
that they're coders. I recall
having a chat one day with

121
00:06:47,890 --> 00:06:50,470
someone and I explained that I
was a research software

122
00:06:50,470 --> 00:06:54,010
engineer. And she said that that
was kind of interesting. But

123
00:06:54,010 --> 00:06:57,460
also not something she had any
knowledge about. And then later

124
00:06:57,460 --> 00:07:00,010
in that same conversation, she
mentioned that she actually uses

125
00:07:00,010 --> 00:07:03,190
our every single day. So I'm
like, well, this is this is

126
00:07:03,190 --> 00:07:06,670
code, and it's an important part
of science, you are a coder. But

127
00:07:06,670 --> 00:07:08,950
of course, if she doesn't think
that she's doing code, then

128
00:07:08,950 --> 00:07:10,870
she's probably not going to be
mentioning this in her grant

129
00:07:10,870 --> 00:07:14,380
applications. And as a funder,
we don't necessarily get a

130
00:07:14,380 --> 00:07:17,230
realistic picture of what's
going on if people aren't

131
00:07:17,230 --> 00:07:20,800
actually feeling like they can
talk about code at all, as part

132
00:07:20,800 --> 00:07:23,980
of science. So we really want
people to be able to open up and

133
00:07:23,980 --> 00:07:27,340
share and tell us what they're
doing around code, so that we

134
00:07:27,340 --> 00:07:30,430
can incentivize and reward it
and understand it a bit better.

135
00:07:33,630 --> 00:07:36,660
Another problem in the talent
area is that we are in an

136
00:07:36,720 --> 00:07:39,480
intersection of places where
diversity tends to be really

137
00:07:39,480 --> 00:07:44,580
bad. GitHub survey in 2017 found
that over 90% of open source

138
00:07:44,580 --> 00:07:48,330
contributors were men. And a
recent report from the software

139
00:07:48,330 --> 00:07:52,200
Sustainability Institute in the
UK similarly found that 6% of UK

140
00:07:52,200 --> 00:07:55,470
research software engineers
reported as disabled, and 5%

141
00:07:55,470 --> 00:07:58,680
reported as black or minority
ethnicity were at the

142
00:07:58,680 --> 00:08:02,580
intersection of so many of these
different places, which means

143
00:08:02,580 --> 00:08:04,800
that we're probably not getting
the best quality outputs that we

144
00:08:04,800 --> 00:08:08,340
could get, since we only have a
real small subset of people that

145
00:08:08,340 --> 00:08:10,770
there actually are available
working on this and putting

146
00:08:10,770 --> 00:08:12,420
their views into what's going
on.

147
00:08:55,280 --> 00:08:58,730
I will also talk a little bit
about the Wellcome Trust

148
00:08:58,760 --> 00:09:01,910
strategy. So sort of previously,
I've been talking about some of

149
00:09:01,910 --> 00:09:04,460
the problems in the domain and
some of the things that I think

150
00:09:04,460 --> 00:09:07,970
my team are really interested in
doing. But the Wellcome Trust is

151
00:09:07,970 --> 00:09:12,140
a science funder largely. And
right now we've sort of shifted

152
00:09:12,140 --> 00:09:15,140
our focus to look at really
three primary areas as well as

153
00:09:15,470 --> 00:09:18,620
discovery research, which is a
bit broader, but we have a

154
00:09:18,620 --> 00:09:22,610
really strong focus on global
heating on mental health and on

155
00:09:22,640 --> 00:09:24,050
infectious disease.

156
00:09:27,650 --> 00:09:30,830
And I will also give you some
examples, I've got a couple of

157
00:09:30,830 --> 00:09:34,490
open source software projects
that we've actually just funded

158
00:09:34,490 --> 00:09:37,430
really recently both of these
have been co funded with other

159
00:09:37,430 --> 00:09:41,990
funding organisations or in this
case, afrimapr was co funded

160
00:09:41,990 --> 00:09:46,940
with Wellcome's Open Research
team. And it is a Shiny/R-based

161
00:09:46,970 --> 00:09:50,750
tool that actually allows you to
look up healthcare facility

162
00:09:50,750 --> 00:09:54,290
locations in Africa is
integrated data both I think

163
00:09:54,290 --> 00:09:56,810
from citizen science data as
well as governmental data

164
00:09:56,810 --> 00:09:57,500
sources

165
00:09:59,870 --> 00:10:04,220
Another open source project that
we've recently co funded is

166
00:10:04,250 --> 00:10:08,750
OpenSAFELY, which allows people
to allows

167
00:11:07,070 --> 00:11:11,000
that you can find maintenance
money for, for example, and that

168
00:11:11,030 --> 00:11:14,270
you feel like you're justified
doing documentation for users

169
00:11:14,270 --> 00:11:17,600
and software developers who want
their code to be accessible, we

170
00:11:17,600 --> 00:11:21,500
want to make sure that... whilst
FAIR, FAIR is great, and this

171
00:11:21,500 --> 00:11:23,870
time I'm also talking about
things like making sure that

172
00:11:24,140 --> 00:11:27,350
your code works with screen
readers, if that's relevant, or

173
00:11:27,350 --> 00:11:31,820
that there is captioning, for
example, if there is audio

174
00:11:31,820 --> 00:11:37,280
content, and a huge, huge part
of open source. Success is

175
00:11:37,280 --> 00:11:41,930
community. And, as I mentioned,
many of the people here and we

176
00:11:41,930 --> 00:11:44,510
are an open source software
conference, people are aware

177
00:11:44,510 --> 00:11:47,420
that very often putting your
code online isn't enough to make

178
00:11:47,420 --> 00:11:50,750
people contribute. But actually,
you may want community managers

179
00:11:50,750 --> 00:11:54,710
and people who spend a lot of
time thinking and working with

180
00:11:54,710 --> 00:11:57,620
people to actually bring in
contributions. And this is an

181
00:11:57,620 --> 00:12:00,590
important part of measuring the
sustainability of software as

182
00:12:00,590 --> 00:12:00,980
well.

183
00:12:29,670 --> 00:12:35,520
here's a couple of calls to
arms. And so if you, if this has

184
00:12:35,520 --> 00:12:38,280
sort of spoken to you, and you
have some examples of things

185
00:12:38,280 --> 00:12:39,480
that you would like to share,

186
00:12:40,770 --> 00:12:43,410
please, please contact me my
email address is on the next

187
00:12:43,410 --> 00:12:46,920
page, we are launching at the
minute a blog to spotlight good

188
00:12:46,920 --> 00:12:49,260
practice in open source research
software.

189
00:13:23,440 --> 00:13:27,190
is that we want to be funding
software tools. So if you think

190
00:13:27,190 --> 00:13:30,160
you fit into life science
domains, either as a really

191
00:13:30,160 --> 00:13:34,900
foundational software tool, or
one of the three specific areas

192
00:13:34,900 --> 00:13:37,510
that I mentioned, that Wellcome
is focusing on - mental health

193
00:13:37,540 --> 00:13:41,710
and infectious diseases and
global heating, then please,

194
00:13:41,740 --> 00:13:45,250
please do contact me. So my
email address is here. And

195
00:13:45,250 --> 00:13:47,710
there's also the whole team's
email address if you prefer

196
00:13:47,710 --> 00:13:48,040
that.

197
00:13:49,200 --> 00:13:52,140
But I'm also happy to sound off
ideas and just discuss things if

198
00:13:52,140 --> 00:13:54,030
you think that there's something
that you're not sure what might

199
00:13:54,030 --> 00:13:56,610
be a good fit, or if you need to
talk through a bit more before

200
00:13:56,610 --> 00:14:00,660
spending time and effort
thinking about it. And with

201
00:14:00,660 --> 00:14:03,330
that, I'll say thank you very
much and happy to take any

202
00:14:03,330 --> 00:14:04,590
questions. Thank you