how much the star dims depends on what colour filter you
observe through. If you can detect only red light, you’ll see the
star dim less than if you can only detect blue. This is not behav-
iour readily caused by solid objects (although some have—I think
in jest—suggested that that we’re seeing alien Christmas lights
hung on the ‘outside’ of their space station). Rather, it likely indicates that the light is being blocked by something like a cloud of
dust. We don’t have all the details yet, but we seem to be begin-
ning to close in on the end of the mystery.
Stories like those of the Voorwerp and WTF star are fascinating
to me because they are the kind of thing I imagined astronomers
did when I was that small kid with a telescope. The discoveries
may be more about finding new ways to investigate the Universe’s
mysteries rather than something to name after me, but who
cares? This image, and this way of working, couldn’t be further
from the usual rhetoric about the onward progress of ‘big data’, in
which we solve problems by writing database queries.
So what had to happen for these discoveries to take place?
First, large surveys—the biggest of big data—need to exist. You
can only find the unusual by fishing in a very large sea; fewer
224 Is It AlIens?
than 1 per cent of galaxies have a Voorwerp and the WTF star is
at the very least only one in 150,000, and so we do need to collect
many, many data points. Then we need to pay individual atten-
tion to each, asking if it is unusual and worthy of interest.
This step is what the Zooniverse projects provide. Professional
astronomers couldn’t possibly look at each system individually,
and—even better—people like Hanny become advocates for
their objects. It’s not that the public are more curious or inter-
ested than professional scientists, but my guess is that citizen sci-
entists are more likely to interrupt their core task to consider a
curiosity. We’re all walking along the same path, if you will, but
if you’re not being paid by the mile you’re much more likely to
stop and smell the flowers, and notice an interesting insect or
two while you do so.
Enough data, and enough citizen scientists, and you can spot
the truly interesting stuff, but that’s not sufficient. Not every-
thing that is interesting is significant. Galaxy Zoo volunteers
found galaxies shaped like letters of the alphabet, and Alice
Sheppard, our moderator, adopted a beautifully penguin-shaped
galaxy as her avatar. Plenty of volunteers have been taken aback
by the sight of a bright green streak shooting from one side of an
image to the other—not (sadly?) an alien space laser but the track
of a satellite captured by the telescope by accident. Here, the for-
ums that are attached to the project—and more to the point the
communities that gather in and around them—are extremely
important.
On projects where we’ve seen this sort of serendipitous dis-
covery happen, it’s usually been because of a community of citi-
zen scientists who can sort through the novel discoveries of
thousands of classifiers, distinguishing the humdrum satellites
from the unusual galaxies. Often, this group do plenty of work
before turning to the professionals, aided by access to raw data
Is It AlIens? 225
from public surveys; Daryll’s investigation of the strange WTF
star is a case in point. Together with their classifying colleagues,
citizen science communities provide a wonderful filter to iden-
tify the most unusual and interesting objects. That’s what I think
is the greatest single reason for trying to preserve citizen science
like the Zooniverse as machines get better; collectively we can
find not only unusual objects but also new questions.
One way of thinking about this was given to me while listen-
ing to talks by the team behind another Zooniverse project, in an
area far from astronomy. Shakespeare’s World asks volunteers to
transcribe material from the sixteenth-century collections held
by the Folger Library in Washington, DC, in order to try to help
understand what life was like back then and to trace the history
of language. The project is a distant descendent of Old Weather,
and, as in that project, participants have paused along the way to
investigate all sorts of curious finds.
A volunteer whose screen name is mutabilitie (sadly, we don’t
know their real name) found a 1567 letter containing the lines
‘Albeit I do assure you he is vnsusspected of | any vntruithe or
oder notable cryme (excepte a whyte lye)’, the oldest recorded
instance of the phrase ‘white lie’ by more than 150 years, now
recorded in the Oxford English Dictionary. You can also see why the appearance of a recipe ‘To make mackroones or portugall farts’
drew the attention of volunteers; ‘farts’ were little, light pastries, no more than puffs of air.
One of the researchers concentrating on recipes for farts and
other things, Lisa Wright, explained to me that what the volun-
teers were doing was a technique known as ‘close reading’, com-
mon enough in studies of literature. The idea is to pay attention
to each individual word in a text, working out what each contrib-
utes as well as considering it in its own right. Of course, the prob-
lem is choosing where to focus, but here the volunteers were
226 Is It AlIens?
providing close reading at scale. Because we have a crowd of
volunteers, we can scan every word in a large corpus of material,
or pay individual attention to a million galaxies. Unlike most
attempts to use large data sets to do research, the point isn’t to
take the traditional close reading—close study, if you prefer—of
individual words or objects and replace it with some database
query, clever visualization, or statistical analysis, but rather to
keep the traditional method of analysis alive. The way the crowd
behaves allows us quickly and collectively to home in on the
examples where it will be of most use.
Citizen science, seen like this, is a way of finding the interest-
ing stuff and focusing rigorous, sustained, detailed attention on
it. By being distracted, we can appreciate and try to understand
the unusual. It’s a wonderful way to make discoveries, and a
lovely way to do science. But does it have a future?
9
THREE PATHS
The Zooniverse project has grown so fast and so far that a
decent description of all of our projects we’ve done would
have filled this book and more. I recently spent a day trying to
complete a single classification on the more than seventy pro-
jects currentlylive and wound up exhausted and overstimulated.
In that day, I’d helped biologists map the ocean, had transcribed
ancient Hebrew texts, done all manner of astrophysical tasks,
and measured stuffed birds from the Natural History Museum
collection. I used to know every project intimately, spending
time thinking about the design an
d data of each, but especially
since we launched a tool that allows researchers to quickly build
their own projects instead of relying on web developers, those
days are long gone.
What people are using our tools for is constantly humbling.
Brooke Simmons, the member of the Galaxy Zoo team who led
the work on bulgeless galaxies described in Chapter 4 and now
an astrophysics lecturer at the University of Lancaster, has led an
effort to try to build what she calls the Planetary Response
Network. When a natural disaster happens, Brooke works with
networks of first responders who will be flying in to help with
the relief efforts. Sometimes, as with the earthquakes in Nepal in
228 Three PaThs
2016, the area affected is so remote that there are literally no
maps. In other cases, like during the Caribbean storms of 2017,
the maps need rapid updating to reflect the effect of the disaster
on roads, buildings, and even people.
It’s amazing stuff, made possible by the profusion of Earth-
observation satellites that now exist, capable of imaging any part
of the globe at short notice. The companies that run these con-
stellations of cameras—most notably the Californian start-up
Planet, who have more than 150 orbiting imagers—are also pretty
generous with their data, making them available for genuine
humanitarian efforts. The results are impressive too; in the case
of the Nepalese earthquake Zooniverse volunteers identified an
otherwise unmapped village within the affected zone, directing
personnel from the UK charity Rescue Global to a place they
might not otherwise have visited. The results of the Caribbean
deployment were less spectacular, but those going to the aid of
storm victims in Guadeloupe, Dominica, and Puerto Rico used
maps which included contributions from volunteers. We’re now
working hard to make it easier to include new sources of images
so that we can respond faster in the event of future crises.
When Galaxy Zoo started, we couldn’t imagine doing any-
thing like this. While Earth observation—taking pictures from
space of our home planet—has long been a major reason to
launch satellites into space, the availability of images that are
sharp enough to pick out details such as a landslide blocking a
road was until recently almost exclusively limited to the military,
along with other government agencies and their partners. Even
when high-resolution images were released, they were typically
out of date; scheduling a sudden imaging campaign following a
disaster was next to impossible. Now, the situation is completely
different. I’m sure military technology has moved on too, but the
small-satellite revolution has changed the game entirely.
Three PaThs 229
The way that space technology has moved from being about
cutting-edge, specialist tech to being about clever reuse of com-
ponents developed for other things—including your mobile
phone—is a story whose consequences I think we’re still trying
to understand, but its effects are becoming clear. Because satel-
lites are cheaper, more can be launched.* Because more are up
there, the chances of one flying over any given location in the
next hour or so have increased dramatically, so up-to-date
images are easier than ever to obtain. At the moment, most of
this data is private, used by commercial companies for every-
thing from assessing traffic flows to directing fertilizer spraying
in fields of crops, but I think the day is coming when either public
space agencies like the European Space Agency or NASA, or pri-
vate companies with different business models, will make large
amounts of high-resolution Earth imagery open to anyone for
free. At that point, as long as the tools to use this wealth of data
are also made available, we should expect a flood of citizen sci-
ence projects similar to that seen in astrophysics in the last
decade.
What might those projects look like? There are already
examples of craters associated with asteroid impacts being spot-
ted in satellite images; one, the Kamil Crater in the middle of the
Egyptian Sahara, is forty-five metres across and sixteen metres
deep, yet was first identified by scientists using Google Earth! We
have had projects pitched to the Zooniverse that want to use sat-
ellite imagery to assess the number of street traders in southern
African cities, and thus work out how much of the country’s
economic activity might be taking place in this informal way
* The increasing number of ways of getting your satellite to space helps too; Elon Musk’s SpaceX have played a large role in making it cheaper to launch things, but there are plenty of innovative companies building small and medium-size rockets capable of launching whole constellations of satellites.
230 Three PaThs
rather than in the more traditional economy that shows up in tax
returns. The number of suggested projects which involve assess-
ing human activity around the world is increasing, though so far
we haven’t promoted any of these projects on the Zooniverse
platform; we need more advice from people who aren’t moon-
lighting astronomers before I’m comfortable.
More obscure projects are possible too. My favourite example
is an attempt to settle what is apparently a vigorous scientific dis-
pute between two sets of researchers. One group believes that
cows are sensitive to magnetic fields, and will tend to align with
the prevailing field (the magnetic field, not the farmer’s field),
while another thinks this is nonsense. The only way to settle the
matter is to collect more data. Cows are visible in many satellite
images, so all one would have to do is find a set of volunteers will-
ing to mark bovine orientation on a sufficiently large number of
images.
The examples and possible projects seem endless. But things
are very different now from the time the astronomical citizen
science revolution was beginning, as you’ll have gathered. Ten
years on, the idea of it being a novel thing to invite the public to
participate seems quaint, and new projects don’t get the kind of
media attention that drove the initial success of Galaxy Zoo.
More to the point, machine learning has improved to the point
that building an image analysis project without considering the
complex interactions between humans and machines seems
negligent, if not downright unethical, in the way it would waste
people’s time.
So what future does citizen science of the sort carried out by
the Zooniverse volunteers really have? I’m going to concentrate
on my home turf, on astronomy, and even here I think there are
three different possibilities, three possible paths that we might
find ourselves on which lead forward from where we are now.
Three PaThs 231
Which one we end up with will depend on how much effort
we’re prepared to put into open collaboration, and what kind of
science we want to do. I hope that as many projects as p
ossible
will make the right choices; we certainly intend to.
The first scenario is the one we will reach if we don’t do any-
thing about it. It seems obvious that the current improvement in
machine learning, powered by research carried out not only in
the computer science departments of universities but also by the
increasingly large machine learning teams at Google, Facebook,
and in the rest of Silicon Valley, will continue. Companies from
these giant firms all the way down to the newest start-up clearly
see being ahead in artificial intelligence as essential for twenty-
first-century business, and while at present that mostly means
having a larger labelled training set to teach your robot new
tricks, it also includes innovation in techniques, many of which
are aimed at the kind of problems we encounter in astronomy.
You can make a reasonable case that machines—specifically
convolutional neural networks—can now be trained to do basic
galaxy classification as well, if not better, than the crowd. If you
just want to split spirals from ellipticals, for the vast majority of systems no human intervention is necessary, and it is beginning
to look likely that a system trained on one survey, such as Sloan,
may be able to cope easily with galaxy images coming from com-
pletely different surveys and therefore with different depths, col-
ours, and characteristics. This is happening partly because
splitting spirals from ellipticals is the easiest of the problems that Galaxy Zoo posed to its crowd of volunteers, and partly because
it is the question for which we have the largest volume of data
with which to train the machines.
It shouldn’t be a surprise that this question might pass from
the realm where we need human intervention to that where
machines rule. The same thing happened to the task of separating
232 Three PaThs
images of galaxies from stars, the latter appearing as sharp
points, easily contrasted with the fuzzy blobs of distant star sys-
tems twenty or thirty years ago. I wrote in Chapter 6 about the
supernova-hunting project that put itself out of business in just
this way. What of questions where the training data is not so
abundant?
My group in Oxford now includes a PhD student who is an
expert in machine learning (and critically, articulate in explain-
The Crowd and the Cosmos: Adventures in the Zooniverse Page 27
