elliptical galaxies to use. Looking up, I saw that the man inter-
rupting was on the other side of a long table, short, dark haired
and smartly dressed. I was wearing a suit (usually a giveaway in
academia that a job is on the line), but my questioner was smartly
dressed too, the only other person in the room wearing a jacket.
I hadn’t noticed him when I walked in, and had no idea who he
was, but now he was leaning forward, hands steepled together
as he tore into the way I’d selected the galaxies used in my study.
The problem, it seemed, was that I hadn’t been nearly careful
enough in picking out a selection of nice elliptical systems. I’d
assumed the red galaxies in my sample were elliptical, but, the
heckler said, this isn’t true. Not all red galaxies are elliptical, he said, and he knew this because he’d looked at them.
I eventually just shrugged and said I’d done what many others
had done in picking out galaxies by colour and moved on, but I
was distracted for the rest of the talk. I must have said something
sensible, though I couldn’t tell you what, and a few months later
* We blame a mysterious force called ‘dark energy’, which we will meet again in Chapter 6.
60 The Crowd and The Cosmos
I found myself wandering up the concrete stairs of the Denys
Wilkinson Building, home to astrophysics in Oxford, to begin
my post-PhD life as a postdoc in Oxford.
My task was to look at how the chemistry associated with star
formation might change in different galaxies, and for that I
needed to find star formation happening in as wide a variety of
galaxies as possible. That meant I couldn’t do what I’d done
before, and just select red galaxies in the confident belief that
most of them would be elliptical. Instead, I found myself chatting
to the smartly dressed chap who’d nearly derailed my interview,
and who turned out not to be a senior lecturer with a grudge but
a precocious graduate student from Switzerland named Kevin
Schawinski.
Kevin had demonstrated that it was possible to distinguish
ellipticals from spirals, but he hadn’t enjoyed the solution very
much. Looking at an image of a random galaxy and being able to
distinguish fuzzy spiral arms turns out to be something of a
human speciality. Just before I turned up in Oxford, blithely
throwing my sample of ellipticals together, Kevin had spent a
week doing nothing but looking at images of galaxies. He would
eventually work his way through 50,000 Sloan images, demon-
strating that at this particular task humans still have the measure
of computers.
The fact that the software that runs on the lump of matter
between our ears can outperform that running on our laptops
perhaps shouldn’t be too much of a surprise when it comes to a
pattern recognition task like this one. The truth is that evolution
has left the human race staggeringly well equipped for galaxy
classification, albeit as an unintended consequence of making us
good at pattern recognition in general. Think about walking
through a busy town centre, preoccupied with a shopping list or
the cares of the world. You pass people, tens or even hundreds of
The Crowd and The Cosmos 61
them, before stopping dead as an old friend or new acquaintance
hoves into view. This is the kind of pattern recognition task we
all perform thousands of times a day without even thinking
about it, but it’s still hugely challenging for computers. Yes, facial recognition software has come a long way—my computer makes
a half-decent stab at guessing the identities of my few closest
friends when they appear in uploaded photos—but only after
the investment of hundreds of millions of dollars in research
funding. Sitting in Oxford, thinking about classifying millions of
galaxies, the best Kevin could do was knuckle down and start
looking. The experience left him with a nice clean sample of
elliptical galaxies and, as I found out, some strong opinions about
the right way to do things.
Kevin was right to be convinced about the virtues of visual
inspection. Unfortunately, the research I wanted to do required
classifications for the entire survey. In the old days, back when
surveys contained only hundreds or thousands of galaxies, this
wouldn’t have been a big deal, but neither Kevin nor I were keen
to spend the best part of five months doing nothing but classify-
ing galaxies. There was a bigger problem, too. Kevin’s classifica-
tions are all very well, but any truly radical result that came from
using the classifications would be vulnerable to the charge that
he simply didn’t know what he was doing. Don’t agree with the
classifications? Criticizing the classifier would have been a sensi-
ble tactic, and the only way to ameliorate its effect would be to
get a second classifier to work through the same set of galaxies,
turning months of effort into years.
Clearly this wasn’t a sensible use of anyone’s time. I tried buy-
ing Kevin plenty of beer in the Royal Oak, the traditional water-
62 The Crowd and The Cosmos
ing hole for Oxford astronomers,* but despite this lubrication he
wasn’t keen on further classification either. Sitting among the
grand old beams of the pub, we realized that the only solution
available was to call for help. Since I’d been a kid with a telescope, I’d known (and dreamed about) the discoveries that amateur
astronomers could make; here was a chance for them to help me.
Selfish, perhaps, but the thought did occur to me straight away
that this was a way that anyone could contribute, without need-
ing to spend thousands of pounds on a telescope.
The plan, quickly formed, was simple. Leave the pub. Call in a
few favours from people we knew who could build websites to
get something simple put together. Give talks to local astronom-
ical societies, including increasingly desperate pleas to help with
galaxy classification. Say fifty people in each audience, each of
whom do 200 classifications each. Give two talks a month for
five years, and we should have had everything classified once by
2012. Would it work? Would people really give up their spare
time to help with my work?
I was confident this might be a plan crazy enough to work.
I’d heard about—and tried to participate in—a project called
Stardust@home which had sent tens of thousands of people
searching through blurry images of dust grains which had been
brought back from a comet by a robotic spacecraft. (I say I tried
to participate; Stardust@home had a test you had to pass before
being presented with the real data, and I could never do well
enough to get in. Though I’d spent several years thinking about
the chemistry that happens on them, it turns out that even with
the training the website provided I couldn’t recognize an inter-
stellar dust grain if one was staring me in the face.) ‘If people will
* For complex sociological reasons, we’re now more likely to be found in the Lamb and Flag.
The Crowd and The Cosmos 63
look at dust grains’, went the logic, ‘surely they will look at gal-
axies’. And look they did. The flood of traffic that prevented me
getting to the website on that summer’s day back at the Royal
Astronomical Society was testament to how powerful the call
for help actually was. Volunteers flocked in from all directions;
an appearance on the BBC Radio 4 flagship Today programme
sent us plenty of traffic from London’s political classes (if the
email addresses I noticed were anything to go by), and an appear-
ance on the Wikipedia home page sent us a collection of people
who were used to rolling up their virtual sleeves and getting
stuck in online. So overwhelming was the demand that the ser-
ver which provided the galaxy images, which had been serving
astronomers happily for a year or two from its home in Alex
Szalay’s lab at Johns Hopkins University, buckled under the
strain.
That could have been it for the project, but to my immense
relief and overwhelming gratitude the team in Baltimore took
pity on us and got the server back online. Soon, more than
70,000 classifications were flowing into the Galaxy Zoo data-
base every hour. Better than that, it was clear pretty quickly that
the classifications were good, probably close to if not better than
Kevin’s. But sorting out exactly how good they were would take
some effort.*
One dark evening a day or two after the Royal Oak discussion
with Kevin, I was sitting at the bar of another historic Oxford
* This book isn’t a history of Galaxy Zoo, nor does it focus on web development. But I’d be remiss if the names of Phil Murray, the original designer, Dan Andreescu, the original developer, and Jan Vandenberg, the sysadmin at the Johns Hopkins University who saved the day on our launch didn’t appear somewhere.
Nor was launch day the only time I had cause to be grateful to Johns Hopkins; a few weeks after launch, we discovered a bug in the code which meant that classifications were being wrongly recorded. Luckily, the problem was interesting enough to be worth Alex’s time, and he was able to straighten everything out.
64 The Crowd and The Cosmos
pub, the Eagle and Child, with Kate Land, my officemate.* A bril-
liant cosmologist I’d long since forgiven for beating me to several
scholarships, Kate was best known as the discoverer of astrono-
my’s ‘axis of evil’, an alignment of features in light emitted just
400,000 years after the Big Bang (the cosmic microwave back-
ground) that just shouldn’t have been there. Before leaving
astronomy for a hedge fund, Kate managed to publish a paper
using newer data that made the axis mostly vanish, but on this
particular evening she was exercised by a different problem.
Staring into her glass of Zinfandel blush, she told me about a
paper which had just appeared on the arXiv preprint server.
While astronomers still publish papers in traditional journals,
the main way they’re shared is via this thirty-year-old website.
In some fields, especially cosmology, papers are published on
arXiv for comment even before they are submitted to a journal; it
makes for more rapid communication and allows ideas to be
bandied about long before they are ready for more formal review.
This particular paper, which had crossed Kate’s desk because it
mentioned her axis of evil, was by Michael Longo. Longo, an
emeritus professor from Michigan, was a distinguished particle
physicist who had recently become interested in astrophysics,
and specifically in the question his paper set out to answer: ‘Does
the Universe have a handedness?’
To answer this apparently obscure question, he’d looked at a
few thousand galaxies in Sloan, selecting the spirals and record-
ing whether they appeared to be rotating clockwise or anticlock-
wise. (The direction of the arms tells you, in most cases, which
way the galaxy is turning; they drag behind the direction of rota-
tion.) He found, surprisingly, that there were more anticlockwise
* You might conclude from this part of the tale that pubs are important in British astronomy. You might think that. I couldn’t possibly comment.
The Crowd and The Cosmos 65
than clockwise spirals, a result crazy enough to scare cosmolo-
gists (Plate 5).
If this result is real, it means two things, both of them dramatic
body blows against the modern cosmological consensus. First, it
suggests that some force is capable of organizing galaxies scat-
tered right across the enormous volume covered by the Sloan
survey; Sloan, remember, covers a quarter of the sky. Second,
it violates the nearly sacred rule known as the cosmological
principle—the idea that any large-scale observation of the
Universe should not depend on your position within it—an alien
astronomer looking on the same set of galaxies from the other
side would see the measurement reversed, so that more clock-
wise than anticlockwise galaxies appear. Messing with the
cosmological principle is bad news; a violation of it means that
we can’t trust our own view of the Universe.
There was more to it, too. Longo looked for an axis of sym-
metry in the data, a way of splitting the Universe in two such that
(most of) the anticlockwise galaxies were on one side and (most
of) the clockwise galaxies were on the other side. In the case of a
very strong effect, you might be able to literally divide clockwise
from anticlockwise galaxies with a single axis, but even Longo
wasn’t claiming our Universe was like that. What he had found
was that if you took the line that was closest to that ideal case—
the line that did the best possible job of dividing those galaxies he classified as clockwise from those he recorded as anticlockwise,
then it aligned almost perfectly with Kate’s axis of evil.
As the axis of evil that Kate had found was in the cosmic
microwave background, it was a feature of the Universe in its
very early days. To find that such a feature persisted in the popu-
lation of galaxies we see around us more than 13 billion years
later suggested that we didn’t really understand galaxy formation
at all. The growth of everything we see would have had to pre-
66 The Crowd and The Cosmos
serve this Universe-scale feature across the aeons, and there’s no
good explanation for how that could happen. If you believed this
paper, then modern cosmology and astrophysics were about to
fall apart.
I should probably emphasize that it wasn’t despair for the state
of modern cosmology that drove Kate and me to the Eagle and
Child that night. For one thing, we simply didn’t believe the
result. Longo simply hadn’t classified enough galaxies, it seemed,
to be able to make such claims, any more than you could toss a
coin twice and conclude on the basis of just those flips that it had
heads on both sides. More spiral galaxies, and hence more clas-
sifications, were needed. A couple of extra buttons were easy to
add to the Galaxy Zoo interface asking volunteers to record the
direction of rota
tion of spiral galaxies, and we could test Longo’s
challenge to conventional cosmology.
Actually, because it is such a clean measurement, checking for
any rotational conspiracy turned out to be an excellent way of
testing the Galaxy Zoo classifications. With so many people tak-
ing part, we were able to have many people look at each image. In
turn, this meant that for each image we didn’t just have a classifi-
cation, but also some idea about how confident we should be.
There is a world of difference, as it turns out, between a galaxy
which ten out of ten people agree is spiral and one where only six
out of ten click the spiral button. After cleaning up the data,
removing the classifications of the very small number of people
who seemed to have clicked randomly, we were left with ‘clean’
samples of hundreds of thousands of both spirals and ellipticals.
The former was the perfect set to test Longo’s claims, which
we were sure were nonsense. You can imagine, therefore, the
confusion in the office when it became clear that, despite hav-
ing a hundred times more spirals available, Galaxy Zoo also
had an excess of anticlockwise galaxies. Was this evidence for a
The Crowd and The Cosmos 67
universal magnetic field? Was the Universe small and shaped
like a doughnut, as one theorist who shall remain anonymous
suggested?*
As it turned out, probably not. Before we rushed to publish
our Universe-shaking paper, we took the precaution of flipping
some of the images in Galaxy Zoo so that people were suddenly
classifying mirror images of the galaxies. Had the Universe really
preferred anticlockwise galaxies, for whatever reason, then we
would have seen a flood of clockwise classifications of these mir-
rored systems. No such flood occurred. In fact, we still recorded
an excess of anticlockwise galaxies. The fault, it seemed, was not
in the Universe but in ourselves. Something in the way that the
human eye and brain process images makes it easier to see an
anticlockwise than a clockwise spiral. It’s not that people are see-
ing a clockwise spiral on Galaxy Zoo and mistaking it for an anti-
clockwise one, but rather that we just miss the spiral arms in a
small proportion of clockwise systems. That slight bias in our
perceptions, added up over the classifications received for thou-
The Crowd and the Cosmos: Adventures in the Zooniverse Page 9