The Crowd and the Cosmos: Adventures in the Zooniverse

by Lintott, Chris

  quite-irony searching for the remains of Toby the pig, the first pig

  to visit the Antarctic twice.*

  But there was no escaping the fact that one of the things that

  might be threatening the well-being of the charismatic, if stink-

  ing, penguins was our own presence alongside them. Part of the

  point of our mission was to try and understand human impact

  * Toby had sailed south with a Uruguayan expedition, who sold him to a French ship on the way home. Toby’s second visit to the Antarctic was thus part of an exhibition organized by the French explorer, Jean-Baptiste Charcot, whose crew were one of the first to deliberately overwinter on the Peninsular. Charcot is remembered now for his superior planning, which allowed for a modicum of comfort which he believed was important for the crew’s well-being. There’s a photo of him breakfasting—at a carefully laid table—on the ice, butler and champagne on hand, and the ship’s crew each morning enjoyed the daily paper from Paris—just distributed precisely a year late.

  Too Many Penguins 145

  on the landscape. Tourism was in evidence whenever we were

  there, for obvious reasons, and science—either by our own

  efforts or those of the research bases we visited—also had a

  visible impact. Two hidden threats need to be taken into account,

  though.

  The first is the one you’re thinking about. It’s hard to mention

  ‘ice’, now, without immediately thinking of the melting induced

  by climate change that is afflicting the world’s ice caps and gla-

  ciers. The last few summers have been brutal in the polar regions,

  with the Arctic ice cap frequently so far from its usual extent for

  the time of year that it’s obvious even to most casual observer

  looking at satellite photographs. The Antarctic Peninsula, too, is

  warming, although the complexities of the flow of water around

  here have complicated the picture somewhat. The other threat

  lies offshore, and comes as something of a shock to those who

  have come here precisely because it is an unspoilt wilderness.

  ‘Fishing?’, they say, as Tom explains his idea that the penguin

  colonies we are visiting are suffering from the effects of over-

  zealous human fisheries. The colonies look robust enough, but

  data from the camera traps show that the numbers of one species

  in particular is declining. It is the species of penguin that depends most of all on krill, the diverse and nutritious tiny crustaceans

  that swarm throughout the Southern Ocean.

  Krill have been harvested seriously in these waters for the last

  couple of decades. That sounds surprising, because you’ve never

  ordered a krill burger, but in addition to food for fish farms the

  bountiful harvest produces gallons of sub-standard cod-liver

  and fish oil. Buy a generic supplement from your pharmacy, and

  without reading very carefully you’ll be competing with the

  Antarctic penguins for their primary foodstuff.

  There is a lot of krill to go round. One source, the Commission

  for the Conservation of Antarctic Marine Living Resources,

  146 Too Many Penguins

  points out that the mass of all the Antarctic krill in the world

  outweighs us humans. Unlike human flesh, though, most of the

  biomass which exists in the form of krill is eaten each year and

  then replaced. This rapid turnover makes it even harder to believe

  that human intervention could make much of a difference, but it

  actually makes the problem worse.

  As the krill disappear, harvested for our consumption, those

  penguin species which aren’t able to adjust to find other food-

  stuffs suffer. Tom’s camera network has already picked up a dif-

  ference between the resilient gentoo penguins, which have been

  able to switch from krill to other food, and populations of less

  versatile chinstraps.

  Cameras can only tell you so much, which is why I found

  myself reaching tentatively into a fridge full of a summer’s worth

  of guano samples. Tom and team have long been in the habit of

  collecting penguin poop from most of their sites, hoping to

  marry serious lab analysis with camera data. The cameras tell us

  how the colony is doing, and the lab work will tell us how at least

  some individual penguins are doing, bearing information on

  diet, on health, and on any infections the penguins carry.

  First though, we had to get the samples back to Oxford, and as

  the ship sailed around the peninsula there was a rare opportun-

  ity. One of our stops was Port Lockroy, a British base more than

  a hundred years old and home to the only functioning Post Office

  on the continent. The volunteer staff from the charity UK

  Antarctic Heritage, who run the base, act as curators, mainten-

  ance staff, wildlife recorders, and more during their stint there,

  but during the tourist season they also run a thriving gift shop.

  Must-have items include an Antarctic tartan tie, but what every-

  one really wants is to send a postcard back to friends at home.

  There is, therefore, a working Post Office, though you have to

  wait for the next ship before mail leaves the base. If Tom and I

  Too Many Penguins 147

  could package the samples that had been languishing in his

  fridge, they could be dispatched directly from the base to Oxford.

  That meant diluting each sample with stabilizing chemical,

  which meant donning latex gloves and squeezing into the tiny

  bathroom attached to Tom’s cabin.

  I’ve already mentioned the unappetizing smell of the penguin

  colonies, and the stench of their droppings within the small cabin

  was unforgettable. It got into one’s nostrils instantly, onto

  clothes, and, I feared, into my skin. After a while on the produc-

  tion line, handling tubes Tom passed to me while pointing them

  as far from my nose as possible, I pleaded for a break and headed

  out to get coffee.

  I got back to the cabin to notice one of the ship’s efficient crew

  fiddling with pipes in the corridor outside. I was just explaining

  this to Tom when a knock on the door revealed the ship’s purser,

  in pursuit of an unearthly smell that was disturbing those paying

  passengers who were trying to have the holiday of a lifetime

  around our scientific expedition. Apparently he wanted to see

  Tom’s bathroom, where he feared the smell now working its way

  through the ship’s air conditioning originated.

  I don’t really know how to describe what we looked like, two

  unshaven researchers with blue hospital gloves, inane grins, and

  the realization of what we’d done slowly showing up in our

  expressions. Somehow, between our shock and his confusion,

  we agreed he should come back and inspect the plumbing later

  and got on with the job. I’ve never worked faster in my life, and

  somehow we got the samples safely packaged before a more for-

  cible intervention arrived.

  Reflecting on the morning’s events in the ship’s bar later, won-

  dering if people were avoiding me because of a lingering stench,

  I realized just how close to astrophysics Tom’s research was.

  Obviously, I’m rarely called in to deal with galaxy excreta, at least

  148 Too Many P
enguins

  directly, but Tom (with my inept help) had become, while still

  doing science, a professional and an expert in data collection. His

  command of the moving network of ships and people, and the

  resulting spread of cameras and data, reminded me of the unsung

  heroes of the Sloan Digital Sky Survey, the engineers and astron-

  omers who spend huge amounts of time gathering the data on

  which the rest of us depend.

  The fact that we’d rushed to treat the samples before the smell

  contaminated a cruise ship won’t ever be mentioned in a scien-

  tific paper. The careful day-to-day diplomacy that ensured that a

  ship employed on quite some other purpose delivered the team

  to each of their cameras is reflected in an unbroken data series,

  but won’t ever be commented on in formal publication. And the

  fact that Tom was able to leap up that hill, and I wasn’t, won’t be

  recorded anywhere but in the pages of this book.

  Similarly, without people who understand how to make the

  telescope perform, to keep the camera operating at peak per-

  formance, all the science that uses this data is measurably poorer.

  Without people who are really good at being Sloan’s ‘cold observer’*

  my measurements of galaxy properties would be less accurate.

  It’s rare, in science, to pay much attention to these hidden parts

  of the process, which since the nineteenth century have become

  increasingly professionalized and, for most of us as we’ve entered

  the digital age, increasingly remote from our day-to-day lives.

  When you go to that much effort—whether in the surpris-

  ingly chilly New Mexico night or the much more predictable

  cold of the Antarctic—it’s important to make the best of all of

  the data you can obtain. For Tom and his team that means shar-

  ing the images his camera network takes, and which they go to

  * This is a real job title; the cold observer is out with the telescope while the

  ‘warm observer’ is inside with the electronics.

  Too Many Penguins 149

  such lengths to return to Oxford, with the entire world, via a

  Zooniverse project called Penguin Watch.

  Penguin Watch is perhaps the simplest of all our projects, so

  straightforward that I know 5-year-old children have taken part.

  Presented with an image from the cameras, all you have to do is

  count and then click on the penguins. This information seems

  almost banal in the context of a single image, but over time we

  learn how colonies shrink and expand, how much time individ-

  ual penguins are spending out at sea, and even what their feeding

  behaviour is like. These details can then be compared to weather

  and climate records, to changes in fishing permits and protected

  areas, and to visitor numbers to get a sense of what’s really hap-

  pening.

  For that to work, the data must be accurate. An individual pen-

  guin counter can make a mistake, and so the key is to combine

  everyone’s penguin counts to produce a consensus. All of our

  citizen science projects depend on multiple people looking at the

  same data, but there are a few maddening quirks about penguins

  that make it more difficult. First—and you might have to take my

  word for it—there are a lot of them. We, of course, are after an

  accurate, scientific count, but some of the cameras capture hun-

  dreds of penguins in a single image. That’s not ideal, obviously,

  but the cameras are set up once and then left for the year and so

  when things change, so does their view.

  The problem with having too many penguins is that people

  baulk at counting them. Being presented with an image of hun-

  dreds of the critters is, for many people, more annoying than

  interesting. Our designers and developers know this, and so

  Penguin Watch reassures you that after your penguin count

  reaches thirty it’s OK to move on. (It turns out there are people

  out there, many of them attracted to Penguin Watch, who deeply

  resent this message. They are people who like order, who like

  150 Too Many Penguins

  completing a challenge no matter how many penguins it involves.

  And so we learn once again that people are complicated.) For

  these busy images, therefore, few people cover every penguin.

  Some click the front row, others a cluster near the back, and so

  on and so forth.

  What we’re left with once many people have seen each image

  is a mosaic of things that at least one person thought was a pen-

  guin. What the team need is a list of likely penguins, ideally with

  some sense of how likely each possibility is to be real. This

  requires some careful data handling, but the basic idea is sim-

  ple—if two people mark a penguin in roughly the same place,

  then each marking counts as a ‘vote’ that there’s a penguin there.

  The nice thing about this is that the researcher working on the

  data, has only got to make two decisions. The first concerns how

  close together two markings have to be for them to be in ‘roughly’

  the same place. That can be found by trial and error. If you have a

  few images that experts have gone through, then you can just

  adjust the parameter until you get results that look pretty good.

  If you fail, then you need more people to look at each image so

  that you get more data. This is essentially what we do when test-

  ing a project.

  That’s the easy part. If you want to make it complicated, there

  are reams of computer science papers that deal with this sort of

  clustering problem, and plenty of researchers who will make it

  more complex for you. The degree of proximity required to have

  the algorithm decide that two marks refer to the same penguin

  need not be a constant, for example, but could depend on how far

  the markings are from the camera, or the time of day, or how

  many other penguins are in the image, or a host of other varia-

  bles.

  In general, though, because Penguin Watch volunteers are

  pretty good, there’s not much need for a complex solution to the

  Too Many Penguins 151

  problem. The second decision you have to make is much more

  difficult. How many people have to have marked the same spot

  for us to conclude there is a penguin there?

  We want accurate data, so the temptation is to say that we

  need lots of people for each penguin. That’ll produce a set of

  places where we’re really, really sure a penguin is—but we’ll miss

  most of them. If we want complete data—if we want to catch

  every penguin going, even the ones at the back disguising them-

  selves effectively as rocks—then we need to relax, make the algo-

  rithm less picky, and include places where only a few people say

  there’s a penguin in the final list.

  That will decrease the accuracy, so there’s a trade-off to be

  made between accuracy and completeness; in a problem like

  this, you have to choose which you care more about. This turns

  out to be a general feature of this type of problem, and for differ-

  ent scientific problems you might pick different combinati
ons.

  If you wanted a few excellent images of penguins, for example,

  you might go for high accuracy and low completeness. If your

  research called for an upper limit on the number of surviving

  penguins, then completeness becomes more important than

  accuracy.

  But this isn’t the end of the story. We haven’t used all of the

  information we have to hand, and in trying to squeeze more

  from the data, in order to do justice to all the hard work that went

  into collecting it, things get interesting. So far, we’ve treated everyone’s classifications as being of equal value, but it must be true

  that some will be better, or more diligent, at penguin counting

  than others. For example, we know a large proportion of people

  who take part in Penguin Watch in particular have ‘Mom’, ‘Mum’,

  or ‘Dad’ in their usernames; it seems reasonable to assume that

  these represent households where participating in science by

  152 Too Many Penguins

  counting penguins is a family activity, involving children too

  young to have their own account.

  It’s possible too that these kids are less good than their older

  counterparts at the task, and thus we should pay less attention to

  them. On the other hand, I could easily believe that the combin-

  ation of growing up as digital natives, smaller hands, and the

  insatiable desire for repetition that characterizes most 5 year olds

  of my acquaintance might make them killer classifiers. We don’t

  have to decide in advance; with only a small number of images

  labelled by experts, we can find the people who are best at pen-

  guin counting, and pay attention to them.

  It’s a simple and obvious idea. The ability of websites to under-

  stand us based solely on our interactions with them is one of the

  things that drives the digital world. If Google can discover just by

  watching the information I happen to give it where I work (after

  a brief, embarrassing period during which it insisted the Lamb

  and Flag pub was my office), and if Facebook can serve up the

  memory of the long-forgotten party I want to see each morning,

  then surely discovering whether I’m any good at penguin count-

  ing by asking me to count penguins is the digital equivalent of

  child’s play.

  Once we start thinking like this, there’s much we can do to