Fake? AGN Galaxies!
Hidden deep inside the center of most massive galaxies is a central super-massive black hole. However, a in a few percent of galaxies, the black hole is growing in size by accretion of matter. These galaxies are called Active Galactic Nuclei (AGN) . By studying these galaxies we can piece together the picture of how galaxies assemble themselves, growing as stars form and their central super-massive black holes accrete matter. In the local universe , we used Galaxy Zoo’s classifications of AGN host galaxies for a study that revealed “The fundamentally different co-evolution of super-massive black holes and their early- and late-type host galaxies”.
But to really understand how most galaxies were built up, we need to look farther back in time, out to where most galaxies in the present day universe were growing most of their mass. Because the light from distant galaxies takes so long to reach us, in Hubble Zoo we are able to see galaxies as they were in the distant past. (In Hubble Zoo most of the galaxies we see are images from ~ 5-7 billion years in the past !). However, as you may have noticed, the Hubble Zoo galaxies appear much smaller. Because the light emitted near an actively growing black hole can be comparable to the light that we see in from an entire galaxy, in Hubble Zoo we have to be careful when classifying galaxies with these luminous centers. To understand what we’re seeing, a team of Galaxy Zoo Scientists have created images, artificially adding the luminous centers of AGN to some of the Hubble Zoo galaxies. These Fake AGN will allow us to determine if how accurately we can classify the host galaxies of the actual AGN in the Hubble Images.
Personally I find the experiment very, err, distasteful. The fakes are often too obvious. There is no classification for AGN, only later in the forum, at which time a poster should have been looking at the available reference material, and the lie is exposed. So who are you fooling, the newbie? Congratulations! I think it’s a shabby way to treat your volunteers.
I agree with Paul. We’re trying to help out here with some boring science work, we’re not stuped guinea-pigs, clicking away on fake pictures.
It seems not only you fake images but also you fake the scientific process of the revised photometry of the eighth SDSS data release:
You went out of control, by not following any correlation comparison done in SDSS3:
quoting the paper “We quantified the accuracy of bright galaxy photome-
try by adding 1300 artificial galaxies at random positions
to SDSS imaging frames, processing them with both the
old (DR7) and new (DR8) versions of photo, and com-
paring the results with the true input values. The sim-
ulated galaxies, which have S´ersic radial profiles with a
range of inclinations and S´ersic indices, follow the ob-
served correlation between apparent magnitude and an-
gular size seen for real galaxies”
It’s obvious that this check was not in use for GZ.
I wish we had actually been given a little science behind this and some preparation as we were with the flipped images three years ago. At that time there was an obvious case to make, not to mention the “ACW/CW” buttons. This explanation makes no sense to me whatsoever and I’m already getting complaints about it – folks, no good complaining to me, I didn’t even know it was going to happen!
(After a great deal of confused thought) Carie, we need to know what you actually want us to DO. Do you want us to try and find out? Do you want us to post them on the forum? Or do you just want us to classify as normal and see if having an apparent AGN there makes a difference in classifying? If we know what this is for and what we’re doing, we’d feel our usual selves – delighted to be part of a team, and seeing if this is another exciting thing to reveal about the way our brains work. At the moment, I’m just extremely confused.
You have disapointed me. I have only been on here a week or so and was excited to think I had found not one but 3 Quasars. Now it looks like they are nothing but these stupid fakes. THANKS A LOT. If you whant to keep people like me on this site treat us like we know what were doing. even if were still learning the ins and out of it all.
Wouldn’t it make more sense to put a controlled set of actual Hubble-AGN pictures in the mix and see how this set gets classified?
It hits like a rod on a pig! I don´t what the meaning of this is, as neither do you. As a complete amateur I classify galaxies. In those pages I even can´t choose whether or not there is an AGN. If and only if I decide to look closer at the thing I will explore. And then I only can see if there is an AGN if there is any info in for instance Simbad or NED.
So what´s the bloody use of this trick? There will be experts around who can tell an AGN maybe by its spectrum. The rest of us can´t!
They also can draw birds in the middle of a galaxy and ask us if we see them!
It´s useless, stupid and offensive.
People who know this now, could be tempted to consider any picture, of which they think it´s tempered with, as an artefact and dismiss it.
Was it it done, for example in the case of Reference 90045113?
(There is a remark on the examination page, below the Survey reference: “This image has been modified: Click here to view the original image”)
posted in GZ forum Hubble: QSO’s
« Reply #167 on: December 31, 2010, 03:50:23 pm »
of a unique survey object 90040755 giving two new ID siblings
teasers from zookeepers !
same hoax as Zutopian 90045113, with rotated image and no change in reading parameters (mag. , Kron, z)
Yesterday a Newbie posted following on the forum:
I got this one today. I’m assuming its an anomaly with the lens or from image modification, but thought I would post since it looks like an eye. lol ID: AHZ40007rj Survey reference: 90045113
I wanted to respond to your concerns here, most of which are fair enough. Firstly, you can always identify those images we’ve changed via their examine page, and to help you out all the altered images and only they begin AHZ7….. This will hopefully help you avoid wasting time talking about exciting discoveries that turn about to be fake.
The reason these images are being added is because we think that our results are being skewed by the presence of AGN in the distant sample. In some of the Hubble images, the AGN can be as bright or even brighter than its host galaxy, and that clearly alters the classification made. If we want to write papers about the interaction between the AGN and the galaxy, we need to be able to measure the size of this effect, and that’s what this latest experiment is designed to do. In other words, we’re not testing whether you can spot AGN or not (although watching the discussions on the forum suggests that many of you can), but rather to see if the presence of AGN changes our conclusions about galaxy shapes. So just classify the galaxies as you would have done before.
I want to emphasise that this isn’t because of anything you’re doing wrong, but this sort of sanity check is needed in any classification project, whether it uses citizen scientists or computers.
I hope that helps, and thanks for the feedback. Sorry for not announcing this properly – it was a last minute addition to the latest release, and I’m afraid it was just an oversight on our part.
P.S. The question about why we can’t use known AGN is a great one – the problem is that we don’t have a good catalogue of AGN throughout our sample, and so the problem would still remain for most galaxies.
[quote author=paulrogers link=topic=271742.msg520882#msg520882 date=1294856116]
[quote author=zookeeperChris link=topic=271742.msg520823#msg520823 date=1294844906]
The only way to deal with the situation is to measure the sensitivity of our classifications to this effect, and the only way to do that is to carefully add some AGN to existing images and then record the classifications. I want to emphasise that this isn’t because your classifications aren’t any good, but rather something we’d want to do with any classification method. We have a very strong track record of pinning down biases in our data set (for those who’ve been around a while, think about the black and white images in the original Zoo) and we want to do a good job here, making sure that the results you’re providing lead to accurate, exciting science.
Of course, the downside is that those of you who like to follow things up here might get confused. …
[color=yellow]No, Sir! That was [u]not[/u] the only way! As a matter of fact it betrayed the trust of your volunteers. It was just about the worst way possible you might have chosen! I said it was a shabby way to treat us, and that’s only my polite version. I haven’t run into anything like this since the experiments we were forced to participate in during freshman psychology pseudo-science class. But at least there we knew walking in that we were participants in some cockamamie experiment. (And even that didn’t help Ted Kozinski in one of the psychology experiments he was subjected to, with unimagined later consequences.) After you betray someone’s trust, how long does it take to get it back?
I presume you remember the “Galaxy Wars”. You compared our barred & spiraled galaxy classifications with others that were similarly classified. With the GZH images you changed the classification scheme to add a clumpy pathway. If AGN’s were of interest, you might have added an AGN option to the Oddities, and then ran “Galaxy Wars” comparisons to known and/or suspected AGN’s. That would have accomplished your goals, but not betrayed the confidence of your volunteers.
You might have even setup a part of the GZ site where, like the common police suspect line-up, you showed us a half dozen or so suspected and/or confirmed AGN’s, and asked us to pick out the “perp”. So this shabby experiment was [u]not[/u] the only way to evaluate our ability to pick out AGN’s. It was just a lack of imagination, within a decision process that did [u]not[/u] consider the trust relationship you made with your volunteers–your claim you were showing us real images of natural object to classify.
It’s not even good science! In the first place, after seeing just a few of them it was clear, at least to me, somebody was messing around with us. The doctored parts of images were obviously unnaturally sharp, circular, and plagued by square overlays with different tints. But let’s be clear about the “scientific” question this scheme asks: If we slip in some doctored images, the classifications will be rubbish even if we [u]had[/u] a classification for AGN’s which we don’t, but what will people say about them on the Forum? You’re asking about the doctored images! That has no relationship to natural images. As someone has suggested, you might have drawn little birdies on them to see if we could tell they aren’t natural.
I remain undecided if continued participation in GZ is acceptable. I’m not here to play games.[/color]
I’ve replied on the forum to this message, but let me post it here too :
Hi Paul and others
Thanks for the further response; let me start by saying that I’m sorry we didn’t discuss this in advance – we should have done, but got carried away adding things at the last minute and we shouldn’t have done that.
That said, I think that your comments miss the point of what we’re trying to do. The goal is absolutely not to assess whether Galaxy Zoo volunteers can identify AGN visually – if we wanted to do that, then the procedure you set out (involving running a trial, or adding a button) would make perfect sense. Instead, as mentioned here the goal is to quantify an effect we think is showing up in our classifications of galaxy shape – the core mission of Galaxy Zoo – which seem to be affected by the presence of AGN.
The new images are a small fraction of the total, so the vast majority of galaxies you see are untouched. The fake ones were generated according to the best code that is available, but as ever with these things, some don’t look at all realistic. Many of the ones I’ve seen look like normal galaxies. We’ll have to account for that in the data analysis. We’ll also review the results in a few days, and if the fake galaxies aren’t producing useful results, they’ll be gone.
I hope that helps. I’m sorry we didn’t say this first – we should have done – but I”m confident that we’re doing what we need to produce science from your classifications.
I can understand what Chris is saying and would point out that in Planethunters website there are regular sets of fake date inserted as a test so that the Team can better assess the types of transits missed by the volunteers, as a test of the completeness of the analysis of the data. But there is a warning about this in the tutorial and after the data is classified the volunteer is told if the data was test data.
Wow I’m so sorry that this has been so concerning to so many!
As Chris explained, we are interested in trying to quantify the influence of a central point source on a galaxy’s morphological classification. We want to understand the morphologies of AGN host galaxies, but were concerned that if we looked at the results for the few galaxies that happened to host AGN (there are maybe a few thousand out of the total sample of galaxies from HST), we might not be able to compare the results to similar galaxies without a central point source. We thought a unique way to test this would be to add a few made up images, which would have one of the regular galaxies, plus the point source seen in an AGN. We could then determine whether the bright point at the center changed how the galaxies themselves were classified. With that knowledge we can then use the results for the actual AGN host galaxies with confidence.
This type of data-verification can be very important in interpreting any scientific result. So please do not thing that we are trying to waste your time or that the results aren’t valuable to real science.
As Chris pointed out, if you can easily see the fakeness of the added AGN, then we’ve done our job poorly. Our hope is that you would classify these test images the same way that you classify all of the others.
This is very disappointing. I hope in time I can regain the enthusiasm felt when joining a few years ago – almost at Day One.
This should not occur in science, especially in a field where accuracy and clear communication are of the utmost importance.
This now raises doubts about other segments of GZ. I feel duped. I have not spent many hundreds of hours to see them treated like this.
This page has some bad html errors in it.
Several of the comments are not fully viewable, and even this area of the page has overlapping text on it;
here is a link to 100 COSMOS galaxies which “do” have real point source AGN embedded in a disk, I think these were the galaxies that were trying to be simulated in this experiment
keep in mind that these are all real images and have not been manipulated in any way.
Let me know when we can examine real life pictures without fakes. That is what we use some of our spare time in doing to help you and because we enjoy seeing what is out there in space, not what has been constructed artificially.
If you want to experiment to see the effect of AGNs do it on a separate site where it doesn’t affect those of us wanting to observe things in real life.
In the meantime- bye.
Insertion of synthetics is nothing new in this type of work. Some of you are making waaay too much out of this. If anything, it will help to improve the overall quality of the end results.
Because someting is nothing new does that make it right?
I’ve been trying to classify galazies for some time now — I know a 1,000 doesn’t seem like much put to a complete amateur like mde it’s a lot. Now I read all this AGN junk and I wonder if my time would be better spent at other volinteer locatioins, bubbles ands such. I’ve been trying to do the right thing, but now I feel at a loss. CRAP..
As a new-be all this, I feel dishearten that the scientific community would be so underhanded. It seem that they should have been more up front in the beginning. As mentioned before, perhaps it might have been better to use Known AGN as a check points. Or you should have rewrite the program for classifying these AGNs to better single out what you are looking for.
I have thought about this and can see a soret of reason for the experiment – ie to calibrate the “accuracy” of classification – but do think that we ought to have been advised beforehand. If we couldn’t be told which was which (fairly obviously) then we should have been given a heads-up in some other way.
Indeed why stop at AGNs? It might be very interesting to see how accurate other classifications are… NO, let’s not go there!! But perhaps the current menu categories are too limited to allow us to differentiate well in any case and that on its own might well be sufficient to screw up any attempt at measurement of accuracy.
“Many faults are detectable only indirectly as a result of performance disorders that manifest as anomalies in monitored system or sensor data Anomaly detection therefore is often the primary means of providing early indications of faults As with any other kind of detector one seeks full coverage of the detection space with the anomaly detector being used Even if coverage of a particular anomaly
detector falls short of detectors can be composed to effect broader coverage once their respective sweet spots and blind regions are known….” – From Anomaly Detection in Embedded Systems by Roy A Maxion and Kymie MC Tan
I cite this because it provides the necessary information for scientific precaution. Without assurance of your data one is left open to losing its validity in full. To achieve this, as in any system, one must anticipate or retroactively define that which appears to be adverse to your data set.
This is merely a way of saying that they saw, or would like to prepare for, is the possibility that the AGN were causing the galaxies to be misclassified in some fashion. If we, the human detector, continue to label the galaxies and the AGN posits no influence on the results then the outcome is the same for you as a reviewer as if it had. There is no underhandedness in its application. There is no discourtesy to the reviewer whatsoever.
After all, Science without data is moot.
I therefore thank them for the opportunity to be a part of the entire process and I laud the lengths that the GZ team has gone to in assuring data consistency. Remember that the entire astronomical community can use the data that is being gathered here. It is certainly an interesting diversion to those participating but, that does not mean that the process of personal and peer review should be removed.
Please, join me in thanking the team for all that they’ve done so far and for all the projects to come.
I think that most people who complain about synthetic data are completely missing the point. The point is not to find AGNs, nor to be excited about them. We’ve volunteered to classify galaxies. We should keep doing precisely that. The project organizers want to find out whether our classifications will be altered by presence of AGNs. End of story, right there. Classifying such synthetic images provides the project with the data it needs. It’s no less valuable than classifying unmodified images. I feel at loss as to why some people get so emotional about it. Your provide the project with valuable data, what the heck more do you want?! Somehow working with synthetic data makes your work less valuable in your eyes? That’s at best silly, IMHO.
I participate to see what the universe looks like.
I want to see real, raw images. (Thank you, GZ, for the privilege.)
In exchange, I carefully and thoughtfully classify what I see.
If valuable data is generated from my looking and classifying, that’s great, but it’s not why I volunteer, or why I am emotionally invested.
There are ways to calibrate the human detector that do not adulterate the image stream. However, add any fake images and all are suspect, in the same way that adding a teaspoon of sewage to a bucket of clean water makes a bucket of sewage. That is why some react emotionally.