10/31: Calling Bullshit

[smiklin]

Comment below with questions or thoughts about the reading for this week's workshop.

Please make your comments by Wednesday 11:59 PM, and upvote at least five of your peers' comments on Thursday prior to the workshop. You need to use 'thumbs-up' for your reactions to count towards 'top comments,' but you can use other emojis on top of the thumbs up.

[SiyuanPengMike]

Thanks a lot for the interesting topic. This might be the most relaxing issue we have ever had for the workshop. As for the views you showed in your papers and materials, I agree that with the emergence of the we-media, the quantity of bullshit has sharply increased. Thus, I'm wondering what kind of constraints we could make to moderate this kind of situation? Similar to the finance domain, shall we also set up qualification examination for those journalists to make sure they are indeed professional and would not produce bullshit, instead of improving readers' ability to judge bullshit?

[vinsonyz]

Thank you for your presentation! We know that in the era of big data, there are many misleading information. People can make some irrational judgement based on biased information they received. When someone take the correlations as the causal effects, the messages become bullshit. Then my question is how can we avoid producing and receiving bullshit?

[sunying2018]

Thanks for your presentation. Just as you mentioned the word is awash in bullshit, politics, science, education, culture and so on. But how could we spot bullshit and check on the plausibility in these different fields which may need different prior knowledge to identify bullshit. Is there any analysis framework or tests to facilitate it?

[ChivLiu]

Thank you for presenting such an interesting topic! Many pieces of research now tend to use tailor-made data and use that data to fit into the algorithm the researchers built. Those kinds of research usually try to prove a general conclusion with data collected from a test group. For example, a researcher could use sleeping data from students at several colleges to support a hypothesis assumed upon all young adults. Also, like the data used to test the sexual orientation learning model, the data was all images from dating sites, and normally people on those sites would like to post pictures that help to impress their friends. Therefore, the model might not be able to detect a person's sexual orientation based on random daily pictures. I wonder that if a paper determined a general-applied theory only from a model built on incomplete data, would that paper have a higher possibility to be called bullshit than the others?

[Yilun0221]

Thanks for sharing! I think the study of using algorithms to predict human sexual orientation is very interesting. But I have a problem: sexual orientation should belong to one's privacy, but according to the algorithm of machine learning, others can know the sexual orientation of a person. Is this not a violation of personal privacy?

[luyingjiang]

Thank you for your presentation. In the sexual orientation case study, the authors proposed two extremely claims but neither was considered particularly likely a priori. Do you think that the authors were aware of this kind of extreme interpretation (i.e. they were aware of the fact that they might be creating bullshit)? But in order to publish a study that was exciting and fresh, they had to be a little over-interpreting. How can we reduce this kind of bullshit?

[chiayunc]

Thank you for the wonderful presentation. My question is related to 'empty words' candidates make during elections, words that are accessible, influential and so empty that it is really hard to call bullshit on. A newly elected mayor in Taiwan won the election by saying these type of empty words. It made him look "relatable" and "down-to-earth". For example, his primary goal for his city is to "make people rich", nothing more substantive, not even slightly. It is so easy for those words to sound good and prosperous, and the fact that those words are so empty makes it very hard to call bullshit on, or for fact-checker to rebut. Would you characterize these as bullshit, and how does it fit into your framework?

[romanticmonkey]

Thank you so much for the presentation! This is a fascinating topic. I first encountered this line of study from SPSP 2018 (I forgot the speaker) where I was introduced to wisdomofchopra.com, this random generator of seemingly profound philosophical quotes. When I try to read into these quotes, they actually start to make sense vaguely, warning me how ridiculous my brain can sometimes be. Other than this funny phenomenon, I'm interested in looking into educational applications. Do you think we should give this bullshit education as a curriculum early on (like in high school) or do you think we should only give it to college graduates? I was wondering if an early establishment of this ability to discern bullshit would be helpful to the student. If given the chance to incorporate it into the school's curriculum, which year of a student's life would you add bullshit education into his or her curriculum?

[huanye]

This is an interesting one. I am also curious about those "second hand" bullshits. Some message may not be born with a bullshit quality, but in the process of their transmission or broadcasting, they become bullshits gradually and thus in this sense, this kind of bullshit can be more like noises. So to reduces those noises, I wonder if we can borrow some ideas from the communication theory to restore the original message either by designing a redundant message producing system, or to increase the "power" of the original message to increase the "signal/noise" ratio in different message producing environments.

[dhruvalb]

Thank you for sharing your work! As I was reading through the material, I was thinking - what makes someone what to intentionally misrepresent truth? Is it because people are rewarded for the narrow purpose of accomplishing what they need to do? That is - advertisers sell, politicians wins votes, media gets views etc. How can we broaden the responsibility of individuals to find and speak the truth, no matter what one has to face professionally or personally? Additionally, how can we combat exhaustion that comes from mistrusting what anyone says and staying vigilant all the time, especially on matters that are not one's expertise?

[fulinguo]

Thanks for your presentation and the interesting topic. People or organizations may face risks when they bullshit, which might be the negative consequences if their bullshit behavior is discovered by the public. I was wondering how we could learn the bullshit phenomenon by conducting this type of cost-benefit analysis. Also, are there any situations when bullshit is helpful? I mean from a moral perspective, bullshit is unpleasant, but from the perspective of utilitarianism, is bullshit necessary or valuable in some cases, like bullshitting for higher or longer-term goals? Thanks!

[skanthan95]

Thank you in advance for your presentation, and for sharing these amusing-yet-informative reads. Bullshit poses a unique problem, given that many people forgo authenticity and intellectual integrity in conversations that are spurred by emotion rather than a shared understanding of facts. What's the best method of responding to it if you encounter it in a debate, and how can we de-incentivize it as a mechanism to save face in this context?

-

[JuneZzj]

With the prevalence of the application of machine learning, people are obsessed with the fascinating power of the computer. Different algorithms indeed bring many important implications in real life. However, I do not see the point of comparing the ability in capturing information and dealing with massive data between human being and computers. The article by Wang and Kosinski tries to tell the fact that computers outperform in the case where the task of detecting information than human brains. I presume in the future there will be more debates about whether human or computers do a better job in a certain field of tasks. My question here was, how to combine the ability of both human beings and computers. Should it be a more important job to do? Since computes are competent in the aspect of detecting and processing information, why we could not make it more productive if we have our own powerful strength?

[TianxinZheng]

This is interesting as well as thought provoking topic. People like to make bullshits just to draw attentions. In academia, researchers are also sometimes obsessed with buzz words and fancy algorithms. In the "machine learning about secual orientation" case study, is there any possibility that the authors intentionally over interpreted their results to make it novel and "interesting" to attract more attention? What do you think would be good rules in academia to help researchers call bullshit?

[Yawei-Li]

Thank you in advance for this interesting and insightful presentation! I did enjoy the readings much. In the era of big data today, we as MACSS students should fully be aware of the side-effects of seemingly informational bullshit. And technologies even contribute to this negative trend of bullshiting. From your perspective, what could we young students do to try to improve our ability to tell bullshit from real data and information? What other measures should individuals and authorities take to prevent and stop the possible harm?

[lyl010]

Thanks for your presentation! I am especially interested in telling a story through visualization, but sometimes good work should be descriptive instead of one hundred percent right, especially when our audience does not care about the details of data. So how can we keep a balance between a fancy graph and a modest data interpretation?

[Yiqing-Zh]

Thank you for your interesting presentation. In the case study "Machine learning about sexual orientation", you mentioned that the authors went too far when they tried to explain their result and make them corresponding to some existed theories. In today's scientific community, a good story like this one can be accepted more easily, especially in some subjects like economics. Do you think that we should impute the phenomenon to the system where research projects are funded, reviewed, published?

[AllisonXiong]

Thanks for sharing you groundbreaking research with us! It has occurred to many of us that there are misleading news and information around us, but never have us theoretically interpret the logic beneath. However, it's way easier to produce and spread rumors that to repute them. An impressive story like the Stanford Prison Study can have strong and long-lasting impact despite all its logical fallacies. However, it may take generations to stop our blind inference of this experiment. Do you think the phenomenon would continue, or it could be mitigate by some deliberate training or some future development? Thanks!

[ydeng117]

Thanks for the presentation! During the course of studying computational social science, I have always had the question about what is the fine line between a reasonable big data study of humanity and a ridiculous discussion of the connection between two far-fetched things. I always think that there will be always something that machine will never calculate and learn about humans. On the other hand, with such a plethora of information that can be collected. A machine could probably learn something that hides behind our unconscious behavior and traits. So, my question is, how should we draw this fine line between bullshit and serious study when we even do not know the maximum potential of the machine?

[ShanglunLi]

Thank you for providing such an interesting paper to read! Since bullshit is vague to some extent, and it can be represented in a misleading way, how can we improve the integrity of it by streamlining? Thank you.

[yalingtsui]

Thank you fo your presenting! That's a really interesting topic, I enjoyed read it a lot. I agreed that "bullshit" is common in reality, but I noticed that sometimes people "prefer" say bullshit even thought they know that. I guess there may be some physical reasons behind that. Additionally, comparing with truth, bullshit is easily spread among the public, it seems like people love dramatic comments, even unconsciously choose bullshit between truth and lie. What do you think about that?

[ellenhsieh]

Thank you for your interesting paper on bullshits! For me, I am wondering how people can distinguish the bullshit and useful information. What is the definition for bullshit here? It seems that this term is vague to some extent just like @ShanglunLi has mentioned. Also, in your study case about doing machine learning on sex orientation, you said that people don't have to equipped with certain professional background knowledge to think critically. However, how can people know that the process in the black box is always accountable and reliable? Would you mind to elaborate on it more and maybe give some more examples for that?

[WMhYang]

Thank you very much for bringing the amazing topic. The principles in visualization really help me find the origin why I sometimes feel confused when reading graphs. My question is related to a point made by Professor James Evans. In academia, the reseaches are becoming more and more difficult to catch up with for people who do not have specific trainings. As you mentioned in On Bullshit, bullshit is inevitable when someone is talking about some topics that he does not realy understand. As a result, people are not familiar with the researches cannot talk about the topics without making bullshits. However, two weeks ago, Professor James Evans present a paper saying that we actually should encourage the combination of different subjects, especially unexpected ones, in order to make new breakthroughs. So how could we deal with bullshits when we try to mix different subjects together?

[mingtao-gao]

Thank you in advance for your presentation! This topic is very fascinating and it speaks out the truth that is happening everywhere now in academia, industry and society. The course on how to identify bullshit or identify someone is calling bullshit sounds very interesting and it correlates so closely to our perspective course that teaches us ways to find out misleading conduction and conclusions made by the researchers. My question is, based on all readings, I understood how to better perceive data and models fed to me everyday, but are there any means that the institution (academia, schools, media, social platforms) should take to prevent nowadays overwhelming bullshit?

[cytwill]

Thank you so much for these interesting topics in data analytics! As Big Data explode, we do need more precaution when presenting or interpreting the data in case making so-called bullshit! I have two questions here regarding the content of your articles:

1. How do you identify such "bullshit" when facing certain unsatisfactory data visualization or interpretation results? In some cases, maybe we feel inconvenient to obtain the information we need from a certain chart or figure, but I think we can not simply say that data presentation is poor because maybe the creator just wants to reveal something of his/her needs instead of ours.

2. When doing data interpretation, should we have some prior expectations for the result? Especially, when the results seem not to be supportive of our theories, shall we modify our theories or shall we change for other data source or analysis techniques to defend the theories?

[bazirou]

Thank you so much for the wonderful presentation. From your perspective, what could we young students do to try to improve our ability to tell bullshit from real data and information?

[caibengbu]

Sorry for late interaction! (Hope this still count as a valid post) I am extremely interested in the machine learning model for sexual orientation. It was echoed in social media and the author was accused by providing tools for sexual orientation prosecution. Ethics has always been a important point in all kinds of studies. I still clearly remembered the horrible news that Chinese “scientist” He trying to created gene modified babies just earlier this year. What do you think is the ethical boundary of science? Do you think that the negative exposure that a research bring to human can be regulated in other ways like legislation?

[Panyw97]

This topic is interesting and instantly caught my eyes as I saw it. It's critical to nowadays data analysis in research. I also concern about the validity of data interpretation since we can't avoid subjectivity while analyzing the visualization or regression results. And actually most of the time I would have different understanding of the presented results with the authors. It's a serious problem in social science study.

[zeyuxu1997]

Thanks for presenting. I have a question about your explanation on "humbug". The example you provide is someone says that he has 20 dollars in his pocket to mislead others to think that's his belief, but that's not actually a lie. I check the definition of "lie" in the dictionary, which is "to say or write sth that you know is not true". In that example, the man has the motive to mislead others, and say something wrong intentionally, so I wonder why it's not a lie. Could you please use some other examples to show what's "short of lying"?

[ZhouXing-19]

Thank you for your presentation. I have fully the same feeling when I was doing internship using machine learning methods to predict the price fluctuation of cryptocurrency, it's more like throwing data into a blackbox and just tune the parameters so that the model fits my "arbitrary" criteria. This even happens in some academic work, especially when some blindly worship the p value as the only criteria about significant casual relation. I believe that people should get more training in the technical side before they truly use these tools in production.

[smiklin]

A lot of you (e.g. @romanticmonkey @PAHADRIANUS @liu431 @tonofshell @nwrim) are discussing what I think is a really important part of the problem: 'bullshit' is really effective and often it seems like calling bullshit, on its own, is not very effective. The issue of 'bullshit' seems to run a lot deeper, from confirmation bias, power of certain ideologies and even the difficulty of certain truths, to what might be an overall low level of 'statistical literacy' and understanding of logical fallacies in the general public.

Thinking about the sociopolitical implications of 'bullshit', in which directions should the work of 'calling bullshit', and teaching to 'call bullshit, expand to achieve the most impact? And what are the limitations of this work, being that 'the truth' (if there is such a thing!, some would say) does not hold the same kind of importance to everyone?

[jamesallenevans]

@sanittawan : In response to your question about whether machines can do better than humans at capturing BS, it may be that computational algorithms, because they have different tendencies and human biases, they may be able to help us see things that are hard for us to see--they may not fall prey to the same attentional channels that shape us. For example, they could amplify voices that are hidden by the deluge of information falsely amplified.

[jamesallenevans]

@Yilun0221 : You raise a very interesting point about privacy. AI, because of its power of prediction, makes data that our algorithms uncover liable to violate privacy norms. Data that wasn't privacy violating previously BECOMES privacy violating precisely because we can predict things that we didn't know previously.

[jamesallenevans]

@linghui-wu : Great question! This is something that we need to think about it in our CSS studies, and science in general. "Featuring the surprise" is a pathway to silence the vast majority of insights and findings.

[ctbergstrom]

@PAHADRIANUS You're asking the million-dollar question. I am afraid that I do not have a satisfying answer to provide.

That said, I believe strongly in the power of education, but of course that has to happen before the university level. In collaboration with teachers around the world, Jevin and I are piloting versions of our class in high schools and these are going brilliantly. We also run a program called MisInfoDay, which we hope can become a national model, in which we bring hundreds of high school students to the UW for a day to study misinformation and information literacy. These are just scratches at the surface of a massive problem, though, and it would take commitment on a massive scale to have substantive effects at a national level.

A couple of comments about your examples. While Michael Kosinsky did answer our initial query with a long and somewhat hostile response, that conversation did not continue once we revised our article in response to his criticisms. So that was a partial success, but only partial. On the other hand, the response of Mike Lauer at NIH has been unacceptable. I have emailed him repeatedly as a professional scientist (not just from Calling Bullshit), I have left comments on his blog, I have reached out through his staff, and never received so much as an acknowledgment of my comments. As you can see from the piece I wrote, this involved substantial work on my part to try to improve the funding environment in the US by pointing out fundamental errors leading to decisions about NIH policy. The (lack of) response is extremely disappointing.

[ctbergstrom]

@wanitchayap I agree with you that the Kosinski and Wang paper is oversold, and I agree that researchers face strong incentives to publish in high profile journals and thus to pursue exciting findings. Indeed in my own research I think a great deal about how the incentives faced by scientists influence the knowledge we ultimately acquire or fail to acquire. You can read a 500-world summary of this perspective here: http://ctbergstrom.com/understanding-science.html

I also agree that we need to explore reforms to the formal and informal systems of credit to scientists, in order to reduce the incentive for "spin".

That said, I don't think that the need to get into a good journal was necessarily a major motivator for this particular paper. The paper was announced in a series of popular articles in top news media before it was ever published, and this drew far more attention than any publication would. Of course this is mere guesswork, but I'd conjecture that the fame/notoriety was more the aim here than getting a high-profile publication.

[ctbergstrom]

@policyglot You ask good questions about literature reviews in the social sciences. I was not even well aware of this pattern, though maybe I should have done a better job of induction from the evidence I do have. My short answer is that I do not know the answers to your questions, but I'll ask around. Thank you!

[ctbergstrom]

@sanittawan Great question. We talked about this in class a bit, but in short I'm pessimistic about machine efforts to identify bullshit partly because context and interpretation are very hard, and ultimately because adversarial machine learning should make it easy to defeat any such system even if someone developed a good one.

To answer your first question, I think that we have to work on the assumption that any training data drawn from society are inherently biased, because we live in a society that is deeply biased in many ways. Jevin has been telling me a little bit about the use of synthetic data to get around this problem; basically you create data signature of the world you want, and train based on that. I don't yet know anything about how that is done.

I think that algorithmic auditing is extremely important: if people use AI/ML to make consequential decisions, those algorithms should be available for intense auditing so that outside parties can analyze their behavior and performance, and uncover bias where it exists.

[ctbergstrom]

@nwrim In short, you note that there are two aspects to bullshit control: prevention and cure. Our course is good for prevention, not necessarily so good for cure.

I agree.

We recognize we are playing the long game with what we are doing. It's not obvious that we are going to chance a lot of minds of people who have already committed to believing bullshit. Our hopes is that a new generation can learn not to believe it in the first place.

As an aside, the Nyhan papers you refer about the backfire effect have not generally replicated; it's not clear that there actually is such a thing as a backfire effect.

But that's really neither here nor there. Jevin and I are hopeful that by engaging people in good-faith conversations and giving them the opportunity to grapple with their own misperceptinos around issues, we can have some progress in helping people cast aside bullshit with which they have been indoctrinated. It's an open question whether that will be successful.

[ctbergstrom]

@SoyBison This is an astute observation, and it's one of the reasons why my own definition of bullshit veers away from Frankfurt's. The argument is most coherently laid out in G. A. Cohen's essay, reproduced here: http://learning.hccs.edu/faculty/robert.tierney/phil1301-6/bullshit/g.a.-cohen-deeper-into-bullshit/at_download/file

G. A. Cohen (2002) Deeper into Bullshit. Buss and Overton, eds., Contours of Agency: Themes from the Philosophy of Harry Frankfurt Cambridge, Massachusetts: MIT Press.

[ctbergstrom]

@goldengua Thank you! I really like the way you summarized the purpose of our essay about Wang and Kosinski's work: "Rather than hypothesizing that machine could detect some subtle cues beyond human perception, it is more likely to conclude that machine is better at integrating cues and updating the posterior probability than human beings."

You ask about protocols for linking hypothesis from the results of complex algorithms to the theories we care about, and whether there is any protocol for such processes. I think that this is the art of experimental design, to some extent. In general I tend to be skeptical about having humans compete with AIs and using this process to guess at the causal factors involved in AI's discrimination ability. There are various ways to probe what the AI is doing, of course. Heatmaps hint at what part of an image is being used to discriminate. For example, in our Criminal Machine Learning case study (https://callingbullshit.org/case_studies/case_study_criminal_machine_learning.html) you could try masking out the mouth; if the AI is still able to discriminate, our hypothesis of a smile detector is not entirely supported. And I'm sure there are much more advanced techniques for reverse engineering what an AI is doing, but Jevin probably knows a lot more about those than I do.

[ctbergstrom]

@hihowme Thank you for your question. If I understand correctly, you are asking how we can discourage people from publishing flashy bullshit and encourage them to publish rigorous science instead. It's a huge question and something I spend a reasonable amount of my time thinking about. It comes down to changing the incentive structure that researchers face. To solve the problem, you need to understand (1) what motivates researchers and (2) who has the power to change the rewards they receive AND also has the inclination to do so.

For example, getting a paper in a high profile journal is worth a huge amount now, more than ever before. There's a recent paper that talks about how the average economist would give up (most of) a thumb for a paper in the top econ journal. This incentivizes flashy work on hot topics. But the Science/Nature/Cell publishers don't have much incentive to change what they publish. They publish what people want to read. Other people could look less at where a paper is published when evaluating a researcher, though. The DORA principles advocate doing something like this. But if admissions committees and hiring committees think that this is useful information in judging a candidate's quality, they will not be eager to simply discard it.

And so on, and so forth. I could write a book about this issue.

In general, I think that often the funding agencies have the most leverage. They have the purest motives (more or less) to promote good science, and what they start to value soon becomes what university administrators value, researchers value, etc.

[ctbergstrom]

@anuraag94 It's always fun when people call bullshit on us. So thank you in advance.

You wrote:

"However, you conclude by proposing a most likely explanation for the results—that sexual orientation influences grooming and self-presentation. You can use Occam's razor to rule the author's interpretation out of consideration for the problem at hand, that's valid. However, you propose several candidate models that might rebut the authors' interpretation, yet somehow select and expose one of them as your most likely choice. Without further inquiry, it's an overreach to use Occam's razor to adjudicate between the candidates."

I think this not a fair characterization of what we did. We wrote

"We suspect the most likely explanation for their findings is that untrained humans are poor at integrating multiple cues to make good probabilistic estimates, and that sexual orientation influences grooming and self-presentation. Both of these are well-known facts; the former can explain why the neural network outperforms humans and the latter can explain why the neural network performs substantially better than chance. To invoke mysterious features below the threshold of human perception, or effects of prenatal hormone exposure on facial structure, is to abuse the principle of parsimony."

He were reporting our posterior degree of belief associated with various alternatives. Given that above we wrote "Our point is merely that if one considers the PHT unlikely prior to reading the present study, the results do little to change one's mind", this is really just our prior. We are most definitely not saying that this study provides strong evidence for these claims; rather we are saying that given other evidence for these claims, this provides a much more plausible explanation for the observations.

In retrospect I do think we could have written this more clearly. Reading back over it, the discussion seems a bit overly verbose and more to the point, the use of "parsimony" and "parsimonious" is not very precise. If I have a bit of time, I'll go back and try to clean up the prose.

Thanks, Carl