5/6: Michal Kosinski

[ehuppert]

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.

[rkcatipon]

Hi Dr. Kosinski, it is quite the coup to have you speak at our workshop! My questions are:

• Given the instability of categories like gender and ethnicity what implications are they (if you believe there are any) of including these categories in ML analysis?
• Would it be possible to view the classification accuracy on an intersectional basis, for example how the classifer performed for senior women versus young men?
• I saw that informed consent was obtained for the Facebook data, was it also obtained for the dating website sample?

Thanks and looking forward to the discussion!

[JadeBenson]

Thank you for this interesting research and for coming to our workshop! I’m curious how you navigate the ethical complexities of your work? It’s so important to make the potential privacy-violating uses of these technologies visible so we can attempt to resist them. However, I do worry about inspiring their further use and reifying dangerous ideologies in the process (i.e phrenology). How do approach these ethical problems and what are your recommendations for us to carefully consider our own impact?

[k-partha]

Thank you for presenting at our workshop, Dr Kosinski. This research is fascinating on two fronts: The very fact that face-based classification can predict political orientation at such high accuracy, and the immediate moral dilemmas and privacy concerns it raises.

1) Following up on Jades's questions, to what extent should we be concerned that this technology is already being applied, unbeknownst to us? In your view, how should we adapt, socially and ethically, to a post-privacy world?

2) It is quite shocking that the prediction accuracy is so high despite error associated with the political orientation measurement. To what extent do you think such results motivate and provide clues to further scientific research into the socio-cultural, environmental, and/or innate causes of these correlations between facial features and political orientation?

[Lynx-jr]

Thank you for presenting at our workshop, Dr. Kosinski! I‘m having questions similar to Jade's @JadeBenson and @k-partha about the ethical implications of your work. Moreover, personally speaking, Stanford's IRB wasn't very clear in terms of criteria for approving a study like the 1st paper) and I think such works peek behind the "veil" covering reality that as well as threat those who live in totalitarian states.

And maybe it's just my own bias, I disagree a lot with your past criteria about "public data" originated from your 2015 paper (Kosinski, Matz, Gosling, Popov, & Stillwell, 2015), that is, using public Facebook profile data (e.g., data that are publicly available to any Facebook user and not only to one’s Facebook friends) should not require participants’ consent if conditions such as the following are met: 1) assume that the data were knowingly made public by the individuals; 2) Data are anonymized after collection, and no attempts are made to deanonymize them; 3) there is no interaction or communication with the individuals in the sample. It's easy for MACSS friends to understand my viewpoint, the 1st and 3rd reasonings run counter to what we called the "Respect for Person" principle. However, I appreciate your effort to reuse the data instead of doing a new experiment, and I do saw you said "asking for consent for the FB dataset" in the 1st paper.

P.S. Political orientation is a much milder topic comparing with the gender detection paper. P.S. for anime watchers: The Psycho-Pass world will come to a reality when the 1st paper commercialize and be used by the police lol...

[jinfei1125]

Dr. Kosinski, thank you so much for coming to our workshop! Your papers have shown at least three times in my reading lists of different courses! It's an honor to have you here! Welcome!

I guess my question would be: Both models of your papers, one based on digital traces/footprints and the other based on facial images, can predict human's attributes (personality, political orientation, even sexual orientation, emotion, and probably more attributes in the future) more accurately than a human, so I am a little concerned about the example from movie you give in the personality paper--what if machines understand us more than our closed friends or even ourselves? But it seems that the development of computer vision and artificial intelligence won't slow down, so how should we be aware of that possible future?

P.S. I really love the visualization of both papers that presents the results so vividly. I really learned a lot from them!

[alevi98]

Hi Dr. Kosinski,

I'm interested to hear your talk, especially given the controversy surrounding some of your past work. I'll see you tomorrow.

[bjcliang-uchi]

Hi Dr. Kosinski, thank you so much for the presentation! Regarding the political orientation paper, I am curious about how to understand the "liberal vs conservative" differences in different countries. It seems that the current political context of what it means to be liberal could be very different across countries. For example, UK voters may be motivated by issues like Brexit which is not relevant to the U.S. voters. It makes sense that machines may detect certain personalities that are relevant to political orientations, but I am wondering how exactly it works: emotions --which this algorithm relies heavily on--can change very fast and is context-based, but one tends to be relatively consistent with their party affiliations.

[bakerwho]

Hi and thanks for presenting at our workshop.

As AI doomsday predictions of events like a singularity abound and realities of machine bias, algorithmic unfairness, and digital policing are discovered to be ubiquitous, the burden of effect is rightly placed on AI researchers who actually develop algorithms. Your work on facial features being predictive of other characteristics is reminiscent of physiognomy and early criminology - fields that were inherently white supremacist and racist. In your own research you caution against the dangers of the very applications you design.

What are you doing to actively mitigate harms from these new algorithmic applications, some of which you yourself have created and shared with the world?

[Tanzi11]

Thank you Dr. Kosinksi for your presentation, I look forward to your thoughts on facial recognition technologies and its risks for being a possible hazard!

[anqi-hu]

Thank you for sharing your work with us! The realities of what platforms and big techs are capable of with our data is truly daunting. I have always found the boundaries of improving user experience to be ambiguous when it comes to privacy. Where do you think these limits exist and how do you think they should be established?

[yongfeilu]

Thank you very much for your presentation! How do you deal with the privacy issue, especially your research will collect people's physical information? Is there any better ways to conform to ethics and meanwhile guarantee the data used for research?

[MkramerPsych]

Dr Kosinski,

Thank you for sharing your research with us! I am curious as to what you may think are the factors responsible for the performance of your political orientation classifier. You make a point to state that your results are not evidence for liberals and conservatives having different faces, but if this is not the case what do you think could underlie this relationship? I am also curious if your data could be utilized in a reverse correlation analysis to synthesize the "average conservative" face and the "average liberal" face to determine if any differences result.

[sabinahartnett]

Dr. Kosinksi - thank you for sharing your research with us! I am looking forward to the talk and really interested to hear your opinions on facial recognition technology, implementation, and some of the recent (US & elsewhere) legislation surrounding it.

[mingtao-gao]

Thank you for sharing your work with us. This is an extremely interesting paper. It's surprising to learn facial recognition technologies can be used to identify political orientation. My question is do you think the pattern found in the results may reveal some stereotypes people might have regarding political orientations? For example, as you mentioned, liberals are perceived to be more open and less conscientious.

[MengChenC]

Thank you for sharing your findings. Even though we have recognized the power and potential of artificial intelligence, it is still startling to see their performance in facial and psychological recognition. This may be a rudimentary question, while I am wondering how you chose the method in the personality judgment? Did you try out other more complicated or non-linear approaches since the big 5 characters are not necessarily all have a linear relationship? Thank you.

[ginxzheng]

Thank you so much for coming!! I really enjoyed reading your paper about facial recognition technology. I would like to hear more about how do you validate the accuracy and how would you define each type of judgment? Thanks!

[xxicheng]

Thanks for sharing your amazing work with us! I wonder about your future steps in facial recognition.

[Raychanan]

Thanks for sharing your work!!

[TwoCentimetre]

Does this provide potential rationale for discrimination and stereotype?

[wu-yt]

Thank you so much for presenting this interesting work? How do you think big internet companies will respond to your research?

[linghui-wu]

Thank you for sharing such fascinating work with us, Dr. Kosinski.

[vinsonyz]

Thank you for sharing your work with us, Dr. Kosinski. My question is how your work could be applied to other fields such as economics.

[kthomas14]

Thank you for sharing your research Dr. Kosinski! I was wondering about the applications of your work in a neuropsychological setting. Do you believe that with some development and public acceptance, these ML techniques may be able to be used alongside other forms of neural analysis to make formal predictions about patient health?

[mikepackard415]

Thanks for coming to the workshop! One of my reactions to your work is that I add it to my growing list of reasons to just use social media less and to produce less internet-based personal data. Would you say that reducing our digital footprints is something you are seeking to encourage with this research? Thanks, looking forward to the talk.

[chiayunc]

Thank you for coming to the workshop. In Wu et al.(2014), it's stated that personalities can be predicted automatically without social-cognitive processes. I wonder to what extent is this statement reliable. Thinking about the fact that digital footprints left by users are not capturing all personal activities and cannot 100% capture what people would do in actual social settings, I myself feel constantly that I don't act the same way I do in the real-world online. Do you think it would be better justified to say that we can automatically predict digital personalities(for lack of a better term)? and that there are nuances in human-societal activities that are not reflected in the virtual world? I think it's quite dangerous to mirror personality predictions based on digital data onto real-world people.

[YuxinNg]

Thank you so much for the sharing. I would like to hear more cases in which the predictability of psychological traits bring risks. Also, I think different people or country may hold different opinions on the definition of "unethical". Should / How should we take that into consideration? Thank you and Looking forward to the presentation.

[adarshmathew]

Thank you for presenting your paper at our workshop, Dr. Kosinski.

I'll leave the substantive questions on your paper and the associated Google doc to my more capable peers on this thread, and take this opportunity to ask you something else. I've gone back and forth on asking this question and framing it right, so I hope you'll forgive any incoherence or combativeness.

My first question is about your specific research agenda vis-a-vis Computer Vision. Your oeuvre has been to use these new methods to showcase correlations between facial/physiological features captured by images and personality, political preferences, sexual orientation etc. It's a well-noted departure from traditional social science research, in that you don't attempt to provide a plausible causal story or explanation or study design to explain these findings. Being trained in psychology itself, what would you argue the value of your research is to the broader scientific endeavor? Unlike your Gaydar paper where you at least posited the prenatal hormone theory to explain your findings, there is nothing of the sort in your Nature paper here. Is it just a case of "Hey, this is an interesting and disturbing result, it's up to you to figure out why this is happening", which eliminates the difference between a CS researcher proposing methods and one trained in the social sciences trying to understand the causes of social phenomena?

My second question has to deal with the role of researchers who straddle social science and machine learning. Your work, for all its impressive results and caveats, does not deal with the intellectual and political histories of practices like phrenology and physiognomy. Given that you are presenting a blueprint for scaling up these politically motivated practices with a dark and terrible history, along with highly contested scientific foundations, how do you choose to navigate the trade-off between "Hey look at this cool result!" and "Here's a blueprint for getting the most of your political mass surveillance apparatus"? My question is motivated by the work of the Philip Kitcher, an eminent philosopher of science, and his arguments on the role of Science and the Public Good in his book Science, Truth & Democracy.

[chrismaurice0]

Looking forward to hearing you speak tomorrow Dr. Kosinski! My question is do you think campaigns should use facial recognition to target voters, or is that taking technology a step too far?

[skanthan95]

Thanks for speaking at our workshop this quarter! Looking forward to your talk. My question is about your thoughts on HireVue, as a hiring mechanism.

[Yilun0221]

Thanks for the presentation! My question is about people's personality and the photos. I think there might be some inconsistency between the two things, since people will pay attention to their images on public platform. I wonder how you think of this? Thanks!

[yutianlai]

Thanks for coming! Can we broaden the research scope to other fields?

[Qiuyu-Li]

Hi Dr. Kosinski, thank you so much for the presentation! I really enjoyed reading your paper about facial recognition technology. You remind me of another infamous paper which tried to identify homosexual people. Since it is somewhat ambiguous and subjective to distinguish ethnically-questionable paper, I would like to know your opinion about the feasibility about developing a uniform rule to guide ethnic review that everyone is convinced of, just like the strict restrictions on performing gene editing on human in the Biology academia. Thank you!

[AlexPrizzy]

Thank you so much for attending our workshop Michał. I've been interested in your work for quite some time now and know there has been public backlash against your "Gaydar" study. At first look such as study raises ethical concerns, though you make a good point by showing potential dangers of new technologies.

Would you publish the sexual orientation study again and were you concerned about similar backlash when publishing the study on political orientation?

[lulululugagaga]

Thanks for sharing the work! I think your discussion is really important and provides insights for how we address ethical issues in facial recognition. Looking forward to this talk!

[YanjieZhou]

Thanks so much for your presentation! I think HireVue is a very good point to research, but could you elaborate on it like its specific restrictions?

[LFShan]

Thank you for the presentation. I am personally very interested in finding out how online behavior reflects offline activities. I am looking forward to the presentation.

[a-bosko]

Hi Dr. Kosinski,

Thank you for coming to our workshop!

In the article about computer-based personality judgments, it is mentioned that computer predictions are more accurate than those made by participant’s Facebook friends using a questionnaire. Although this might be vaguely related, the concept of computers outpacing humans reminds me of the AlphaGo versus Lee Sedol Go match in 2016, where Google’s DeepMind program beat the world champion Go player.

With this in mind, do you see computers taking over in the realm of psychology? Specifically, will computers be able to tell more about an individual’s personality traits or future outcomes than even a human can predict?

[MegicLF]

Thank you for sharing! I’m curious about how you approach the ethical complexities of your work and why you do it that way.

[luxin-tian]

Thanks a lot for sharing! This is really an important topic and provides insights on the ethical side of pattern recognition.

[hesongrun]

Thank you so much for sharing! How can we apply this method to other social science disciplines such as economics and finance?

[egemenpamukcu]

Thanks for presenting at the workshop, Dr. Kosinski. I would like to hear your responses to some of the questions about ethical concerns raised by my fellow friends above. Also, what do you think are some of the intermediary mechanisms between facial expressions/features and political ideology? Why do you think these algorithms can perform so well with very little information?

[JuneZzj]

Thank you for presenting in advance. It is fascinating to learn the accuracy rate is high using this model. As you mentioned, this high accuracy is enabled when we use multiple images per person. Except for those, do you have some other ideas for us if we want to avoid the concerns of particular specification of the model we trained. Thank you.

[Anqi-Zhou]

Thanks for your sharing!! It's really an interesting topic. You have mentioned it could be used in the education of management, can you please explain a few examples to us? And do you think it will give scholars a fresh perspective on investigating the consumer behaviors?

[ydeng117]

Thank you for sharing your work. With the development of big data technology and our lives being mediated by digital devices, I am worried about the power dynamics that large companies process by controlling our data. What measure do you think we can hold these companies accountable?

[Jasmine97Huang]

Fascinating work! I am interested to hear more about the ethical implications of the methods that you seem to favor. Looking forward to the presentation/

[YaoYao121]

Thank you Dr. Kosinksi for your presentation. This is really a interesting research topci. It provides a different angle about analysis on facial recognition technologies and its risks. Besides, I am a little curious about the practical implications from your conclusion. Could you explain more about the external validity of your results, please? Thanks!

[ChivLiu]

Thank you for the presentation! I like the way you present the profiles on the internet and how we could use them. User Profile Analysis has been signified by internet companies for years, and they even use AI algorithms to benefit more from frequent users. How would you comment on those behaviors?

[tianyueniu]

Thanks so much for sharing your work! Looking forward to the presentation!

[minminfly68]

Thanks for coming to the presentation and can I kindly ask how much this work would be applied to other fields outside the scope that you have been mentioned? Many thanks.

[FranciscoRMendes]

Thank you for coming! I am also concerned about the ethical and moral dilemma of having such a technology. Excited to hear you speak!