Hi everyone and all the best in the new year!
It is this time of the season that numerous resolutions emerge and a lot of people decide to start being the better version of themselves - gyms are full, diets start, enrolments for educational courses plummet. This week I decided to suggest an article from the researchers who have developed an app which predicts age and attractiveness level. Their work showed significant results in predicting ratings on previously unseen photos uploaded to a popular dating website.
As a discussion starter I would like to ask your opinion on the following questions:
- What are the biggest obstacles for preference prediction algorithms ?
- In your opinion, should attractiveness be measured? Alternatively, does it cause any major ethical problems?
- Is one photo enough to measure attractiveness? What other features researchers might use?
- What are the possible applications of age, gender and attractiveness prediction algorithms?
1. The biggest obstacle for preference prediction algorithms is the change of people's preference over time. What one like in one day, might change in the other, which might be difficult to mimic utilising algorithms.
ReplyDelete2. In my opinion attractiveness is the personal view and is measured whether we want that or not. It's however the subjective assessment of others, which is proved by scientific researches stating that first 3 sec makes an impression of the person observed. I can't see any ethical dilemmas associated with attractiveness measurement, given it is commonly approved practice - eg. all these beauty contests like Miss World or Miss Poland.
3. One photo in my opinion is not enough for measuring attractiveness, especially the one pimped up with Photoshop. Researchers might and should use the holistic approach to assess the persons attractiveness measuring not only the generally accepted standards of beauty accompanied with the IQ scores.
4. The possible applications of attractiveness prediction algorithms can be used for speed dating or web based dating services.
Attractiveness and beauty in general is a very subjective matter. Someone can find me handsome, but another person will say I'm not her type or that I'm ugly. So in my opinion, such algorithms and applications are not universal from their very beginning, nature and this is the biggest obstacle. I don't think that one photo is sufficient to measure one's attractiveness level. A person can look very differently on each and every photo she took. As per possible applications of attractiveness prediction algorithms I agree with Mateusz that they can be used by speed dating, online dating services, tinder etc.
ReplyDeleteHello,
ReplyDeletethis topic is quite weird, because if the want to classify people as beauty and ugly or young and old ... good luck! I would like to give an example ... of course beauty is something what will change in the future. Please take a look how canon of beauty has changed for last decades :). If the ask your child: Is somebody old then we will receive answer yes (in most of cases if somebody is over 30 years old). But if the ask 80 yrs old grandma then :) 40 is still very young ;).
Are we able to prepare AI, which will realise this kind of classification - yes. Regarding to yours 2nd and 3rd questions - yes, but everyone has own preferences. I have seen perfect women's face visualisation that was based on the math pattern ... she was nice, but was not perfect in my opinion. We can calculate proportions and compare to a pattern.
I have seen this kind of application and I treated it as a toy. We can use it in shops - pretty person will see cosmetics on a screen and ugly person will see stronger set of cosmetics :)? (joke)
Hi Pawel,
DeleteI agree that the topic is very subjective. This is why I find it very interesting - we can tackle the problems which have some clear metrics but when it comes to perception such as art, aesthetics, beauty, stylistical similarity or even user experience it is so much more difficult to analyse.
1) I would say that the biggest obstacle is the fact, that preferences are not stable in time, and that preferences, contrary to what economist like to assume, are not transitive.
ReplyDeleteAlthough people tend to disagree on whether given celeb is attractive or not, I would say that automatic beauty prediction is possible, we are talking about 'mean' and 'averages' here for crying out loud.
When people talk about that 'the beauty is in the eye of the beholder' or that 'it's all subjective man, I have my preferences' what they forget is that matter is the crowd consensus.
2) Well, should anything be measured? The measurement is not the problem, it's the obsessive comparing with each other and stratifying people according to those benchmarks, what's the problem.
4) I second the motion of @Marcin Mogilski, that attractivenes could be used for matching people on dating websites. Alternatively, such system could warn the used to take another foto. Or just to hang oneself ;)
Thanks Wojtek for your feedback!
DeleteSimilar systems to rank people on dating websites apparently are emerging already and it is in every user's interest to keep them secret :)
https://www.fastcompany.com/3054871/whats-your-tinder-score-inside-the-apps-internal-ranking-system
Hi, all the best in the new year for you too! Most of these new year resolutions come prematurely to an end. I've already seen numerous apps that supposedly guess one's age and even how one will die. ;)
ReplyDeleteLet me try answer your questions:
1) I think the biggest obstacle for preference prediction algorithm is the high amount of different preferences that people have. Beauty is a subjective quality, although research shows that statistically speaking most people will agree that symmetrical faces are considered better in this aspect as they indicate higher probability of healthy status (better reproductive traits for offspring).
2) As long as it is handled with care (not stigmatising people) I don't really find it too unethical. Physical appearance is just one of many characteristics/traits that a person can have. It is like asking if IQ or maturity tests are unethical.
3) That depends on the algorithm, I guess that for a very basic evaluation one photo could do the trick, unless it if of bad quality.
4) A service that would make use of all three of the mentioned parameters is probably a dating site. But that was already mentioned. ;)
Thank you for your opinion Michail! To your experience, were these apps accurate (guessing your age? ) ? Do you have any interesting examples?
DeleteIvona – thank for an interesting and funny article. I receive this article with mixed feelings. On the one hand I believe that all issues can be counted by computers but on the other hand in my opinion attractiveness can’t be measured by computers. Therefore the biggest obstacles for preference prediction algorithms are human and their different tastes. Depending on for example part of the day, our mood, our age etc we once choose one type but another day we choose other type of attractiveness. In my opinion human is unpredictable generally and this kind of technology to predict best matches without human are rather useful. Moreover I reckon attractiveness can’t be measured because everybody is different. One likes this type of human but other likes different types. In my opinion only big set of personal photos, answers and questions etc can help for example computers try to measure something called attractiveness.
ReplyDeleteThank you for an interesting insight! I agree with you that it is extremely difficult to have any subjective metrics for beauty evaluations, thus perhaps computers and machine learning could help create some beyond the human comprehension/evaluation (e.g. face symmetry)
Delete1.What are the biggest obstacles for preference prediction algorithms?
ReplyDeleteI agree with previous comments that the biggest obstacle are preferences that are changing rapidly. But some general prediction of attractivity is - my opinion - possible.
2. In your opinion, should attractiveness be measured? Alternatively, does it cause any major ethical problems?
I'm ok with that. I think phisically attractive people often know that they are attractive, as well as those who are not attractive phisically know that too. But I attractivness of a body is not everything.
3. Is one photo enough to measure attractiveness? What other features researchers might use?
I think that for very basic estimates it would be enough. But maybe tone of a voice could be used too to evaluate attractivness of a person.
4. What are the possible applications of age, gender and attractiveness prediction algorithms?
Nothing more comes to my mind than things other people mentined - dating websites.
1. What are the biggest obstacles for preference prediction algorithms ?
ReplyDeleteI think that the biggest obstacle is that sometimes even people don't know excatly ahat they prefer :) sometimes they find someone attractive despite this person doesn't have any features they like in most of the people. Also people change their prefferences based on experience.
2.In your opinion, should attractiveness be measured? Alternatively, does it cause any major ethical problems?
I don't see any problem wuth that. People measure their attractivnes in competitions so why not do it by neural networks.
3.Is one photo enough to measure attractiveness? What other features researchers might use?
One photo is not enough but can be a good clue. Atractivness is not only phisical appearance but also behaviour etc.
What are the possible applications of age, gender and attractiveness prediction algorithms?
I won't add anything that haven't been mentioned. Dating websites is the most obvious application.
What are the biggest obstacles for preference prediction algorithms?
ReplyDeleteI think that there are exceptions's rules.
In your opinion, should Attractiveness be measured? Alternatively, does it cause any major ethical problems?
If the man himself on this agreement, it is ok. If the algorithm used data without authorization it would be unacceptable.
These photo is enough to measure Attractiveness? What other features researchers might use?
It seems to me that one picture is enough if they are captured on the corresponding features.
What are the possible applications of age, gender and Attractiveness prediction algorithms?
I think that portals for singles are a great place for this type of algorithms.
Hi Ivona, thank you for sharing this.
ReplyDelete1. What are the biggest obstacles for preference prediction algorithms ?
There are quite a few issues, both technical and non-technical ones.
Given my background (I was Globalization Software Engineer for many years), I think the biggest obstacle lies in culture. They say that beauty is in the eye of beholder. And indeed, it is. What is considered beautiful in India, is not in Central Europe, especially if we focus on males. There just simply could not be one, superior algorithm that would judge the attractiveness globally.
There are quite a few technical obstacles as well. For instance the size of a training sample (and of course subjectiveness of training classes). Quality of samples (i.e. resolution) could be a problem as well. I am stating the obvious, but how the picture is focused, and what a subject wears make whole lot of difference. A woman in her hijab might be mis-labeled for instance.
2. In your opinion, should attractiveness be measured? Alternatively, does it cause any major ethical problems?
Everything depends. It might help find right matches on a dating site. Whatever we say about this, first impression is an important factor. For better or worse.
However, I would be strongly opposed to use attractiveness (or in more general physical appearance) as a hiring factor in a recruitment process. That would simply be unfair.
3. Is one photo enough to measure attractiveness? What other features researchers might use?
It depends on the photo. Sometimes, one is more than enough.
Other features... Age could be learned from a voice sample.
Larger voice sample (or text sample) could also be used to decide whether somebody finds the person interesting (but not necessary attractive). I bet that this could also lead to age and racial/social profiling, if you're given large enough corpora...
4. What are the possible applications of age, gender and attractiveness prediction algorithms?
There are many possible applications; most of them are controversial. The obvious one is dating. Another (usually illegal) is profiling in a recruitment process. Please let me stop here as my thoughts are gravitating towards concentration camps; something must have gone into wrong direction.
1. What are the biggest obstacles for preference prediction algorithms ?
ReplyDeleteWell it reminds a bit of this tv show “Married at First Sight” where some specialists chosen two people based on science and let them married first and after that to get to know each other. This time there is algorithm… As we know even from our personal lives, many times we are surprised when it comes to choice of the parter. People that seems to live in completely two different worlds are together and they seems to be happy. In tv show mentioned by me almost none of these marriages lasted, everyone got a divorce. So basing only on some scientific knowledge I don’t think that results in real life will be promising.
2. In your opinion, should attractiveness be measured? Alternatively, does it cause any major ethical problems?
In the majority of cases that is true that people who are attractive are looking for someone who also take care about himself/herself. That is true that attractiveness does matter, but it is not 100% of cases. For sure people claim that outside look is not everything, but we know how it works.
3. Is one photo enough to measure attractiveness? What other features researchers might use?
Well in this tv programmed attractiveness was measured by some proportions. People who have exact hight, weight, proportionality and many other physical traits were connected by measurements. I don’t think that photo is enough to collect data, firstly one photo differentiate from another one, secondly normal physical measurements will be more accurate.
4. What are the possible applications of age, gender and attractiveness prediction algorithms?
Maybe some dating websites will use it. I don’t see any important usage.
1. What are the biggest obstacles for preference prediction algorithms ?
ReplyDeleteI think that prediction won't be 100 percent accurate and that is because peoples preferences are always changing and second thing is that, that I always thought about this kind of alghorithms as some sort of average from data on which this alghorithm is trained. So for me the biggest obstacle will be data "freshness" because we can have data about peoples preferences from 2013 and we want to do experiment in 2017.
2. In your opinion, should attractiveness be measured? Alternatively, does it cause any major ethical problems?
I don't think that measuring attractiveness is ethical problem. Everyone of us is searching something different in other person. so when algorithm is traying to create perfect men or women he is getting mode from data and then creates perfect companion. But when you are searching differenet kind of atracvtiveness that was served to you, person may seem ugly for you.
3. Is one photo enough to measure attractiveness? What other features researchers might use?
My answer will be quite similar as one above.
4. What are the possible applications of age, gender and attractiveness prediction algorithms?
I don't trust this kind of algorithms and that is because one day I have upload my photo into program made for age prediction (and at this time I was in my late 20s) and this algorithm predict my age for about 50. If I will ask human (even now) my age would be around 22 but now I am 11 years older. So right now I don't see any application for this kind of stuff.
1. What are the biggest obstacles for preference prediction algorithms?
ReplyDeleteThere is serious problem with all prediction algorithm. You could process to small amount of input data and receive inaccurate result. Another problem is when you process to much data in neural network and you receive such enormous abstract generalization that in no way present real world answer for investigated problem.
2. In your opinion, should attractiveness be measured? Alternatively, does it cause any major ethical problems?
At the beginning I would saw a fear and call it eugenics. Next I would state an assumption that every healthy human being it almost every worthy minute.
3. Is one photo enough to measure attractiveness? What other features researchers might use?
It is usually known - first impression - and it might be true love seed on the other hand it might be a beginning of total tragedy. I would say that true attractiveness is worked out in long term. However there is something special and unique for every single person that some photo profiles tend to be more liked by them.
4. What are the possible applications of age, gender and attractiveness prediction algorithms?
Ask anyone who know art about importance for this prediction algorithms.
What are the biggest obstacles for preference prediction algorithms ?
ReplyDeletePeople? Generally, it's hard to match human beings. Even though, we can share common hobbys, have specific look and so on most algorithms will fail in the real life. We are far too complicated and if I take myself as an example, I can bet that not a single one would be predict my choice :D
In your opinion, should attractiveness be measured? Alternatively, does it cause any major ethical problems?
It should, but how? This is the question, there is a quote: It is not beauty that endears, it's love that makes us see beauty.
Is one photo enough to measure attractiveness? What other features researchers might use?
It isn't. Inquiries?
What are the possible applications of age, gender and attractiveness prediction algorithms?
Currently, we have thousands of different dating websites and applications. Do we need them? Who knows...
1. What are the biggest obstacles for preference prediction algorithms ?
ReplyDeleteI think that everything which is measurable will not be a problem for the algorithms. Look and as voice can be certainly processed by a computer. The main abstacles may be binded with the senses of taste and olfaction.
2. In your opinion, should attractiveness be measured? Alternatively, does it cause any major ethical problems?
I don't see any ethical problems, unless, the photos from CV/resume are not taken into consideration. On the other hand, if one of the requirements is the attractiveness, why not?
3. Is one photo enough to measure attractiveness? What other features researchers might use?
Provided, that the photo fulfills some requirements - the whole silhouette, visible face, one photo may be enough. As I have mentioned, maybe some samples of pheromones should be examined?
4. What are the possible applications of age, gender and attractiveness prediction algorithms?
To be honest, I don't see many applications. However, I won't be surprised if supermarkets predict the age and gender of its clients.
1. What are the biggest obstacles for preference prediction algorithms ?
ReplyDelete2. In your opinion, should attractiveness be measured? Alternatively, does it cause any major ethical problems?
3. Is one photo enough to measure attractiveness? What other features researchers might use?
4. What are the possible applications of age, gender and attractiveness prediction algorithms?
I think that it is impossible to measure things like attractiveness. There is no way to find scale that will be interpreted in the same way by every person. But if we think that we can and should measure it that I am sure that we need much more than one photo. We need a full sequence from different perspectives. Of course we need to remember that we need same pattern for photos for each person that we want to measure. I am afraid that such measurements will create even bigger problems in our community and problem of the segregation will come back
What are the biggest obstacles for preference prediction algorithms?
ReplyDeleteIn my opinion the biggest obstacle may be the variety of human preferences. There is no one canon of beauty. It varies between countries, cultures, personal opinions and also age. Of course there are people that are "pretty" to the most of society, but they might not be the "perfect match" for a lot of people.
In your opinion, should attractiveness be measured? Alternatively, does it cause any major ethical problems?
My personal opinion is no. The attractiveness should not be measured. I think that the whole impression that a person makes consists of many factors, like appearance, character, sense of humor etc.
Is one photo enough to measure attractiveness? What other features researchers might use?
It depends. Photo can show how a person looks in every day life, but also can give a better or worse impression of the photographed person.
What are the possible applications of age, gender and attractiveness prediction algorithms?
There was an application some time ago (on facebook I think, but I am not sure) that was guessing age by analyzing photo. I didn't use it, but a lot of my friends did.
Attractiveness can be measured by where there are generally acceptable ideals. Then measure the attractiveness informs us about the extent of compliance with this ideal. But as a measure of the attractiveness of compliance with the chosen ideal - it's a little objectification of people. In my view, the attractiveness is very broad. If we think about the appearance, intelligence - we can measure it, we know here examples of metrics but other issues - not really. The next question is how many features we want to measure - though for different people, different issues are important. The attractiveness depends on your personal ideals which are sometimes variable - unfortunately. Moreover, I do not want to be by metrics attractiveness especially for dating - especially when it comes to personality traits of character - if you imagined how many appeals would be, maybe even some applications for compensation ... Unfortunately, almost all algorithms led to artificial segregation and the "invisible" selection.
ReplyDelete1. What are the biggest obstacles for preference prediction algorithms ?
ReplyDeletePreferences always depend on every human and his outlook. Different races have their own culture and understanding of beauty. So the main problem I see in definition of some basic preferences.
2. In your opinion, should attractiveness be measured? Alternatively, does it cause any major ethical problems?
Yes, I think it may causes some problems. Attractiveness consists of few main factors, as physique or character. Some people can look not so great but in spite of all they can be still attractive. It may be unfair in such situations.
3. Is one photo enough to measure attractiveness? What other features researchers might use?
For face assessment it may be enough but not for whole attractiveness. Other features I mentioned above.
4. What are the possible applications of age, gender and attractiveness prediction algorithms?
At the moment I see few possibilities in dating systems, where people can be matched based on their preferences but as a first step of selection of course. Maybe in police search it can be useful to age assessment of wanted.
What are the biggest obstacles for preference prediction algorithms ?
ReplyDeleteI think the biggest obstacle for preference prediction algorithms is unstructured data.
In your opinion, should attractiveness be measured? Alternatively, does it cause any major ethical problems?
I could not be measured because it is classification problem rather related to statistics.
Is one photo enough to measure attractiveness? What other features researchers might use?
We can talk about good problem classification when we have enough big data set, so in my opinion one picture is not enough. If it comes for feature engineering in my opinion we should relay on previous classifications.
What are the possible applications of age, gender and attractiveness prediction algorithms?
I can't imagine application for such algorithm.