UNBIAS Game by The University of Edinburgh https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05 en-us 2020-08-04 08:51:13 UTC 2026-01-04 17:36:25 UTC hello@padlet.com Demo task output moocdeliveryteam https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/668876959 2020-08-04 08:51:13 UTC https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/668876959 UNBIAS Game moocdeliveryteam https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/668876960 2020-08-04 08:51:13 UTC https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/668876960 Hate Speech & Cyberbullying https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/935689530 Do we have the right to share things that we like or dislike even when some consider this hateful or harmful?  Surely we have the right to pursue our own values; if we enjoy sharing our thoughts why should this be restricted?  Sharing our thoughts allows for personal development since others can help us to find the truth when we err.  Sharing offensive ideas sometimes leads to major progression i.e. Copernicus would have been deemed offensive when he explained that the world revolved around the sun.  By limiting freedom of speech will we miss important scientific and social developments?  
The problem here is that others have rights to pursue their happiness as well and hate speech and cyberbullying can massively affect the wellbeing of others.  There have been cases where addresses have been posted online to target individuals this is inciteful of hate and does not protect their personal data.
How can we solve these problems?
With great difficulty....
There is currently a white paper suggesting legal intervention in case of online harms in the UK. https://www.gov.uk/government/consultations/online-harms-white-paper
But do we really want to allow our government to limit freedom of speech?  The right by which we defend all other rights?
Social media aims to monitor hate speech but unfortunately this is done by algorithms not people and even when reported the user may not be helped if it does not tick certain boxes defined as harm by these companies.
My personal thoughts are that hate speech is always unnecessary, we can diverge in viewpoints but share this in a constructive way.  The problem is that due to online anonymity there are often no consequences for users engaging in such acts and The Stanford Prison Experiment shows that people can do awful things when empowered.
Perhaps one solution would be an of-com style body which monitored the web for abuse and helped users who are targeted. This body would most likely be able to respond to issues far quicker than the court systems could. The only problem would be to define what constitutes hate speech, is it harm? If it is would offence constitute a harm? This is a tough question that needs much consideration.]]> 2020-11-18 09:01:05 UTC https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/935689530 Personalisation and Employment Rights https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/953061940 I think this is already covered by the UK DPA 2018 and GDPR.]]> 2020-11-23 16:41:38 UTC https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/953061940 Hate speech and cyber bullying https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/953861255 The cards had a strong theme running through them, that of power and the abuse of power, in this case in a work place setting. Work is an environment full of hierarchies and power structures, and it is not uncommon for that power to be abused. One form that this takes is bullying behaviour. One way that this toxic behaviour can be tackled is with appropriate monitoring, but of course the definition of appropriate is the difficult issue here. It is an expectation in highly regulated industries such as finance that their is a fair bit of surveillance of communication and activity, but less so where there is less regulation. The solution for providing a fairer outcome in terms of communication given a dynamic situation of health and safety considerations, power context and data led framework is probably some level of agreed communication analysis - this could be sentiment analysis or a hybrid approach where user feedback is considered along with signals. In any situation however, the underlying theme that runs across these random cards is an interesting one!]]> 2020-11-23 19:58:39 UTC https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/953861255 Can algorithms be racist? https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/955770253 Well yes! Algorithms can be and are racist at times as they were 'fed' data that was already biased. It is not surprising then that in this context the decision about what is professional vs unprofessional hair was racist. I don't think the people in those photos would have ever agreed to their personal information about ethnic identity being used in this way and I consider this an infringement of their equal rights. The negative consequences for this could be reinforcing existing bias when it comes to job searching but also making these people feel like the don't belong, like they are different and not part of their community. The solution is to obviously check the data you feed into your algorithm and see if it is biased and use more objective, scientifically gathered data.

]]> 2020-11-24 11:27:27 UTC https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/955770253 Hate Speech and Cyber Bullying https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/958192507 My example was Hate Speech and Cyber Bullying, the Value was Hedonism, the Data was Health Records, and the Rights were Consumer Rights when purchasing something (not the best for health records).  
One ethical problem would be someone accessing health records and selling them to others without consent.  This could then lead to hate speech against someone who had an illness or bullying about the costs of treatments.  The consumer/patient's rights would have been violated by the health care provider.  I'm not sure how Hedonism comes into play here unless someone finds pleasure in bullying others or in revealing health data.  
Solution: Higher security on health data which requires encryption and physical security measures regarding access and storage.  As for algorithms for cyber bullying, possibly look for derogatory speech relating to health and illness?

]]> 2020-11-25 00:02:42 UTC https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/958192507 Gender Bias https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/959592468 My example was Gender Bias, the Data was Personal Data, the Value was Commerce and the Rights were Equality Rights.
The Gender Bias example looks at men and women being shown different job adverts, specifically that men were shown jobs of a higher salary than the women. When combined with personal data this becomes even more problematic as it could be seen as companies potentially not wanting people from poorer areas to apply for their positions. 
This works quite well with the equality rights issue as when personal data is being used which covers gender, race, sexual orientation, as well as the maternity/paternity issues it can be seen as directly violating the Equality Act of 2010, as men are being favoured over women for highly paid jobs. This is sexist but it can also be seen as pushing for employees that either don't have or don't want a family so they can avoid paying for maternity/paternity leave. ]]> 2020-11-25 12:09:27 UTC https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/959592468 Automated Moderation and Hate Speech https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/959639793 My ethical problem was Automated Moderation, the data was Politics and Beliefs, the values card was Tradition, and the rights card was Human Rights. The example detailed how many online platforms automatically detect content that violates their T&Cs or represents hate speech. Facebook, Twitter, and other social media platforms have been struggling with doing this as the American presidential election (politics and beliefs) brought to light how polarized the country has become as a result of the clash between tradition (white people in power, status quo being upheld, systemic racism staying in place) and human rights abuses in the form of denying women bodily autonomy and locking Hispanic children and their families up in cages and sterilizing the women, as just a couple of examples. All of this was discussed in the comment sections of Facebook, on Twitter, and in the forums of sites like Reddit as well. These sites have only recently begun to flag misleading information and hate speech and taken steps to moderate these problems on their platforms using automoderation. It’s an imperfect solution because though automated moderation is needed due to the volume of content handled by these sites (human moderation just isn’t scalable in this context), the AI behind the automod doesn’t understand context or colloquialisms, opening up the potential for false flags and hate speech flying under the radar. One solution could be feeding the algorithms the millions of examples of hate speech flowing through these platforms as a training dataset so the AI could learn to identify and flag hate speech in all of its permutations. ]]> 2020-11-25 12:30:26 UTC https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/959639793 Personalisation and financial data https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/960874131 My problem card was about the increasing personalisation of the content we see online, including recommendations and advertisements. (As the chap mentions in the example video, online recommendations can be hilariously inaccurate. For example I often used to get amazon recommendations for different editions of a book I'd just purchased. Bit of a no brainer there..)
My data card was financial data, and this made me think of the move lots of banks are trying to make towards more personalised products (I work in the sector). For example from an analysis of someone's transaction it would be possible to identify when someone was likely getting ready to move house, or maybe about to have a big life event such as a wedding or a baby. Using this information you could target advertising for financial products and offer personalised rates. Open banking adds to the possibilities here. Thinking about data protection, this made me think of the point made in the interview with Zuboff that it's not just the data we give companies and consent to provide, it's the 'surplus' I think she called it. I.e. what other information about us they can infer from the data we willing provide. I may be wrong but I don't think it's clear legally what the boundaries are for using such inferred data for things like targeted advertising. And are consumers ready to receive targeted and personalised products in an area like banking? Would they trust they were getting a good deal? And how would you know whether what you were being offered was fair? ]]> 2020-11-25 18:06:09 UTC https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/960874131 Racist algorithms https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/966146155 Algorithms trained on official identity data can be, if not well thought, specially racist. Information like immigration status, nationality and even the employment status and employer details contain clear encodings of ethnicity and origins of a person. 

This is specially sensitive if those algorithms are developed and used to dictate a treatment for a health issue, in light of offering a more scientific and informed treatment. While in the light of personalized treatment, where it has been proven that tailored treatment can be more effective for some types of cancer for example, it can also create disincentives to not develop new medicines for a group if an algorithm has evaluated that there is no treatment due to the ethnicity of a person. 

 Here, as patients, we would need to request an enhancement on patients rights, so they include the ability to understand and challenge the algorithms that make de decisions for us, receiving information on other treatment options, and their risks and benefits. ]]> 2020-11-27 18:18:09 UTC https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/966146155 My selected card was about historical bias. I considered the impact of data gathered from internet searches on a person`s human rights in the free world if he values altruism. If someone values helping humanity and the natural world, he is likely to succeed in contributing properly if he educates himself both the good and the bad around the topic he cares about. This will enable him to make an educated decision about his theme. For example if he wants to work against discrimination based on race, gender, religion etc. he needs to read what are the kinds of discrimination certain people may face. To read he may search and downloads from internet such material which are racist and biased against some gender or religion.If this kind of internet search goes into an algorithm, he himself will be treated as discriminatory due to bias created after considering only one aspect of his behavior. The controllers of the Algorithm is unlikely to ask the reasoning from him, so this “misconception” will not get cleared. The problem will be multiplied if the information is used by a third party for targeting news or product to him or if this information is made public during some lawsuit. He may himself get clandestinely discriminated against. Capturing of data of internet usage or social media usage needs to be broadly regulated. Sharing of person specific data should be made exception rather than the rule. Specific and detailed regulations like EUGDPR does not ultimately help consumers and common people. Although awareness about data privacy has increased, the legal clauses for large corporations force consumers to accept these clauses to use the services, so life of consumers actually become difficult. The solution to me seems to be broad mandatory directives making it transparent as to which business models are allowed and which not. The obvious problem with that is the regulators and activists often have less technical and financial resources that the large corporations resulting in them being always a step behind. https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/966862264 2020-11-28 09:46:11 UTC https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/966862264 Telling Stories https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/968083681 My first card was on the subject of FAKE NEWS. Apart from the pavlovian shudder I experienced owing to the overuse of this phrase by a certain person, it's an interesting area where algorithms used to promote news stories were doing so in a way that amplified 'fake' news - untrue stories that were submitted to Facebook. 

The data is CRIMINAL RECORD, the value AESTHETICS, and the rights PATIENT RIGHTS. It's a slightly difficult mix, but this is how I interpreted them.

It is not uncommon for healthcare professionals to share their experiences on social media. Whilst there are clear guidelines on what should, and should not, be shared, it is easy for some people to share in a way which might compromise patient anonymity (see RIGHTS). IN a world driven by likes and retweets, it can be easy for those posting to embellish the facts, or even completely fabricate stories, under the guise of artistic or creative expression, (values: aesthetics). This can then be subject to the unseen algoriothmic hand and drive promotion.In this case, the data may be a patient's criminal history woven into the clinical story, so a paramedic may report on tending to  the injury of a person in police custody, or on the street after a brawl, or a drug worker may write a blog post on an inexpertly anonymised patient story in jail. 

The posting behaviour of healthcare workers has been posted this week in the BMJ in an editorial piece: "Should we bring the curtain down on NHS social media performances?" https://www.bmj.com/content/371/bmj.m4497 

The author suggests that more care is taken by those who may bring the profession into disrepute through sharing 'fun' activities, but the sharing of patient stories is also mentioned. 

A solution here should in part lie with greater education and training for healthcare staff, careful enforcement of the guidelines and employment contractual restrictions, and possibly even algorithmic surveillance of all HCPs posts - although this raises a whole new range of ethical concerns!]]> 2020-11-29 08:43:20 UTC https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/968083681 Algorithmic Beauty Contest https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/982471083 My cards were:
- Example: Algorithmic Beauty Contest
- Data: Location
- Values: Altruism
- Rights: Human Rights

When looking at location-related data, we see that areas with higher economic resources might allow for people to spend more time on their physical appearance (fitness, beauty products, healthier food, etc.) then in the case of lower income regions where people might be busy gathering the necessities of life for their family.
Based on the values of altruism, diversity is one aspect which that algorithm obviously didn't take into account.
In terms of Human Rights, an important right is the right to non-discrimination, meaning that being physically different shouldn't impede an evaluation of someone's beauty (whatever that words actually mean).

I also think that the concept of beauty and fashion all over the world have been corrupted by a Western-centric mindset where many ethnicities still perceive a lighter skin tone as an element of beauty based on older values where richer people did not have to do back-breaking work in the fields, scorched by the sun. Many aspects of Asian beauty for example, includes a lighter skin tone which would lead the algorithm into thinking that "light color sking = more beautiful than darker skin" hence the results.

As a solution, I think there should be a focus on the fact that "diversity = beauty" which would lead to the algorithm in finding winners that are "beautiful" but also different from each other. Also, I would include in the diversity aspect not only a diversity of ethnicity but also of age, in order to reduce agism and train the algorithm in understanding that some older people, based on their body language can also be beautiful.]]> 2020-12-03 03:40:08 UTC https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/982471083 Algorithmic Justice https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/989000520 I picked Algorithmic Justice as the ethical problem, Financial Records as the Data implications, Security as the Value effect, and Human Rights as the Rights impact.

The risk assessment algorithm COMPAS overestimated the risk of black and Latino offenders when compared to their white counterparts resulting in possible higher bail amounts and sentencing terms. The algorithm amplified an injustice in the historical/training data in ways that will negative impact the victims’ financial records such as credit score and their ability to secure loans under a misguided and distorted view of the value Law and Order when in fact the algorithm was exacting the opposite by violating the human right of fair trail and the right to non-discrimination of the victims.   

]]> 2020-12-04 19:40:26 UTC https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/989000520 Algorithmic Altruism https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/992020703 I picked up the hate speech and cyberbullying as the ethical problem. Following by personal information by the data card with altruism values; and lastly, Human Rights chosen as the right card. 
One ethical problem would be someone accessing to sensitive data and selling them or spreading them for reasons including: hate, envy, contempt… and of course without consent. Automatically, leading to a hate speech -often in social media- creating insecurities, mental breakdowns to the victim. In this case, his/her human rights have been violated. Thus, the main solution here is education, educated society according to certain values such as the right to respect for privacy life; equality; justice; helping others; caring; empathy; morality; compassion; betterment of humanity. 

]]> 2020-12-06 19:15:00 UTC https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/992020703 Algorithmic beauty https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/995075838 I chose 
Example - beauty contest. 
Data - financial. 
Values - Affiliation and belonging. 
Rights - for patients.
I chose my cards using a randomn number generator, which is an algorithm itself. The potential for data to show bias for patients has been mentioned already, if bias data is fed in as in the beauty contest. white skin might become a proxy for health and better financial performance, leading to discrimination in job market for example. the AI might find correlations in the bias data and therefore exagerate bias. In Asian countries it has been reported that lighter skin signals higher status from not having to do work in sun - would this preference be found in affiliation? would certain groups search for skin whitening products? what would the AI do if it found this correlation and how should developers react this? 


]]> 2020-12-07 17:30:12 UTC https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/995075838 Reputational backlash https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/995922077 I decided to to pick the following cards. 
• Reputational backlash.
• Human rights
• Power
• Personal information 

I have always been concerned with the saying that the internet is always right. I believe in free speech  and differing opinions and fully support those concepts. But when righting arrivals, Opers, social medial post people should state that the following words are their opinions or briefs and not stated as facts. People hide behind free speech and take and have very little accountability for what they say. While I may not like Facebook they have the resources and abilities to deal with fake news with staffs to investigate, ability to remove the the words and or the use of legal staffs. Normal people that can get hurt by malicious wrongful information generally have to bare the consequences. 

]]> 2020-12-07 20:26:01 UTC https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/995922077 Automatic Re-education https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/1005134240 The problems is that moderation of online forums have become automated. This could lead to censorship. Some data collected that could influence this includes what a person consumes, buys, how often, which websites they use, in addition to what they do in their free time – where, with whom, how often. We consider values such as a respect of science – innovation, objectivity, evidence. We must also consider people’s rights, which could include age, race, religion, belief, etc. 

One can imagine a situation in which a person is automatically discriminated against by an algorithm, based on their consumption practices and leisure activities. But that this was unfair, at least in part because this person took certain decisions and completed activities in the pursuit of knowledge and interest.

To imagine a specific example, let us think of China, with their increasingly prevalent social credit system. This credit system attempts to assert a certain set of values. These values seem to be in conflict with many subcultures that live towards the fringes of Chinese society, such as the Uighur muslims. Indeed, although the news out of China is nebulous and patchy, there have been accusations of mass imprisonment and re-education camps. Could the decision to imprison and re-educate these people be due to data collected on them? Even just where they live, what religion they express, what clubs, groups, organisations they have been a part of. Perhaps an algorithm detects that anyone who goes to a certain community building at 7pm on a Saturday is more likely to be radical. This leads to a raid at a book club at that building, where everyone who was reading Atlas Shrugged that month gets re-educated. 

Though these cards were chosen randomly, it’s interesting how easy it is to imagine specific examples related to these criteria. 

]]> 2020-12-10 06:23:04 UTC https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/1005134240 Personalisation/ Otherfication https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/1012710028 I chose the example of Personalisation. Then I chose:
Rights: Human Rights
Value: Affiliation & Belonging
Data: Personal Information

This is a subject I have discussed with several people including colleagues that work in Tech, Digital Marketing, and other related fields but also with friends & family that approach Digital Platforms as users. The main problem I see here is that while I understand the needs for people to have a feeling of belonging when this respectful value is miss-used by corporations and governments to create silos to later target them-either to make them buy products or sell them political ideas, influence their vote on legislation or other socially-relevant matters- we are not only allowing an unethical use of the personal data collected via dubious mechanisms but also we are generating a society that can easily be triggered to see everyone outside their group as 'the others'. History has shown how dangerous otherfication can be and how this otherfication has justified theoretically mature, ethically solid and democratic societies to commit atrocities. I often joke with my 'social platform addicted friends' by telling them 'One day you follow influencers and friends that you like because they think like you and they are hype and the other you end up voting for Trump'. It is a joke that seems extreme as most of my friends are 'lefties' but it is to show that when you do not receive information or input from outside your small bubble it is really easy to sell you a specific pair of sneakers but also to influence (and even change) your vision of the world.


]]> 2020-12-12 20:32:28 UTC https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/1012710028 Historical Bias https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/1013881336 As we have seen, AI can be as good as data and people are the o es to create algorithms. There are números human biases, conscient and inconsciente.
Therefore, it may not be posible to have complétele unbiased. But human and human-made algorithms can checo the data to identity and remove biases.

One solution in order to build trust and to minimize bias is to create data and algorithm tests. In a capitalist world and with the digital world monopolized, the way is regulation. That the GDPR requires control at all stages from data capture to the creation and application of algorithms to test bias minimization.]]> 2020-12-13 17:30:54 UTC https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/1013881336 Clearly, the image search results insinuate that black women have unprofessional hair! Just because of their curly pattern. I have examples from poorly-written “dress code” manuals in large multi-national companies, where special reference is made to the hair and in particular, provide guidance on how the hair should be processed to adhere to a professional style. We can now imagine that this might be a reason for an unfair job termination and even without compensation! A clear sign of discrimination and lack of any of the commonly agreed values like equality, diversity etc!I could imagine also that this kind of appearance would affect someone’s credit score and access to loans and other social benefits or contributions. https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/1030849945 2020-12-18 10:55:06 UTC https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/1030849945 https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/1030850659 2020-12-18 10:55:36 UTC https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/1030850659 Algorithms and racism https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/1032775709 Although I sufficiently mixed up and selected my cards randomly, most seemed to mesh well. These were the cards I picked:

Example: Can Algorithms be racist?
Rights: Human Rights
Values: Science & Knowledge
Data: Internet Search

Questions asked from the Ethics Example was, If a search result appears biased towards or against a particular group, is this simply a reflection of attitudes that exist across the internet? or do these kinds of search results reinforce discriminatory attitudes and make societal problems like racism worse?

This is a bit like asking, which came first, the chicken or the egg. So my initial answer was going to state the latter, biased data makes societal problems like racism worse. We've discussed sentencing on this course, longer sentences for certain races potentially. LFR for example looking at one race more than others and not to mention other police systems, which are biased. Algorithms are based on input data and if the input data is biased, it'll undoubtedly make societal problems worse as these biased algorithms are used in real life situations. Even location data can have indirectly racist results, if for example algorithms are used to grant credit and those from lower socioeconomic backgrounds are associated with certain locations (and these people come from a particular race) the people might unfairly be denied certain credit, rather than this process being meritocratic.

In terms of human rights, there is the right to non-discrimination, if algorithms are "racist" this fundamentally clashes with basic human rights law. Additionally, what about things like the right to property? If an algorithm denies you a mortgage based on bias, rather than actual merit, is that another breach of human rights law? What about sentencing, as I mentioned above and the right to a fair trial. The reason I mentioned the chicken and the egg thing before, is because if this biased data is used - legacy data, it will no doubt seep into society and it can inflame certain groups who then go onto the internet. It is a destructive cycle and then you potentially go into the territory if these groups terrorise someone into the "right not to be tortured or treated inhumanely" or the "right to protection of property" (attacks, theft and vandalism). Obviously I am being very dystopian here and thinking of the worst case, but it is a consideration.

In terms of Data and Values, search terms, search results, clicked links are all related to the Example, you search for a term and it gives you biased results, you click on that link and it may subconsciously influence your mind. In terms of our attraction to people, how much of this is influenced by marketing/hollywood/bollywood and whatever else we're consistently exposed to? Exactly! What I mean by this, is if you are consistently exposed to bias through your searches, can you become more discriminatory subconsciously? These things need to be addressed and as per the Values card, "understanding how things work" is the key to helping to design algorithms, which have more "objectivity" this is not something easy to do as the bias in legacy data is so widespread/entrenched that to displace it,  takes a lot of effort, unless you work on a controlled sample and even then that will slow down the progression of an algorithm. You may even accidentally programme bias into the controlled sample. So this is certainly a "challenge" and is "complex." The card mentions discovery of new knowledge, invention and innovation and learning. What I would add to this card is also representation, trying to diversify the process of creating algorithms, so that biases can be dealt with at the design stage.



]]> 2020-12-18 22:17:26 UTC https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/1032775709 Beauty Bias - The eye of the beholder https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/1032813643 Values: aesthetics There is something here around the algorithm re-inforcing a stereotype of beauty. Is a consequence of AI a homogenising of aesthetics to a vanilla bland version? It seems incredible that surfacing from diversity and complexity an algorithm can give the impression that beauty is reducible to this apparent truth. A bigger issue for me is how this then goes on to shape and homogenise concepts of beauty

Rights - Equality - this absolutely feeds into equality rights. There is a value judgement here that lighter is better.  This has a direct impact on darker skinned people who in this case are excluded from winning this competition for not being beautiful 

Data - It wasn't clear in the example how the data was selected and how the AI was trained to identify beauty? Was it faces linked to words denoting beauty for example?

Solution 1 : We'd need to understand much more about the schema that was used to train the AI.  For example you could increase the data set, analyse the diversity of sources from which the samples are drawn. There is also something about the definition of beauty.

Solution 2 : Just don't run beauty competitions and objectify people! Beauty is in the eye of the beholder ]]> 2020-12-18 22:51:34 UTC https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/1032813643 Historical bias https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/1033509586 Should be developed exemption algorithm, or should not be used at all. ]]> 2020-12-19 16:56:55 UTC https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/1033509586 Fake news, personal information data, affiliation and belonging, and data protection rights. https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/1033741276
  • Problem: The Facebook algorithm which curates news articles on the Trending list is allowing fake news to spread
  • Personal information data: This data can be used to infer someone's political leanings. Pew Research found that Black women, white college-educated women, millennial women, hispanic catholics, and unreligious people tend to lean Democrat, while white evangelicals, white non-college educated men, rural southerners, and Gen X men tend to lean Republican. So algorithm may choose Republican-favoring news if its user base is made up of those likely leaning Republican.
  • Values affiliation & belonging: Facebook is a place where users seek out people who care about the same things and will validate their opinions. This means they are less concerned with facts, and more concerned with their point of view being accepted. This could perpetuate the spread of fake news, because users are likely to share what they agree with whether or not it is true. Since the need for belonging is not political, this behavior is non-partisan, but an algorithm that is bi-partisan based on user demographics, will soon turn promote one perspective and ultimately that party will have more power to spread fake news that works in their favor.
  • Data protection rights: Personal data can be used to target someone, and in the case of fake news, you could determine someone's education level by comparing personal information to their social media activity and know who might be willing to believe and spread fake news. This again could perpetuate the spread of fake news, if a person who is likely to believe fake news is shown "trending" articles that are fake.
  • Solution: Add a fact-checking team to the news department at Facebook who can look at the trending articles the algorithm has surface and remove the ones that can't be fact checked.
  • Solution: Create trending headlines instead of articles. Since factual news is usually reported on across all major news sources, this wouldn't allow one-off inflammatory articles to be surfaced. The algorithm could see several articles from ABC, MSN, CNN, etc, all saying the Moderna vaccine is approved by the CDC, and collect all of those articles under one headline. Then an article that said the Moderna vaccine contains a microchip couldn't be surfaced as it only exists on a single fringe news site.
  • Solution: White list news sources, so that the algorithm only promotes articles from these sources.

    • ]]>     2020-12-19 21:48:56 UTC https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/1033741276     Algorithmic Justice https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/1034473900 The ethical problem selected is algorithmic justice, ethnic identity as the data implication, Security (law and order, certainty and predictability) as the value effect and human rights as the rights impact ]]> 2020-12-20 16:39:27 UTC https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/1034473900 Echo chamber on commercial social media amplifies homogeneity https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/1036908635 The Personalisation is picked as an example for 3 reasons:
    1. Personalisation. On Facebook or Twitter (also other proprietary platforms), political deliberations are shaped by algorithmic codes which are trained by... 
    2. Data. ...end-users data (icl. clicks, likes, shares, their friends and followers), which are profiled, packaged, sold or shared with social quantification industry and/or law enforcement as well as the intelligence community. 
    3. Values. Similar political views drawn from end-users as well as those pages they visit are sorted, proritised and shown to end-users by algorithmic NewsFeed feature, e.g. FaceBook EdgeRank. As a results, end-users are limited to see the reality of their news feed.
    4.The problem of Personalisation violates freedom of expression, b/c algorithmic systems suppress whole information on end-user platforms in spite the fact that end-users work for these firms for free. End-users also face their data protection infringement as they are not sufficiently and explicitly informed about their data being used or shared with third parties. As seen in the political ads scandal involved with the Cambridge Analytica, it may violate enc-user's right to data protection as their data is shared to campaigners who may manipulate it for political purpose.

    ]]>     2020-12-21 23:06:30 UTC https://padlet.com/moocdeliveryteam/jbgsmug2e9xxwz05/wish/1036908635