The World Wide Web: Past, Present and Future by Tim Berners-Lee
I found this text provided an interesting insight into the origins of the internet and the intended use of it. It is clear that the main intention behind the internet was to act as a medium for communication and information interchange for high value items and it was interesting comparing this to how it is used today. This source will provide some useful context and background to my essay, especially with it being written by the creator of the World Wide Web itself. However it does not specifically address my hypothesis.
The tech titans must have their monopoly broken – and this is how we do it by Vince Cable
This news article outlines the problem created by the monopoly between the tech giants, as well as the ways in which it can be overcome. Through my research, I have identified that the problem of algorithmic decision-making is heightened by the fact that there are so few major players within the market for social media platforms. This article provided a useful discussion into what the shortcomings of social media are, and I would like to discuss this within my project as it illustrates why the problem may be significantly worsened. I thought it was interesting that the article picked up upon social media being a carrier of false information after having read Tim Berners-Lee article on the intended use of the Internet (discussed above) as being a place where information is verified and of high-quality.
Having read this article has raised for me more questions about who is to blame. Clearly, citizens are targeted and victimised by this spread of false information combined with a need for staying connected, resulting in their data being compromised. This can then go on to impact families more materially, in line with my hypothesis. However, is it the companies to blame for spreading the false information? Or is it the use of the algorithms that prey upon people’s vulnerabilities and place them into categories to determine their chances of successes and failure that are to name? This article has given me some food for thought, which will make for some interesting evaluation within my project.
Is ‘Big Data’ Actually Reinforcing Social Inequalities? by Michelle Chen
In this article, Chen recognises that though big data can enable insight into situations, it can also be used as a tool of control. As an example, Chen talks about the subprime lending crisis in which those from poorer suburbs were exploited with subprime loans. Further, she highlights how although algorithmic decision-making is seen to be objective, it in fact ‘can reinforce and mask prejudice’. Through reading this article, it seems that this is not a desired, purposefully sought-after consequence; rather, it happens as a result of the algorithms identifying an underlying pattern which might mean that those of a certain race and class are profitable. However this is obviously unacceptable as it will only go on to feed further systemic discrimination. We cannot assume that the output of these algorithms will be unbiased, as If the algorithm is trained on biased data that we generate, the algorithm inherits that discrimination and thus exacerbates the problem.
Researcher David Robinson provides a very concise description as to why this happens, which I found really helped me get to grips with why this bias occurs. Essentially, the data that the algorithms use to make decisions can be used to infer more sensitive details. Moreover, they often use the current and past environments to inform these decisions. Therefore, using algorithmic decision-making can, in the worst instances, automate and perpetuate already existing hostility and inequality. Neil Richards warns that if we don’t recognise the importance of maintaining confidentiality, Big Data can be more dangerous than useful for society, acting as an ‘instrument of repression’. The article picks up upon how the fact that the digital world is so essential in everything we do nowadays means that we have to give up our personal sovereignty for the sake of being able to utilise the resources it provides. This is especially worse for those that are poor, who experience being more restricted by employment opportunities, contrary to what the internet was designed to enable. I found this article thought-provoking, exploring the uses of big data from both points and crucially highlighting how the age we live in calls for new and more persistent efforts to introduce regulations and maintain privacy and transparency.
Bias detectives: the researchers striving to make algorithms fair by Rachel Courtland
Using a real story about how automated tools are used to make ‘predictive policing’, this article highlights how decisions made by algorithms can be life-changing and potentially devastating. As examples, families are unfairly investigated for child abuse and there is a concern that these systems are only worsening the problems that they are supposed to simplifying. However, digital activists such as Kate Crawford and Cathy O’Neil are fighting to make software more transparent and accountable. This article highlights some of the progress we have made. Emmanuel Macron announced in 2018 that algorithms used by the French government would be made open, while the EU has introduced the GDPR which are both steps in the right direction.
I found the discussion about the trade-off between fairness and accuracy especially fascinating as it provided a more technical insight as to what is truly going on, helping me to understand the complexity of training algorithms with numerous data points. The article looks closely at COMPAS, which ProPublica had reported to be biased against black defendants. (https://www.theatlantic.com/technology/archive/2018/01/equivant-compas-algorithm/550646/) I have read the article about ProPublica, so this provided a useful and more recent alternative viewpoint on the case. The issue was that black defendants were disproportionately more likely to be flagged up as likely to commit another crime, but did not go on to do so (false positives). Essentially, there must be some level of trade-off between fairness: it is impossible to have the same level of false-positive error rates whilst maintaining predictive parity, which would skew the results another way. The differences in the false-positive rates occurs due to the differences in base rates, so it is possible that there is another reason for the imbalance – such as a certain group being more targeted for arrest to begin with. Allegheny’s tool to predict child abuse has faced the same criticism, with is acting on biased inputs i.e. poorer families are more likely to be reported meaning they are unfairly scrutinised. This suggests the fault is inherently in our own biases and prejudices which we build into the system.
Although some of the content of this article repeated what I’ve encountered in my past research, it also provided a more optimistic (and recent) outlook as to how we are seeing change happen which I was not aware of before. I think the media industry tend to exaggerate the harms and negatives of technology, seeing it as something that we have no control over whereas this article provided a more balanced discussion as to both the pros and cons of harnessing algorithms for decision-making. I would like to take this into perspective while discussing my hypothesis, recognising where we are making positive changes and where there is still room for improvement.
How payday lenders profit from our psychological vulnerabilities by Molly Crockett
Though a more dated article, I found this an article to be an interesting read from a more psychological perspective as to how and why financial lenders prey upon our vulnerabilities. Crockett explores the idea that during times of hardship, consumers are likely to make impulsive decisions due to stress. Studies carried out have shown that poorer subjects were more likely to over-borrow meaning their future savings were depleted.
The writer suggests that the best way to prevent people from making bad decisions is to prevent them from being exposed to them to begin with. I know from my research so far that Google has adopted a policy of only displaying ads from payday lenders with sensible interest rates, which indicates a step in the right direction. Although this article included lots of details about the psychological working of the brain which will not be relevant to my project, I think the insight it provides can crucially help legal bodies recognise the importance of regulation and shape a future where our vulnerabilities are not exploited.
Artificial Intelligence Has a Bias Problem, and It’s Our Fault by Ben Dickson
This article explores the origins of algorithmic bias and crucially, how deep-learning does not equate to neutrality. I think this did repeat a bit of the research I have already done, but also explored the more technical aspect of it which I found insightful and helped me get to grips with understanding why inequalities occur. Professor Venkatesh Saligrama says that algorithms have ‘deterministic functionality’ which means they inherit biases from the data sets they are trained upon. For example, word-embedding algorithms trained on news and Wikipedia articles associated men with programming related jobs. This suggests that algorithms only replicate the stereotypes within our society; they do not correct them. It also provided what I found to be some truly shocking examples of where AI algorithms have discriminated between people.
However I do not think the luddite approach of being afraid of technology is the solution as there is no doubt that there are some areas in which these predictive algorithms make our lives easier. A popular example is Netflix’s algorithm to predict shows that you will enjoy. Again, however, this will never be entirely accurate as is the case with all of these predictive algorithms. In some instances, obviously, the consequences of this imbalance can be the cause of discrimination and unfairness. The question is how we can build upon what has been achieved in the way of machine learning to improve what we have already. Perhaps this calls for the use of separate algorithms for different groups of people to prevent the reinforcement of already existing bias. For example, within recruitment within the tech industry, it may be more beneficial to evaluate women based on specific criteria separately to men to prevent employment-determining algorithms associating gender (and other inequality-reinforcing inputs) with success. Many data scientists argue that the flaw is in our society (highlighting the issue of ‘GIGO’ which I discussed previously) and it is no surprise that these algorithms produce one-sided results which, given my research so far, seems fair. I think this will make a useful discussion point as it shows that getting rid of these processes altogether is not the solution.
How Cambridge Analytica’s Facebook Targeting Model Really Worked – According To The Person Who Built It
My Facebook Was Breached by Cambridge Analytica. Was Yours?
Researching into the scandal involving Cambridge Analytica, I was interested in hearing different viewpoints on the story to be able to make a thorough and reasoned judgement in terms of its implications. The details of the sources outlined above were useful in helping me to understand the story. Amid all the outcry about Facebook’s data breach, I was interested in finding out about how Cambridge Analytica used data to sway voters, particularly how effective it is. I think it is one thing to identify the potential damaging consequences of algorithms used for decision-making and an entirely different thing for them to have had serious effect which this article by Matthew Hindman helped to shed light on. Aleksandr Kogan, who was the researcher behind the personality test app, talked about how the whole process worked which gives a useful insight into the truth behind the media hype.
In essence, Kogan’s model was designed to predict party affiliation and personality types by using public information about likes, friends etc. It uses the ‘Big Five’ model to predict behaviour and personality types and thus tailor adverts. Such models have been designed in the past, and as a previous article I read identified, are fundamentally the basis of how marketing works. The difference here is that they are being used supposedly to ‘manipulate’ elections which seems considerably more serious. I found this article useful from a technical viewpoint, and I’ve noted that though the model was assumed to ‘work’, in reality it is difficult to tell the extent of its success. Regardless, this incident is a signal to government bodies to recognise how embedded the online world is in our daily lives such that it can be used as a tool to spread fake news and exploiting people’s weaknesses. In a way, I think this resembles the subprime crisis in that it capitalises on people’s vulnerabilities.
Facebook has gotten so big that no one can understand it, and it could be a good reason to break it up
Google and Facebook grab ad growth
I understand through my research so far that the problem posed by algorithmic decision-making is heightened by the existence of so few of these ‘tech giants’ that effectively have control over the data of millions of people. Facebook, for example, owns Messenger, recently bought Instagram and Whatsapp and even tried to buy Snapchat. This is made more clear by the chart from Statista above, highlighting how ¼ of the money spent on ads goes to Google and Facebook. The lack of competition in the market for social media platforms gives these few firms the power to effectively do what they like as there are not many alternatives – and the ones they exist are owned by the same few companies. This particular article from the Business Insider suggests breaking up the company, with it possessing so many qualities of a monopoly. Though this may subdue the issue, I am sceptical about whether it would at all subdue the issue of socioeconomic disadvantage, which is what my research is on. This is because of the issues outlined I have picked up upon through my previous pieces of research – i.e. biases within our society, different base rates etc. Regardless, I think more healthy competition within the market would mean that firms would perhaps respect users’ rights more. Whether and how that would happen is a different question.
IBM uses big data to predict outbreaks of dengue fever and malaria
How Can Big Data Technology Help Fight Poverty?
Big Data and Human Resources—Letting the Computer Decide?
How Big Data and AI are Driving Business Innovation
I was interested in looking at the ways big data combined with algorithmic decision-making has benefitted our society to be able to make a judgement about whether, on the whole, it has done more good than bad.
The first article looked at a project conducted in 2013 by IBM where they successfully were able to predict outbreaks of dengue and malaria. With the use of the cloud becoming so easy now, real-time data can be analysed quickly to inform action which can potentially help save lives. The way analytics can revolutionise healthcare is something that very much excited me, but although not discussed in this article specifically, I think these systems would be especially beneficial in poorer countries. The problem is that health records are highly sensitive and storing them would be a challenge especially with the presence of corrupt governments or private healthcare.
In Borgen Magazine, Wang describes how large data sets have been used to forecast weather patterns to help with farm planning, which will obviously go on to have resounding, positive effects across communities. A different set of data is being used in West African nations to predict Ebola outbreaks.
The third source looked at the use of big data more critically, discussing the different ways in which big data can be used within HR – from recruitment through to workforce management and how implementing these methods can maximise efficiency. Although I’ve looked at these uses before, this particular source weighs up both sides more equally. However this was from a more legal perspective in terms of laws and rights surrounding data collection in different countries which I did not think was especially relevant to my hypothesis.
These sources showed me the different ways in which data can be harnessed to benefit society which I will need to weigh against its shortcomings in my essay to address my hypothesis. I would especially like to address the feasibility of some of the ideas discussed in these sources, with regards to the worsened scenarios that will result if they go wrong.