Riots, bias, IPR and questions on AI technology

Politics this week, in both US and the UK, has featured prominently some of the issues touched on in earlier blogs about AI technology. As always, things move so fast with this technology that politicians and, to a far greater extent, the public are left reacting haphazardly to its effects and uses. The lack of coherence and the serious issues at the heart of using this technology, therefore, demand constant thought and clarity lest we walk into an era of new norms that none of us remember choosing, far less agreeing with.

Social media and public information

With reference to the riots that took place in the aftermath of stabbings in Southport on 29 July 2024, Professor Andrew Chadwick, professor of political communication at Loughborough University, and an expert in the spread of online misinformation, has been quoted as saying that a ‘complicated mix‘ of ingredients converges to bring about the kind of heightened passions and public disorder seen in the wake of the stabbings. He was being asked about the role played by social media rumours that had spread the notion that the stabbings had been carried out by a migrant, possibly a Muslim or Syrian person.

Professor Chadwick emphasised, however, that it was not just the false name or the rumour around it but

the way these are used by public figures and interested parties that is equally important in causing the kind of surprising scenes of disorder witnessed this week.

This issue has been simmering for nearly a decade since information began to come to light about social media algorithms that promote controversial or emotive content much more aggressively than normal material, for commercial reasons. In March 2021, Karen Hao reported in MIT Technology Review that even after the fierce Cambridge Analytica scandal of 2018, in which personal data had been ‘surreptitiously siphoned‘ from millions of Facebook users to sway the 2016 US elections in Donald Trump’s favour, Facebook efforts at mitigation had not targeted the spread of misinformation via their platform.

Instead their efforts had been focussed much more on addressing bias against disadvantaged groups on their platform. While in itself a laudable objective, the better-targeting of marginalised groups actually serves to promote the business aims of social media platforms so it is difficult to attribute this initiative to pure altruism. Hao quotes Professor Hany Farid of University of California, Berkeley to make the point that the aim of business growth at all costs means ‘maximizing engagement’ trumps considerations of veracity and truth. In fact, Professor Farid suggests that ‘harm, divisiveness, conspiracy’ tend to become the friends of businesses in these circumstances by bringing them more subscribers.

With this information in their hands, and social media becoming the major source of socio-political communications, politicians of dubious intent have revelled in placing incendiary gossip, innuendo and out-of-context diatribe on their accounts to gain political traction. The early promise of the internet giving a voice to the people, and thus promoting democracy, has become a far more mixed proposition in light of these developments. Ethical certainties of the past and the role of various players in their upkeep are often exhaustingly complicated now.

Ethical confusion and the need for a new business model

Ironically, while the rumours and misinformation regarding the Southport incidents in the UK spread through X, formerly Twitter, the owner of the platform was accusing Google algorithms in the US of obstructing the flow of information on Donald Trump, though other reporters did not always have the same search experience that he did.

With serious and frivolous, true and false, pleasant and atrocious all jumbled up in the same digital space, and competing for the individual’s attention at lightening speeds and voracious volumes, the ease with which misinformation can be unleashed amongst the public is alarming for supporters of democracy. In his 2016 book Homo Deus, Professor Yuval Noah Harari had pointed out how our private data is the most valuable asset in the digital universe which we end up giving away for free to Big Tech algorithms.

All those thousands of likes and shares on social media, as well as information about our relationships and interests, feeds the algorithms that enable large platforms to attract advertisers. What is more, this information helps build profiles that can potentially be used to exert subliminal control of users by ‘nudging’ them to buy more of what they like or, alternately, to help politicians understand ‘what each candidate needs to say [to each user] in order to tip the balance’ in their favour (Harari 2016, p. 397). This level of control of individual views and choices is pernicious to democratic societies because it subverts popular will, through unaccountable power and influence, for commercial or political gain.

A business model that prioritises economic growth alone, even at the expense of weightier, but often intangible, human imperatives poses a grave risk to societies valuing human freedom and well-being as non-negotiable values.

Facial recognition, bias and shifting boundaries of police authority

At the same time, the ethical dilemmas are deepened by the need of the authorities to gain control of the situation. Reacting to the riots in the wake of the Southport incident, the prime minister indicated that the government ‘might use these riots to sanction greater use of facial recognition cameras‘. This use of fire to fight fire, or technology to combat technology, may seem inevitable, however, much caution has been urged with regard to the use of facial recognition technology for law enforcement.

The use of this technology involves real-time comparison of facial biometrics of members of the public with facial biometrics of people on a police watchlist. Images of people captured through CCTV cameras are converted to numerical data and the data is then compared with similar numerical data for the images on the watchlist. Images that do not match are deleted instantaneously while those matched are then checked by a human operative before action is taken. In the case of Bridges v The South Wales Police in 2020, the judges noted that the South Wales Police had not failed in its efforts to ‘to equip itself with information on possible or potential disparate impacts, based on the information reasonably available at that time’, yet their judgement was that

‘too much discretion is currently left to individual police officers. It is not clear who can be placed on the watchlist nor is it clear that there are any criteria for determining where AFR [facial recognition technology] can be deployed.’

Thus, it is easy, even with the best intentions, for this technology, which one of the experts in the trial testified was often found to contain ‘bias on grounds of race or sex’, to be misused. Facial biometrics form part of a person’s private data and are thus protected under various legal provisions. Even ‘the Information Commissioners [have] advised there must be a “very high bar” for its use‘. Once again, the neutrality of the technology is only as good as the datasets that the underlying AI that does the work is trained on. Due to historic bias against certain minorities in particular, any datasets from historical policing records, have a good chance of passing on that bias to the AI technology.

Yet, as an expert witness to the Court noted ‘the precise makeup, scale and sources of the training data [for AI] used are commercially sensitive and cannot be released’. It is in this manner that intellectual property rules often make it difficult to ensure adequate transparency if bias is to be ruled out with complete certainty. The balance between profit-needs of commercial enterprises and public good seems tilted too far in favour of technology companies at the moment and is justified in the name of constant need for economic growth.

As we inch our way into a human environment clogged with AI applications – from those controlling Google searches, to those controlling our ‘user experiences’ of social media, or banking, or healthcare, or self-driving cars, or legal needs, or predicting our chances of getting cancer, or writing our essays, or drawing our pictures – the convenience and surface excitement often hide from view the impact of this exchange of breadth of information for depth, on human cognition and judgement.

As Henrik Skaug Sætra, Mark Coeckelbergh & John Danaher have warned,

‘…the challenges in question are non-trivial … the system … refers not just to individual AI companies, but to the larger eco-system of companies, social structures, and political arrangements that generate the negative impacts in question … You have to change the entire business model and the supporting legal–political infrastructure.’

Each time a government gives in to using AI technologies that further erode the privacy of the individual that has been hard-won after centuries of struggle, they strengthen that legal–political infrastructure that Sætra et al. are arguing against. The most vital challenge of our times is to stem this tide and then hope to turn it back.

Photo by The 77 Human Needs System on Unsplash