ACS and Big Data ethics - when the best is yet to be
TO extrapolate how far banks and other financial institutions could go with Big Data, one senior DBS executive once illustrated it to this reporter this way: If data analytics predicted that bankers who were former ACS students were more effective at targeting customers who were once their schoolmates, banks could, in theory, deliberately hire ex-ACS students and define sales targets such that they go after other ex-ACS boys - or customers from segments of society or "particular addresses".
Just to be clear, DBS does not do this. But the illustration shows the dangers of technology creating a more reductionist society, one that more narrowly defines and quantifies the worth of each human being. As Singapore looks to unwind the stigma of an average PSLE result, it is also facing a brave new world of Big Data that could run amok. In that world, those three digits may form a tattoo that brands one's trajectory in life from age 12.
Such a reading has dystopian shades. But regulators have stepped up the debate on the ethical pitfalls of both Big Data and artificial intelligence (AI). The Monetary Authority of Singapore (MAS) has begun its work in this area. The Business Times reported in November that MAS has engaged the financial industry on ways to form regulatory guidelines in this space.
MAS is opening up the biggest discussion of our time, though it's confined to the finance industry. It is joining some of its regulator peers in considering how innovation in Big Data and AI, while beneficial in breaking down barriers to finance, can also build new walls. Questions have also been raised about the impact of predictive data on medical insurance. There are worries that Big Data and AI would change the methodology for risk pooling in the area of healthcare protection in a way that exacerbates financial exclusion.
The Council for Big Data, Ethics and Society, a global organisation which boasts several professors from top universities and data researchers as members, has laid out some thoughts on the ensuing debate.
Spurious links
In a white paper published online last year, the council noted that the fast-paced roll-out of Big Data has "bulldozed thoughtful consideration of bias, statistical meaning, and grounded interpretation, in relating any set of data to any other set of data". The study pointed to a website Spurious Correlations, which showed data correlating the number of people drowned in pools to the number of Nicolas Cage films.
In avoiding spurious correlations, corporations have to secure enough relevant data. But to allow corporations to capture so much data, how should regulators calibrate consent?
The philosophy behind consent should be that consumers must have enough information about the intent of data use, and the risks behind granting permission of the use of that data to businesses seeking that consent. But is it enough to have a one-time, sweeping, consent to access of personal data when it is not possible to capture all the benefits and risks of consent from the get-go?
Consumers may have to constantly update their consent on the use of their data according to the change in the risk context. If so, it is unlikely that self-regulation alone by large corporations that are profiting from the data would prompt them to keep consumers informed.
The question of consent is also relevant in the realm of AI. Oren Etzioni, the chief executive of the Allen Institute for Artificial Intelligence, has suggested updates to Isaac Asimov's three laws of robotics. He proposed that an AI system cannot retain or disclose confidential information without explicit approval from the source.
Amid the rising use of smart home devices such as voice assistants, TVs, and vacuum cleaners, reports suggest that transcripts of one's conversations with voice-activated personal assistants could be given to third-party developers. Even if that information is anonymised in theory, a third party holding enough data to cross-reference data sets can in effect pinpoint one's identity. With that in mind, should corporations hold intimate information about room maps and bedroom talk?
Dilemmas
Meanwhile, the advent of self-driving cars has already raised a question of how such vehicles would respond to the seminal trolley-problem - the thought experiment on the ethics of killing one person to save five.
Germany has tried to address this. Its regulators drafted the world's first set of ethical guidelines for self-driving cars this year, and noted that the AI must be designed in such a way that dilemmas in which an automated vehicle has to decide between two "evils" are avoided in the first place.
But the regulators also accepted that there can be no clear-cut way to anticipate the decision of a human driver who should have the moral capacity to make the right judgements.
Therein lies the discomfort that AI as powered by Big Data also reflects the fundamental biases of humans. German regulators pointed out that where accidents become unavoidable, "any distinction based on personal features such as age, gender, physical or mental constitution, is strictly prohibited". They see a need to police potential human discrimination manifesting through technology.
Likewise, Microsoft's AI bot Tay was shut down less than a day after its debut amid criticisms. Tay's interactions with bigoted Twitter users led it to retaliate with bigotry of its own.
What will we do in a world run by ACS boys making more ACS boys richer? What will we do in a world that could quietly determine the value of human life through split-second algorithmic calculations? Can we rein in the social contagion of prejudice?
To consider the ethics of Big Data is to confront social divides, information asymmetry in market-driven economies and the brutishness of human nature. The future of AI and Big Data may come down to just how much of ourselves we can bear.
