Jeremy Pitt has taken a lead from an unusual source in his work on ethical algorithms. He says it has been inspired partly by the work of the late Elinor Ostrom, a political and economic scientist who won the Nobel Prize in 2009 for her studies on the self-governance of common resources.
She observed that people can preserve a shared resource such as water or forestry with a set of rules to regulate their own behaviour for the common good, and went on to formulate eight design principles for the institutions that manage the resource. They include clearly-defined boundaries, cheap and accessible mechanisms for conflict resolution, effective monitoring and graduated sanctions on anyone who violates the rules.
Pitt, who is deputy head of the Intelligent Systems and Networks Group at Imperial College, was already thinking of a related issue. The emergence of cognitive computing systems has stirred up fears in some quarters that they could remove a human sense of justice from decision-making. Pitt began to look at the possibility of introducing ideas of fair play into algorithms, so they could make judgements on human needs and the common good to help solve big societal problems, and has been using Ostrom's eight principles as one of its parameters.
"When we build these systems we have rules for each of these principles, so the electronic systems with ethical outcomes will hopefully replicate her results so we can have an enduring institution," he says.
He has been linking this with the notion of 'design contractualism' – building in the rules that govern human behaviour for the collective good – in an effort to lay the ground for ethical algorithms.
"This has been an ongoing process for three years now, and artificial intelligence has made available techniques that are very well suited to the task," Pitt says. "The key insight was to see how Ostrom's principles could be expressed as protocols to be represented in event calculus, a computer language for reasoning about actions and events. There are specific events in the system that create a narrative that can be processed according to the rules we specify.
"It is complicated to translate these principles from natural language to formal rules, but we have made fairly substantial progress."
Pitt's team have run a series of experiments, the findings of which have been published in academic papers, to show how Ostrom's principles can be used in creating ethical algorithms.
"We have a version of event calculus that is compatible computationally with all the event recognition systems out there," he says. "You can use it to process tens of thousands of events per second."
The ultimate goal is to produce algorithms that can ensure decisions in areas such as resource allocation and dispute mediation are based on a sense of ethics and are seen to be so.
The next move is to incorporate this into a platform for digital institutions being developed by the Institute for Data Driven Design in Massachusetts, which Pitt sees as a step towards building "industrial strength applications". There are also experiments on how the algorithms could work within a smart grid, and an Autonomic Power System project, led by the University of Strathclyde, for the future management of electricity supplies.
The latter provides an indication of how ethical algorithms could work. Pitt says that by 2050 we will have to be a lot smarter in managing energy, making use of systems that manage and heal themselves. Cognitive computing is moving in this direction, but in a field such as this supplies have to be allocated for the common good.
"This is where all our ideas of algorithmic self-governance and fairness in resource distribution become really important," he says.