Doing the right thing: the ethics of AI

When 18-year-old Brisha Borden was arrested in 2014 for taking an unlocked bike and riding it down the street in Fort Lauderdale, she was assessed by a computer program in the federal jail as being at high risk of committing a future crime. A year before, 41-year-old Vernon Prater, a criminal with previous convictions, had been picked up while shoplifting US$86,000 worth of tools from a store. The same algorithm had rated Prater as being at low risk.

Two years later, while Borden had not been charged with any new crimes, Prater was back in prison. And the one crucial difference between the two cases, according to the non-profit investigative journalism organization ProPublica, which analyzed the statistics from the algorithm: Borden was black, and Prater white.¹ Although a person’s features, such as skin color or gender, may not be specifically disclosed, algorithms make correlations based on other characteristics, such as school or place of residence, as well as historical data which may already contain human bias.  But who’s responsible for inaccurate forecasting or prejudice in a case like this?

Where’s the accountability?

Take a moment to sit back, and questions spring to mind. If decisions are being taken by artificial intelligence (AI) machines – or by humans basing their decisions on the data produced by algorithms – then who’s responsible for the outcome? Is it the algorithm or the human programmer or the decision maker? In the UK in 2020, when more than a third of high-school leavers’ exam results were downgraded in England alone, was the AI software developer responsible for their shattered dreams, or the exam regulator, or the members of government that decided to use AI to predict students’ grades?²

If AI systems don’t have legal personalities, how can we define accountability in the future? And when systems need to make decisions that have ethical dimensions, whose moral compass are they using? AI developers, integrators, operators, regulators, and data providers must work together to resolve such questions around accountability and moral responsibility and find a responsible way of working with AI.

Predictive analytics vs. the moral compass

Smart AI technologies are prediction machines, using algorithms to analyze big data and learning over time to achieve better results. AI can analyze a limitless number of documents in a very short time; it can recognize and follow patterns (which may already contain human bias), anticipate behavior, and forecast probability. Precise predictive analytics are already helping to diagnose health issues and suggesting treatment; algorithms can make predictions about people who have committed crimes, like Brisha Borden and Vernon Prater; they can recommend whether or not an applicant is suitable to receive credit or an insurance policy. Faced with the probability of an accident, an autonomous car can “decide” probably quicker than a human driver how to react to protect lives.

But how does an algorithm assess the value of a life, using which principles and whose ethics? AI isn’t human; it mimics human behavior. At present, the use of AI without proper ethical standards in place to protect the violation of human dignity and rights carries the risk of projecting human bias, including racial or gender discrimination, into automated decision making.³ If we use machine learning and algorithms, we must apply our human empathy and human ethics to reach a final decision.⁴ In the earlier example, where the predictive algorithm ascribed weight to schools’ past performance, in many cases to the detriment of socially disadvantaged students, ultimately the results were scrapped. Instead, schoolteachers’ professional assessment of students’ previous performance and overall achievement was adopted, based on human knowledge, expertise, empathy, and a moral compass that has been developing for millions of years.

Against the background of increasingly intelligent, self-learning systems, as well as the rapid advancement of AI, the need for values, rules, and regulation is becoming urgent.⁵ At present, the most secure protection against bias, invasion of privacy, and other ethical issues is tight human control over AI systems.

According to Dirk Wacker, Director Technology and Innovation at G+D Corporate Technology Office, “Ongoing research in AI is helping to develop methods for assessing and improving the fairness of AI systems. However, there are many complex sources of unfairness – some societal and some technical,” he adds. “It is not possible to fully remove bias from decision making, neither for AI nor for humans. The goal has to be to avoid bias-related suffering as much as possible and to agree on the trade-offs between fairness and efficiency.”

Extending the human skill set

Perhaps the most visible manifestation of AI at present is in industry. According to statistics generated by PricewaterhouseCoopers, 37% of workers in the US are already worried about losing their jobs because of technical advancements in automation.⁶

And not without reason. Research from 2017 on future job losses owing to automation shows that the transportation, storage, and manufacturing industries, along with financial services, will face a huge impact from automation job displacement.⁷ As well as threatening the jobs of industrial workers who perform manual tasks, over the next decades AI is set to put many office administration workers out of work. AI can create accurate contracts without any loopholes; it can analyze X-rays and scans, develop software, create graphics, write texts, collect and communicate news, and edit videos.

It’s true that AI can do all this. But this also opens up opportunities for new, more specialized jobs. Lower-skilled workers could have the ability to take on more highly skilled roles, as human decision-making will remain crucial.  A report by Dun & Bradstreet found that only 8% of 100 surveyed business executives said that they are cutting jobs due to AI, while 40% are adding more jobs as a result of AI adoption.⁸ The World Economic Forum’s 2018 “The Future of Jobs Report” even estimated that 133 million jobs would be created by AI by 2022.⁹

Dr. Jutta Häusler, Head of Corporate Human Resources at G+D, says, “Rather than viewing new technology as competition for jobs, we should see it as complementary – a means of benefiting both processes and people. By eliminating repetitive, routine work, AI allows people to focus on the skilled part of manual work as well as creative and meaningful tasks. Therefore AI can make people more inspired by enabling a clear view on the purpose of what they do.”

AI will certainly help machines to support us humans, boosting our efficiency, reducing our errors, and extending our human skill sets to help us find increased job fulfillment and create new jobs for lower-skilled workers. Looked at this way, AI becomes just another means to increase our individual potential and benefit society as a whole, just as the wheel helped agricultural workers, or the microchip increased connectivity, or the medical scan allowed doctors to make more informed diagnoses.

Setting standards for a safe future

Many companies, including G+D, are now starting to define their own position as regards the ethics of AI. It’s important to find an ethical framework for future regulation to be able to continue our tradition of creating reliable solutions that are based on security, confidence, and trust, as well as ethical and legal standards. Microsoft, for example, has said it won’t sell facial recognition technology to US police departments until legislation is in place to regulate the technology, grounded in human rights.¹⁰ IBM and Amazon have both set limitations on plans to sell facial recognition. As Microsoft’s president Brad Smith said at a recent Washington Post press event, “We need Congress to act, not just tech companies alone.”

In a world where important decisions – such as whether a person can receive credit or an insurance policy, or a new lung – may in the future be taken by machines that mimic human intelligence but lack a certain empathy, we need to ensure that uncertainty and fear don’t cast a shadow over the potential and opportunities offered by AI and ML. As the Future of Humanity Institute at the University of Oxford writes, “It is critical that global governance institutions are put in place to steer these transformations in beneficial directions.”¹¹ In the end, the direction we go is in our own, very human, hands.

Further reading

Is AI good for humans? In a world where decisions are increasingly taken by algorithms, what are the challenges and the opportunities of this new technology? Find out in our animated infographic here

Published: 30/09/2020