With rapid developments in artificial intelligence (AI), ethics has now moved to the frontline of many companies as they consider how to use AI to avoid the unintended consequences of large-scale automation and algorithmic decision-making.

After facing an AI winter on the back of overhyped expectations that could not be met when first proposed in the 1950s, AI has seen rapid growth in the past few years, particularly around machine learning, deep learning and cognitive computing.

Increasingly, this is is necessitating assessments and ethical reviews, with an opportunity now to eliminate bias and discrimination in decision-making, says SAS CTO, COO and executive VP Oliver Schabenberger.

“We are talking about replacing the decisions that we make as humans, with decisions that are made by algorithms that are crafted by humans or crafted by machines,” he says, noting that some algorithms are imprinted by machines, with a neural network that is “trained by operating a set of data”.

With the impact AI is set to have on the markets, processes and individuals, there is a need to consider implicitly programmed systems that make decisions.

“We are having to develop a better understanding of how to govern and manage AI systems in an ethical way, [while] ensuring that we keep up with technological developments,” he told delegates at the Analytics Experience 2018, held in Milan, Italy, last month.

AI systems can augment human efforts; however, human efforts are also required to augment AI, with current technology “running ahead” of what is capable of being absorbed and integrated into daily processes.

Citing autonomous vehicles and health applications, he explained that the technology was close to completion; however, questions surrounding responsibility over an accident or misdiagnosis had not yet been resolved.

AI systems can be developed to self-govern, but they still need overseeing to ensure that their governance decisions are correct.

“Many of these decisions will boil down to the question not of what is possible, or what can be done, but rather what should be done, and that is a very human question,” he added.

Intel AI and technical computing director Stephan Gillic, also on the panel discussion, said that there needed to be a series of controls in place, including accuracy of data, programming and verification.

The key would lie in the way the application was applied, said Accenture Applied Intelligence MD Michele Crescenzi during a panel session at the SAS-hosted two-day conference, suggesting the use of a predetermined framework.

One framework for considering ethical AI development is known as Fair, Accountability, Transparency and Explainability, or ‘FATE’.

Fair, or the removal of bias and corporate discrimination, guides the system to help remove human bias, and not to build in new machine biases of its own, with the question of who decides that AI outcomes are right to be considered.

Accountability is the ownership of decision-making and the willingness to take responsibility, while transparency aims to avoid ‘black box processes’ and ensure a clear, good understanding of the process from start to finish.

Explainability, or the ability to explain and make sense of the decision, is about model interpretability and is closely linked to transparency.

These four elements create a framework for effective governance of AI systems.

However, many organisations are already gearing up for more ethical and responsible use of AI, according to a joint study by SAS, Accenture Applied Intelligence, Intel and Forbes Insights.

“Organisations have started addressing concerns and aberrations that AI has been known to cause, such as the biased and unfair treatment of people.”

Ninety-two per cent of AI leaders train their technologists in ethics, while 74% evaluate AI outcomes weekly, indicating that business leaders are taking steps to ensure responsible use of AI within their organisations.

Some 70% of other AI adopters – which now account for 72% of organisations globally – conduct ethics training for their technologists and 63% have ethics committees in place to review the use of AI.

“Despite popular messages suggesting AI operates independently of human intervention, the research shows that AI leaders recognise that oversight is not optional for these technologies,” the study points out.

Nearly 75% of AI leaders reported careful oversight, with at least a weekly review or evaluation of outcomes, compared with only 33% of the less successful AI adopters.

Further, 43% of AI leaders and 28% of AI adopters have a process for augmenting or overriding results deemed questionable during review.

“It stands to reason that companies are taking steps towards ethical AI and ensuring AI oversight, because they know that faulty AI output can cause repercussions,” SAS notes.

However, oversight processes have a long way to go before they catch up with advances in AI technology.

The study, ‘AI Momentum, Maturity and Models for Success’, includes responses from 305 business leaders in the Americas, Europe and the Asia-Pacific region.