AI acting ethically in tomorrow’s world depends on how we shape its governance today

AI tech is not inherently biased or manipulative, it’s all about how we train and use it…

Despite the concerns surrounding artificial intelligence, investments in the technology are growing at an accelerated rate based on the expectation that AI will improve the human condition. There are already many examples where AI has greatly enhanced the customer experience and had a material impact on corporate profits. And the potential societal impact is even more impressive. AI has the capability to fast-track the achievement of the United Nations Sustainable Development Goals. AI can be deployed to address illness and disease. AI can be implemented to broadly improve wellbeing for the majority, particularly those in developing countries. However, despite all the wins with AI, there have also been many disappointing and discriminatory outcomes. One thing is clear: decisions we take today in terms of ethics and governance will be crucial in shaping the impact of artificial intelligence on society in 20 years’ time.

The words are those of Cathy Cobey, EY’s Global Trusted AI Leader. As a keynote speaker at the World Summit Artificial Intelligence (World Summit AI) in Amsterdam-Zaandam on October 12/13, Ms Cobey is an acknowledged expert in the ethical and control implications of AI. On the eve of the conference, we asked her for her perspective on a handful of defining themes related to the societal implications of technology transformation.

1. Ethical AI reinforces our fundamental values

“AI is still in the infancy of its technological journey. Our job right now is to set the ethical and safety rails that AI should operate within, both today and into the future. We need to ensure that AI adheres to well-considered ethical guidelines such as privacy, individual rights and the increasingly complex area of non-manipulation,” according to Ms Cobey. “AI is trained on historical data. Data that incorporates biases and outdated value systems of yesterday. We need to leverage the power of our massive data stores, while preventing historical discrimination being replicated through AI’s decision framework. With proper design and training, artificial intelligence can reverse historical inequality, but only if we consciously build equality into the objectives set for the AI system.

“My point is that – ideally – AI can be a powerful vehicle for good. We should harness AI today to open up education to students all around the world building for each child a customized learning plan. We should embrace AI to meet multiple societal objectives such as driving energy efficiency, while improving individual safety. AI can easily find the optimal schedule for street lighting to conserve energy and decrease aggressive behavior in entertainment areas. We should programme AI to save lives by pre-empting imminent catastrophe and assist our leaders in conducting sophisticated scenario planning, not only to draft but also to implement more timely and effective emergency response plans. AI has proven that there is still a lot of untapped value in our current data stores. EY sponsored a global modeling competition to demonstrate how frog counting is an excellent indicator of an ecosystem’s health. Weather data is a particularly useful data set being used by large investment companies to more accurately predict storm impacts on commodity prices, and by small scale farmers to time their crop planting and harvesting. I truly believe that AI can have a huge positive impact on society, but as great as all those benefits are, it is a technology that can also be used to cause a lot of harm.”

Ms Cobey, who has been helping institutions to better understand and manage business risks related to technology throughout her 29 years with EY, is convinced that “AI can be an enabler towards improving the human condition but, as it builds its own decision framework, ethical and equitable objectives need to be built into its design” – not mitigated for afterwards or accepted as the way it’s always been done. “It’s a no-brainer that the technology will become increasingly available and the onus is on us to grasp it with wisdom, prudence and respect. This is no easy task and will require the joint effort of robust multidisciplinary teams.”

Despite the complexity, it is the diversity of dedicated designers – in terms of gender, age, values and lineage – that will be the backbone of human-technology interaction, working together to improve the quality of people’s everyday lives. “A society that is in some important ways undesirable can – with the help of AI – be reshaped and transformed into a better version of itself.”

2. Human centricity: how do you fit technology to its user?

The million-dollar question is how to make positive human-technology interaction happen in a qualitative, cost-effective fashion. “People need to interact with technology and technology needs to understand the full spectrum of its user,” says Ms Cobey, when asked to consider aspects like cognition, perception and motivation. “Human-technology interaction focuses on technology from a user’s perspective. How should we assess the viability of new technological developments in relation to human constraints and capabilities? Ultimately, we want AI to help improve the human experience. We must, however, bear the following in mind; initiatives that may be beneficial to the majority may well be detrimental to the minority. Many AI models are built to determine the optimal outcome. But there is a design flaw when the historical data it is trained on, would teach AI that the most hired candidate in the past was a white male. Or that husbands, on average, were lent more money than their wives. There is a well-discussed example where a husband and wife – with the same input variables – were given vastly difference lending limits. Considering beyond the individual impact, what impact could there be on society if AI, rather than eradicating historical biases, magnifies them?”