21.10.2019 BlogHow to build ethical AIs?

This summer, ZenRobotics’ Principal Scientist Timo Tossavainen was invited to participate in the first ever European AI Alliance Assembly, held in Brussels, Belgium, on 26 June 2019. The AI Alliance is a discussion forum for AI development and its impact to inform the European Commission’s policy making.

How can we ensure that AI systems promote societal and environmental well-being? That was the focus of the workshop I participated at the AI Alliance event.

In a nutshell, the AIs we develop today should be sustainable and environmentally friendly to benefit all humans, including future generations. That’s part of the work we do at ZenRobotics.

Ethical AI and sustainable robots at ZenRobotics

At ZenRobotics, we develop AI-powered waste sorting robots. The AI software component, developed in-house by us, is an essential part of our product: it enables extreme accuracy, efficiency and flexibility in automated waste sorting.

How do we make our products sustainable from an ethical perspective?

Traditionally, waste sorting has included a lot of manual work that is physically straining and even hazardous. Robotic sorting improves human safety and makes the work physically less demanding. In our experience, robots have been seen as exciting co-workers, who do the heavy lifting.

Environmental sustainability comes as a given in our products. Recycling robots promote circular economy and allow re-using precious raw materials.

With regard to transparency, we do our best to educate our customers and end users on the use and potential of our AI-powered robots. By understanding how they work, and what their limitations and possibilities are, our customers are able to make educated decisions on robotic sorting.

Principles for building trustworthy AIs

Cases on AI’s potential misuse have made headline news during the last years. Cambridge Analytica, profiled advertising and deepfake videos are easy-to-understand examples of possible non-ethical use of AI.

However, when we build AI systems, will we know how they are used in the future? How can we take ethical factors into consideration, let alone design them as part of the systems?

To this end, the AI Alliance’s High-Level Expert Group (HLEG) compiled a set of ethics guidelines for trustworthy AI and launched a piloting process for them at the event.

The ethics guidelines state that trustworthy AI should be lawful, ethical, and robust. For example, an AI system that makes decisions about people should respect non-discrimination laws, but the data used to build such a system may contain hidden biases.

Seven requirements were defined to help developing trustworthy AI:

  • Human agency and oversight.
  • Technical robustness and safety.
  • Privacy and data governance.
  • Transparency.
  • Diversity, non-discrimination and fairness.
  • Societal and environmental well-being.
  • Accountability.

To learn more about these, download the Ethics Guidelines for Trustworthy AI.

Timo Tossavainen has a background in academic research and teaching in visual computing and applied AI. At ZenRobotics, Timo’s work includes developing methods for object and material recognition and robot motion control.