Study: AVs may one day have human-like ethics built in
- A new study from The Institute of Cognitive Science at the University of Osnabrück found that programming human ethics into autonomous vehicles could be easier than previously thought. Researchers once believed that the human ethical decision making process was too contextual to be modeled in an algorithm, yet the study "found quite the opposite," as reported by Science Daily.
- The research included putting volunteers into virtual reality (VR) sets to study human behavior in simulated road traffic conditions. They experienced unexpected and unavoidable situations with combinations of inanimate objects, animals and humans, and were forced to to decide on the spot where to direct the vehicle.
- The researchers were then able to develop a model with "well over" 90% prediction accuracy that corresponded to human decision making in the traffic situations they tested for, as reported by Frontiers.
Robot and artificial intelligence (AI) ethics are hotly debated with big names on both sides. The Ethics and Governance of Artificial Intelligence Fund is offering millions of dollars in grants to explore biases in AI systems like AVs. IBM recently sent a letter to Congress to let lawmakers know AI is not going to lead to a dystopian future — unlike Stephen Hawking and others, who have cautioned against AI.
Other work is being done specifically to deal with ethics within the AI field. Last year, Google, Amazon, Microsoft, IBM and Facebook formed the Partnership on AI to Benefit People and Society, with the goal of creating a set of guidelines for AI development.
Humans are still needed to train autonomous vehicles, which has become a business model for some companies. Mighty AI, a data training company, pays people to carefully label images online so AI systems in automated vehicles know that what it is seeing — whether it is a tree, a child with a balloon or a bag on the street. Starsky Robotics has programmers sitting with truckers as they drive their rigs in order to get the data still needed before trucking could truly autonomous.