People believe that artificial intelligence assistants should be accountable for the results.

2023-08-31

A new study suggests that even if humans view AI assistants purely as tools, they still attribute partial responsibility for decision-making to them.

Future AI-based systems may guide autonomous vehicles without human input. The study shows that people believe these future AI systems are as responsible as humans when making autonomous traffic decisions. However, current AI assistants are far from having this level of autonomy. They provide assistance to human users, such as navigation and driving aids.

So, who should be held responsible in real-life situations when problems arise? The human user or the AI assistant? A team led by Chair of Philosophy of Mind, Louis Longin, has investigated how people assess responsibility in these cases. Their research findings have been published in the journal iScience.

"We all have AI assistants in our pockets," says Longin. "However, most of the empirical evidence we have about responsibility is focused on robots or autonomous vehicles, where AI is actually sitting in the driver's seat, making decisions for us. But in cases where humans still make the final decision, AI is more like a complex tool."

Philosopher Longin, along with colleagues Dr. Bahador Bahrami and Chair of Philosophy of Mind, Professor Ophelia Deroy, conducted a study involving 940 participants to investigate the views of human drivers using intelligent AI voice assistants, intelligent AI tactile assistants, or non-AI navigation instruments. They examined whether participants believed that navigation aids should be held responsible and to what extent AI assistants are considered tools.

The Contradictory State of AI Assistants

The research findings revealed a contradictory mindset: participants strongly argued that AI assistants are merely tools, yet they believed that AI assistants bear partial responsibility for the mistakes of human drivers using them. This responsibility division did not exist for non-AI-driven instruments.

Equally surprising to the authors was the fact that "people may use different moral standards to evaluate praise and blame. When a crisis is averted and no harm is caused, the standards are relaxed, making it easier for people to attribute credit to non-human systems rather than blame them,"

The Irrelevance of Language

In this study, the authors found no distinction between AI assistants that use language and those that warn users through tactile vibrations.

Ophelia Deroy says, "In this situation, both provide the same information, 'Hey, be careful, there's something ahead,' but of course, ChatGPT provides more information in practice." Her research demonstrates our contradictory attitudes towards AI. Regarding the additional information provided by new language-based AI systems like ChatGPT, Deroy adds, "The more interactive it is, the easier it is to anthropomorphize."

"In conclusion, our research supports the view that AI assistants are not just seen as tools but are still far from reaching human levels," says Longin.

The authors believe that the findings of this new study will have a profound impact on the design and social discourse surrounding AI assistants. "Organizations developing and releasing AI assistants should consider how they influence social and moral norms," concludes Longin.

RECENT AI NEWS

RECENT AI TOOLS