Will AI become self-aware? Scientists warn we need to act now to anticipate the societal risks
AI could become self-aware through robotics so we should act now to anticipate the societal risks, according to researchers at the University of Sheffield.
- Artificial intelligence is making rapid progress but is not yet at the stage of developing self-awareness, according to researchers at the University of Sheffield
- Professor Tony Prescott, Professor of Cognitive Robotics at Sheffield, outlines how humans first develop their sense of self and warns that advances in robotics could lead to AI becoming self-aware accidentally, meaning we should act now to protect against future risks
- Professor Prescott makes the call in a special theme issue on the sense of self in the Royal Society’s Philosophical Transactions - the world’s first and longest-running scientific journal
AI could become self-aware through robotics so we should act now to anticipate the societal risks, according to researchers at the University of Sheffield.
In a special theme issue of the Royal Society’s Philosophical Transactions - the first and longest-running scientific journal in the world - Professor Tony Prescott, Professor of Cognitive Robotics at the University of Sheffield, investigates through computer simulation how newborn babies learn to distinguish the self from the world. Understanding the human sense of self will help scientists to determine how AI may become self-aware and how to protect against future risks.
Current AIs, such as chatbots, can give the appearance of a sense of self, however, this is best seen as a form of “role-play”. This new research shows that, through robotics, AIs could develop a more human-like sense of self.
Concerns that AI could become self-aware and the impact that could have on the human race have been long-running in the scientific community, but also voiced by industry insiders and former developers in leading tech companies. In the special theme edition of the journal, Professor Prescott introduces the latest research on the sense of self in humans, including how advances in the understanding of the self could improve mental health, and the question of whether machines might become self-aware, as AI and robotics rapidly advance.
Professor Prescott also argues that robots, like humans, need to be aware of their physical boundaries so that they don’t damage themselves. Advances in robotics could therefore lead to AIs that have a self-world distinction, which is the beginning of the human sense of self. Given other AI capabilities, such as the ability to think about the past and the future, this could ultimately lead to AIs that have human-like self-awareness.
Professor Tony Prescott, Professor of Cognitive Robotics in the University of Sheffield’s School of Computer Science, said: “To understand if AI can become self-aware, we first have to look at humans. Our sense of self develops in early infancy, driven by physical body awareness long before language.
“Because AI chatbots lack bodies, they won't develop a sense of self. Robotics, however, is a different story - body awareness is critical for robots, so developing AI in physical machines could ultimately lead to self-aware AI.”
Dr Luca Manneschi, who led the AI aspects of the study and is also a Lecturer in Machine Learning at Sheffield’s School of Computer Science, said: “We used AI to teach a simulated robot to visually distinguish between its body and the world by observing its own movements.
“As in many modern AI systems, we trained the model in stages. The aim of the first stage was to learn which parts of the scene belonged to the robot’s own body. An additional AI layer then built on that distinction to predict what the robot would see next.
“Our study illustrates how early experiences, such as watching your own hands move as an infant, could help you begin to learn where your body ends and the world begins. It also demonstrates how robotic embodiment could lead to the emergence of self-awareness in AI.”
This research was conducted in partnership with Sheffield Hallam University and Kinematic Trees Ltd, a Sheffield-based robotics startup.
