DeepMind replaces Asimov’s laws with adaptive dataset for robots
Google DeepMind under the leadership of Carolina Parada is rethinking fundamental principles of robot safety and wants to move away from classical Asimov’s laws to a more flexible, trainable system. The new so-called “Asimov Dataset” represents not a rigid set of rules, but an adaptive base of potentially dangerous situation scenarios.
The key difference of the new approach lies in the method of risk processing. Modern robots don’t simply follow preset directives – they learn to analyze context. And make decisions based on an extensive base of examples. When a robot sees a glass on the edge of a table, it doesn’t execute a pre-programmed command. But evaluates the situation and moves the object to a safe position. Discovering an object on the floor, the system recognizes potential danger for a person and eliminates it.
The dataset is formed based on analysis of real incidents from different countries of the world, which ensures diversity of cultural and social contexts. Each scenario is accompanied by visual examples and instructions for risk minimization, creating a comprehensive educational environment for artificial intelligence.
This approach differs with 3 fundamental features: dynamic data updating, hybrid control with human participation and openness for testing by third-party developers. Thus, at DeepMind they believe that the “Asimov Dataset” creates not just technology, but an evolving safety ecosystem.
