Mathematicians have created an artificially intelligent system capable of learning and acquiring new skills by observing the actions of other neural networks or humans in an unfamiliar environment and transmitting the acquired knowledge further.
“Nature Communications” reports that this enables the application of the cultural-technological evolution concept to the advancement of artificial intelligence (AI).
These experiments demonstrate that the transmission of knowledge and the development of technological culture can be achieved with an impressively modest and uncomplicated collection of components. The researchers write that this discovery paves the way for developing general-purpose AI systems via cultural-technological evolution.
GoalCycle3D, an associated learning system and “evolving” neural network, were created by a group of British mathematicians led by Zhang Li, a senior researcher at the London, United Kingdom-based company “DeepMind.”
The organisation has established itself as a frontrunner in developing sophisticated artificial intelligence (AI) systems that can independently learn to play computer and board games, prove theorems, and resolve other intricate challenges formerly deemed unattainable by humans.
Li’s group has recently developed an artificial intelligence system that has acquired an additional ability shared by certain higher primates and animals, including the capacity to learn from the mistakes of others and pass on acquired skills to other members of the species. This capability has historically been regarded as the cornerstone of human cultural and technological advancement, enabling our predecessors to develop ever more intricate implements and lay the groundwork for civilisation.
British scientists have effectively emulated this procedure by employing a custom-built three-dimensional computer game in which a neural network and a human “expert” engage in a collaborative effort.
Researchers have discovered that in such an environment, neural networks with built-in analogues of human attention, the capacity to monitor and link the actions of other players, and both short-term and long-term memory are more likely to develop the ability to learn from others.
Preliminary observations of the neural network indicate that it employs a comparable learning mechanism and rapidly develops its own strategies for resolving previously unknown problems—approximately twenty to thirty hours.
Simultaneously, researchers have discovered that termination of knowledge transmission occurs when at least one element of the system is eliminated, such as attention, memory, or the capability to observe the actions of others.
Scholars assert that analogous learning systems may be applied to the development of intelligent systems in progress and to the investigation of how children acquire the capacity to learn over time. Furthermore, by monitoring the functioning and progression of these systems in subsequent periods, anthropologists may gain insights into how our predecessors obtained the capacity to convey knowledge and advance cultural and technological developments.
