Loading...

  • 21 Nov, 2024

Russian scientists develop self-adapting AI

Russian scientists develop self-adapting AI

A new artificial intelligence model developed in Russia is reported to have the ability to independently learn new actions.

A team of Russian scientists has developed a new artificial intelligence model capable of self-adapting to new tasks and contexts without requiring additional human intervention.

The model has addressed a major limitation in contextual machine learning, according to researchers from the T-Bank (formerly Tinkoff Bank) AI Research Lab and the Moscow-based Artificial Intelligence Research Institute (AIRI), who detailed their findings in a paper published online.

Previously, AI models could learn new tasks when provided with ample data but were restricted by a fixed set of actions. Introducing new actions required extensive data and re-learning, making adaptation costly for some applications.

The team enhanced a machine learning model called Algorithm Distillation (AD), which trains AI to perform tasks by predicting actions autoregressively and using its learning history as context. They developed this into a new model known as ‘Headless-AD,’ which was presented at the International Conference on Machine Learning in Vienna this week.

The Headless-AD model enables AI to learn and execute new actions autonomously in response to new tasks, without further human input or re-learning. The team reported that their AI could perform five times more actions than it was initially trained to do, with potential applications ranging from space technology to smart home assistants.

The AI model can be taught basic actions using generalized data and then adapt to specific conditions in various contexts. Some Russian media have speculated that this AI might be capable of passing the so-called ‘coffee test,’ proposed by Apple co-founder Steve Wozniak. This test challenges an AI to enter an average American household and make coffee, including identifying and using the coffee machine and locating the coffee cabinet.

Most AI struggles with this test due to the variability in household setups, requiring re-learning for each new environment. However, the self-adapting nature of the Russian AI could potentially overcome this challenge, according to the reports.