A group of Russian scientists has developed a new artificial intelligence (AI) model capable of self-adapting to new tasks and contexts without human intervention. In the future, this technology is expected to become a game-changer in the realm of warfare.
The model enables developers to overcome one of the main limitations in contextual machine learning, said the team from the T-Bank AI Research Lab (formerly Tinkoff Bank) and the Artificial Intelligence Research Institute (AIRI) based in Moscow, in a paper published online.
Existing models, although able to learn new tasks when provided with sufficient data, are still constrained by a predefined and static set of actions, the researchers explained. The introduction of new “action spaces” would then require a new set of data, often quite extensive, and retraining of the model. “These limitations make re-adaptation a costly endeavor for some applications,” they stated, as reported by RT.
The team took a specialized machine learning model called Algorithm Distillation (AD) and further modified it to meet their set objectives. The AD method trains AI to perform tasks by autoregressively predicting actions while using its historical learning data as context.
The Russian model, dubbed ‘Headless-AD,’ was presented at the International Conference on Machine Learning in Vienna this week. The Headless-AD approach enables the model to learn and apply new actions in response to new tasks without requiring additional human input or retraining.
According to the team, their AI is capable of performing five times more actions than it was initially trained to do. The researchers noted that this could have broad applications ranging from space technology to smart home assistants.
Such a model can be taught some basic actions on general data and then adapt to the specific conditions of a given context, according to the team’s report. Some Russian media have suggested that the new AI model may be smart enough to pass the so-called ‘coffee test,’ which the now-famous ChatGPT reportedly failed.
First proposed by Apple co-founder Steve Wozniak, the test requires an AI machine to “enter an average American household and figure out how to make coffee, including identifying the coffee machine, figuring out the button functions, and locating the coffee cupboard.”
The problem for most AI systems is that while the average household has many similarities, each is still slightly different, typically requiring the AI machine to be trained on specific datasets related to a particular household to perform tasks there.
Performing the same task in a new household would necessitate retraining on a new dataset. However, the self-adapting Russian AI potentially has the capability to carry out its tasks without this need, according to the report.
- Russian scientists developed a self-adapting AI model, 'Headless-AD,' capable of learning new tasks without human input.
- The model addresses limitations in contextual machine learning by allowing broader action execution, reportedly five times more than initially trained.
- It adapts to specific conditions, potentially passing the 'coffee test.'
- The technology could revolutionize warfare and various applications, from space tech to smart home assistants, eliminating costly retraining needs.
Self-Adapting AI Model
Advancement in Machine Learning
Adaptability and the Coffee Test
Revolutionizing Various Fields
