Russian researchers unveil AI model that adapts to new tasks without human input

07/30/2024 / By Richard Brown

A team of Russian researchers has developed an innovative artificial intelligence (AI) model that can adapt to new tasks and contexts without requiring human intervention.

The model called Headless-AD was presented at the International Conference on Machine Learning, held in the Austrian capital Vienna from July 21 to 27. Researchers from the T-Bank AI Research Laboratory and the Moscow-based Artificial Intelligence Research Institute were responsible for this breakthrough.

The Headless-AD AI model addresses a significant limitation in contextual machine learning, according to Russia Today (RT). Previously, AI models could learn new tasks with sufficient data – but were restricted by a pre-defined and fixed set of actions. Introducing new actions typically required extensive data and model re-learning, making re-adaptation a costly and time-consuming process.

The researchers explained that the introduction of a new “action space” necessitated a fresh dataset and model re-learning, which posed significant challenges for many applications. They enhanced a specific machine learning model called algorithm distillation (AD) – which trains AI to perform tasks by auto-regressively predicting actions based on its learning history dataset – to create Headless-AD. (Related: AI chatbot admits artificial intelligence can cause the downfall of humanity.)

Headless-AD’s capability means the AI can adapt to new situations and tasks seamlessly, increasing its flexibility and applicability. According to the researchers, the AI developed using the Headless-AD model was capable of performing five times more actions than it was originally taught. The researchers believe this capability could have wide-ranging applications, from space technologies to smart home assistants.

Headless-AD can pass Wozniak’s “coffee test”

The Headless-AD model’s ability to adapt to specific conditions based on generalized data makes it highly versatile. For instance, it could be initially taught basic actions using generalized data and then adapt to the specific conditions of a particular context. This flexibility is expected to enhance its utility in various fields, such as space technologies, where adaptability to different environments is crucial

Russian media have suggested that this new AI model might be capable of passing the so-called “coffee test,” a challenge first proposed by Apple co-founder Steve Wozniak. The test requires an AI to enter an average American household and figure out how to make coffee, including identifying the coffee machine, understanding the buttons, and locating the coffee cabinet.

This task is particularly challenging because, although households share common features, they all have slight variations that typically require an AI to be trained on a specific dataset for each household. Most AI systems struggle with this because performing the same task in a new household would require re-learning on a new dataset. However, the self-adapting nature of the Headless-AD model suggests it could potentially handle this challenge without the need for re-learning, making it a significant advancement in AI capabilities.

Head over to Robots.news for more stories about developments in AI.

Watch this video of retired Army Gen. Mark Milley on “60 Minutes” explaining the future of warfare and AI in the military.

This video is from the Daily Videos channel on Brighteon.com.