It was stated that a new artificial intelligence model developed in Russia can learn new actions on its own. A group of Russian scientists have developed a new artificial intelligence model that can adapt itself to new tasks and context without additional intervention from humans.
A team from T-Bank (formerly Tinkoff Bank) Artificial Intelligence Research Laboratory and the Moscow-based Artificial Intelligence Research Institute (AIRI) said in a paper published online that the model allows developers to overcome one of the main limitations in contextual machine learning.
The researchers explained that while previously existing models can learn to perform new tasks when fed enough data, they are limited to a predefined and fixed set of actions. The introduction of a new “domain of action” required a new data set, often quite extensive, as well as relearning of the model. This limitation would make re-adaptation a costly endeavor for some applications, they said.
The team took a specific machine learning model called Algorithm Distillation (AD) and further modified it to meet the set goal. The AD method trains AI to perform tasks by autoregressively predicting actions while using the learning history dataset as context.
The Russian model is called ‘Headless-AD’ and was presented at the International Conference on Machine Learning in Vienna this week.
The headless-AD approach meant that the model gained the ability to learn and execute new actions in response to new tasks without the need for additional input or relearning by a human.
According to the team, their AI was capable of performing five times more actions than initially taught. Researchers stated that this could have a wide range of applications, from space technologies to smart home assistants.
According to the team’s report, such a model could be taught some basic actions on generalized data and then adapt to the specific conditions of a particular context. Some Russian media later suggested that the new AI model might be smart enough to pass the so-called ‘coffee test’, which has now been reported to have failed by the infamous ChatGPT.
First put forward by Apple co-founder Steve Wozniak, the test involves an AI machine “walking into the average American home and having the ability to recognize the coffee maker, figure out what the buttons do, find the coffee fridge, etc.” requires “understanding how to make coffee, including
The problem for most AIs is that while average households have a lot in common, they are still all slightly different; This normally requires an AI machine to be trained on a specific set of data about a particular household before it can perform the task there.
Doing the same task in a new home would require relearning on a new data set. However, reports have suggested that self-adaptive Russian AI could potentially fulfill this task.
