- DeepMind, Google’s AI division, unveiled a new AI model dubbed ‘RoboCat,’ capable of self-improvement.
- Data generated by such AI-run robots allows improvements in technique without human supervision. Such learning, in turn, can be transferred to other robots that can perform myriad tasks.
Google’s AI division, DeepMind, has introduced a self-improving AI model called RoboCat, which could potentially create a way for machines to generate training data to improve their techniques on tasks without human input.
According to DeepMind, the model can learn new tasks with as few as 100 demonstrations, and it claimed that the RoboCat project could bolster research into robotics owing to the reduced need for human supervision.
The RoboCat model is based on DeepMind’s ‘Gato’ multimodal model, which is capable of processing images, language, and action in physical and simulated environments. Consequently, the AI model learns new tasks faster and more efficiently the more tasks it masters.
In early June, DeepMind achieved a notable breakthrough in data sorting, with a 70% speed improvement over conventional methods, under an AI project called AlphaDev. The process uses reinforcement learning to generate optimized algorithms. The developments came soon after Google merged DeepMind with its Brain unit to bolster AI research.
See More: Intel Restructures Manufacturing Business, Intends To Save Billions
DeepMind Co-Founder Promotes ‘Artificial Capable Intelligence’
DeepMind co-founder Mustafa Suleyman stated that conventional Turing’s tests to compare AI capabilities, which have been used since the 1950s, are now outdated. He said that Turing tests could not give data about system capabilities or understanding to make plans over abstract timelines, a key characteristic of human thinking.
The old Turing test simply had human evaluators determine if they were speaking to a human or a machine. If the machine was able to pass as a human, it passed. Suleyman instead proposed a test considering financial feasibility in which AI bots should be able to take a $100,000 investment and turn it into a million dollars.
This could include researching business ideas, generating product plans, sourcing manufacturers, and selling the products. Suleyman expects AI to achieve these capabilities in the next two years.
What role do you think AI developments will play in general-purpose technology? Let us know your thoughts on LinkedInOpens a new window , TwitterOpens a new window , or FacebookOpens a new window .
Image source: Shutterstock
LATEST NEWS STORIES
- Apple Releases Fixes for Actively Exploited Vulnerabilities in Safari, iOS, and macOS
- HPE Discover 2023 Spotlight: Company Launches HPE GreenLake for Large Language Models
- Cybersecurity Researchers Expose ‘Legitimate’ Android App as Malware
- Blackcat Ransomware Threatens To Leak 80GB Data Unless Reddit Withdraws Its New API Policy