Google Has Given DeepMind a Memory, The AI Will not Make the Same Mistakes Again
Google’s DeepMind has been playing video games on the Atari since 2014, and it got pretty good too, beating human scores. The problem was, it couldn’t remember how it did it.
Google’s DeepMind has been playing video games on the Atari since 2014, and it got pretty good too, beating human scores. The problem was, it couldn’t remember how it did it. So, every time a new Atari game was introduced, a new neural network was created, but in doing this, the AI could never benefit from its own learned experiences. However, a group of researchers from DeepMind in collaboration with those at Imperial College London has been busy creating an algorithm that could change all that.
The new algorithm allows the AI to learn, retain, and then reuse the knowledge that it learns. This is done through the use of supervised learning and reinforcement learning tests. For humans, the basis for continual learning is synaptic consolidation. Being able to save knowledge and use it later to carry out a different task is an essential part of how humans learn. This is one area where machine learning has come unstuck.
DeepMind is an impressive system; now it can retain the most vital information from previous experiences to learn. But even so, it has its limitations and still can’t perform as good as a neural network that completes just one single game. If machine learning is to match real world learning the next step is the efficiency of learning. With the new algorithm supporting continual learning just like that of a human brain it’s a huge leap forward for AI and one that will hopefully allow more creative and intellectual tasks to be carried out.
More News to Read
- Breakthrough for Hearing Loss Patients Comes in the Form of Replacement Hair Cells
- Understanding the Epigenetic Clock
- Should We Be Preparing for the Explosion of Betelgeuse Supernova?
- The Truth Behind Asteroid Hunters
- Germany Opens New ‘Fake Sun’ Hydrogen Producing Facility
Related Coverage
- Renewed Google+ Now On Stage
- Do Robots Make Good Teachers?
- Google Uses StarCraft II Game as Testing Platform for AI
- Robots Can Now Talk like Humans with Thanks to Google’s DeepMind Project
- Meteor Shower From Dead Comet Sighted Again
Don't miss new tech stories on Google
Add TrendinTech once in the Google app and our stories appear in your news suggestions.
Daniel Okafor
Daniel Okafor is the Senior AI Reporter at TrendinTech, where he covers large language models, machine learning research and the practical use of artificial intelligence across business and government. He previously reported on artificial intelligence for MIT Technology Review, covering the labs behind the current generation of frontier models and the policy debates in Washington and Brussels. Daniel holds a Master of Science in Machine Learning from Carnegie Mellon University and follows the research community closely, attending NeurIPS and ICML each year to speak with the people behind the papers. He has a particular interest in evaluation: how models are benchmarked, where those benchmarks fail and what that means for the companies betting on them.
All stories by Daniel Okafor (316)