Google DeepMind is exploring AI that learns a bit more like we do. They’ve recently patented technology aimed at letting artificial intelligence acquire new knowledge or skills without overwriting what it’s already learned – a process remarkably similar to the human brain.
The core idea lies in a patent for ‘function-preserving neural network expansion’. Think of it this way: when you learn a new hobby, your brain doesn’t wipe out your knowledge of how to ride a bike. Instead, it builds upon your existing experiences. This technology aims for something similar with AI. It allows developers to expand an AI model’s size or tailor it for a specific task without having to start the learning process from scratch, preserving its foundational knowledge.
This could be a significant step. Currently, significantly altering an AI model often requires extensive retraining. This new approach suggests a way to continuously grow and adapt AI capabilities more efficiently. It might be one of the key background developments enabling Google to work on multiple complex AI models simultaneously, building incrementally rather than constantly starting over. Imagine AI models that can continuously evolve and specialize throughout their lifecycle.
While this “never forget” learning captured attention, it wasn’t DeepMind’s only recent innovation. They also patented tech to optimize virtual tours in 3D environments using popular photo spots and another method to perfectly mix training data – like a chef adjusting ingredients – to boost AI performance before large-scale training even begins.
These advancements highlight Google’s push towards more sophisticated, efficient, and perhaps more human-like artificial intelligence, capable of learning and adapting in ways that were previously much harder to achieve.
| Latest From Us
- Forget Towers: Verizon and AST SpaceMobile Are Launching Cellular Service From Space
- This $1,600 Graphics Card Can Now Run $30,000 AI Models, Thanks to Huawei
- The Global AI Safety Train Leaves the Station: Is the U.S. Already Too Late?
- The AI Breakthrough That Solves Sparse Data: Meet the Interpolating Neural Network
- The AI Advantage: Why Defenders Must Adopt Claude to Secure Digital Infrastructure
