The Future of AI Chips, They Will Be in Everything
A discussion surrounding a future where AI enhances human lives was conducted at Berlin’s TechCrunch Disrupt conference.
Artificial Intelligence (AI) is already pervasive in many applications. Deep learning has happened since 2012 and it’s only growing. We’re just at the beginning of what’s possible.
This is Nigel Toon’s perspective, which he shared at Berlin’s TechCrunch Disrupt conference on Wednesday. Toon is Graphcore’s CEO, a company that specializes in AI chips.
For Toon, AI can allow many different kinds of innovation, deep learning, and probabilistic machine learning to make new breakthroughs.
How do Graphcore’s AI chips function?
AI chips are designed to process data to work like the human brain. The challenge with that is that we don’t even quite know how the brain works. However, what you’re doing with a machine learning model — like a brain — is trying to pull together all the information that it’s learned and turning it into a logical thought.
Toon explained that the way Graphcore’s processor works is that it understands that data structure. Data can be moved when needed as the computer evolves with machine learning.
As Toon pointed out, the CPU in your laptop is working on what your next action is going to be, whereas in machine learning it becomes a smoother process.
How will AI chips affect our everyday lives?
Most of us won’t be using Graphcore’s chip directly. Instead, we’ll be using the software that companies develop new applications from it.
Graphcore wants to make the technology available, so working with companies such as Dell, Microsoft Azure – is helping Graphcore to spread where the tech is available for people to use on a wider scale.
For instance, their tech would enable people to make the next breakthroughs in natural language processing. It’s hard enough for humans to predict what’s going to happen next in real life, let alone teaching a machine to predict the next move.
AI and safety
Toon stressed that these values are set by different nation-states and are most likely going to be different worldwide.
There are always issues with how these tools are being used, and so long as regulations are properly put in place, much good can come from this technology.