In this interview hosted by Andrew Gaule (linkedin.com/in/andrew-gaule-aimava), Graphcore CEO Nigel Toon shares his perspectives on artificial intelligence and the hardware powering the latest capabilities. Nigel and Andrew are mentors on the AI Stream at Creative Destruction Labs based in Oxford University. See more of the topics below.
Nigel outlines how far AI has advanced, from beating the world champion at Go to the recent explosion in popularity of chatbots like ChatGPT. Underpinning these leaps in software are rapid gains in semiconductor chips, improving at an astonishing 25 billion fold over 60 years. Graphcore builds specialty AI chips to provide an alternative to dominant player Nvidia, allowing more researchers to accelerate discoveries.Beyond keeping up with technical progress, Nigel stresses that education must transform to prioritize creativity over rote learning. He welcomes AI-assisted teaching tailored to individuals. Regarding ethical concerns, Nigel argues biases come from flawed data rather than being inherent to AI systems. Still, developers must pledge transparency while testing for unfair impacts on diverse groups. With sensible safeguards, AI can augment human intelligence to solve previously intractable problems. The technology itself is neither good nor bad; it merely amplifies our own goals and values.Here are six key topics from the interview Graphcore's AI chips Enable more parallel processing like the human brain Optimized to accelerate neural networks for complex AI workloads Provide an alternative to Nvidia for AI compute in data centersComparing AI and human cognition Many subconscious brain skills remain beyond AI systems currently Things easy for people often prove difficult computationally Future computing may better approximate biological infrastructureAI adoption in China Leading aggressive deployment of AI across many sectors Rapid integration into education at early ages Authoritarian system enables swift data collection and trialsAI's economic impact Potential to augment productivity on par with industrial revolution Requires rethinking of education models and job training Risk of automating certain jobs must be mitigatedEthics of AI systems Biases originate from flawed training data rather than inherently Guidelines needed to ensure transparency and test for harm Clearly unethical applications should be bannedHardware progress enabling AI Exponential improvements in semiconductors underlying gains Future quantum and molecular computing shifts possible Enormous data digitization also crucial for progress
