Fireside Chat, Interview
Graphcore’s Nigel Toon on advances in semis and an AI-powered future
- Approximately 150 million intelligent machines capable of intelligence comparable to or exceeding humans will be delivered within an unspecified timeframe to operate companies, advise governments, and eventually govern.
- These machines are predicted to defeat all human sporting champions in every category within an unspecified timeframe.
- Intelligent machines, described as "children," are expected to eventually assume responsibility for caring for the aging population.
- Scientific breakthroughs in protein modeling to cure diseases like cancer without side effects are projected within the next 10 years.
- Fusion energy experiments are likely to generate more energy than consumed within the next 20 years, with commercial viability anticipated within an additional 5 to 10 years.
- AI will play a critical role in solving complex fusion challenges, particularly plasma control, within the next 20 years.
- Value within the AI ecosystem is expected to shift from infrastructure providers to application developers within the next 5 to 10 years.
- Direct programming will cease to be a necessary skill for most people, scientists in biosciences and drug discovery, and financial traders, as AI frameworks enable prompt-based problem solving in the near future.
- Semiconductor improvements will continue but will progress much more slowly over the next few years due to S-curve limitations of current technology.
- Recent and future gains in GPU compute are driven primarily by precision reduction rather than Moore's Law improvements.
- The ability of quantum computers to scale to millions of qubits to overcome noise issues remains uncertain.
- Developing functional computers based on molecular approaches is estimated to be at least 20 years away.
- The speaker and their peers may become obsolete before some audience members in the unspecified near future.