Trevor Cohn
Professor Trevor Cohn works at the University of Melbourne in Australia, researching natural language processing, with a focus on translation, multilingual methods and tools for low-resource languages. His work combines machine learning, computing and linguistics.
Website
My favourite AI in culture project at the moment is… DeepMind’s amazing game-playing systems, which learn to play games purely from experience. Their work has shown how to play old Atari games, and more recently the classic “Go” game, which has long been held out as one of the hardest challenges for AI. There’s still a big jump to move to “real” problems, like language understanding, but their work is a key stepping-stone towards better interactive systems.
By 2030 I would like to have an AI app which… monitors and “understands” what is going on in the brain and how this relates to language and cognition. Our understanding of the brain is still pretty cursory as is our understanding of language, language learning and other key cognitive tasks. AI has been ritually overpromising an understanding of the brain and human cognition in the near future, so I’m not holding my breath for substantial progress in the next decade.
My worst AI nightmare is… the hype cycle goes into overdrive, followed by a crash based on the stark realisation that every technology has limits! The adage goes “garbage in, garbage out” which applies perfectly to machine learning-based AI systems, for which high quality data, in quantity, is a fundamental requirement for decent performance.
The future of AI needs… a clear moral and ethical framework. A lot of key questions come up when dealing with systems that track individuals, make important decisions, distil or exacerbate biases from their input data or risk compromising individuals’ privacy. How we handle these questions will be key in AI becoming a social good moving forward.
Trevor Cohn's contribution: Will machine translation ever be on par with humans?
Website
My favourite AI in culture project at the moment is… DeepMind’s amazing game-playing systems, which learn to play games purely from experience. Their work has shown how to play old Atari games, and more recently the classic “Go” game, which has long been held out as one of the hardest challenges for AI. There’s still a big jump to move to “real” problems, like language understanding, but their work is a key stepping-stone towards better interactive systems.
By 2030 I would like to have an AI app which… monitors and “understands” what is going on in the brain and how this relates to language and cognition. Our understanding of the brain is still pretty cursory as is our understanding of language, language learning and other key cognitive tasks. AI has been ritually overpromising an understanding of the brain and human cognition in the near future, so I’m not holding my breath for substantial progress in the next decade.
My worst AI nightmare is… the hype cycle goes into overdrive, followed by a crash based on the stark realisation that every technology has limits! The adage goes “garbage in, garbage out” which applies perfectly to machine learning-based AI systems, for which high quality data, in quantity, is a fundamental requirement for decent performance.
The future of AI needs… a clear moral and ethical framework. A lot of key questions come up when dealing with systems that track individuals, make important decisions, distil or exacerbate biases from their input data or risk compromising individuals’ privacy. How we handle these questions will be key in AI becoming a social good moving forward.
Trevor Cohn's contribution: Will machine translation ever be on par with humans?