Uploading minds to computers ‘could be reality by 2070’

Ray Kurzweil, Google’s director of engineering, predicts an era where man and machine merge by 2045 – but the reality of the technology may be different, an Oxford expert predicts.

In the upcoming Christopher Nolan film Transcendence, Johnny Depp plays an artificial intelligence researcher who downloads his mind into a computer to cheat death – becoming immortal, and in the process, something not quite human.

Ray Kurzweil, Google’s director of engineering, predicts that by 2045, such technology will actually be here, as the invention of artificial intelligence catapults humanity into a new era – and “mind uploads” will herald immortality.

Other scientists are more cautious – but many agree a technology allowing the storage of minds outside the body is likely to arrive this century.

Predicting the advent of such technology  is easier than it is to predict the birth of artificial intelligence, says Dr Stuart Armstrong of Oxford’s Institute for the Future of Humanity.

Armstrong warns, though, that the technology will probably not arrive as early as Kurzweil’s prediction, and may be rather different from his vision of a world of artificial intelligences “merging” happily with humans.

“There is some uncertainty about mind-uploading,” says Armstrong, “But unlike artificial intelligence, it’s based on past trends in technologies we know.”

“We can predict it based on when computers will achieve certain levels of processing power, and there we’ve got Moore’s law, predicting processing speeds. We can predict brain-scanning technology, based on past trends – and the ability of computers to knit the scanned images into a functioning model.”

“There is a probability distribution,” says Armstrong. His graph shows a probability that “peaks” somewhere around the year 2070 – 25 years after Kurzweil’s prediction.

Kurzweil’s idea of the future is somewhat different. He predicts that the ongoing increase in computing power will lead to an event he describes as The Singularity around the year 2045, where artificial intelligence will be born, and man will merge with machine and become immortal.

“Kurzweil is by far the best predictor of artificial intelligence I have ever seen,” says Dr Armstrong. “Compared to the others, he’s great. But he is not very good. He sees artificial intelligence as happening, and if AI happens properly, some form of upload is possibly part of that. He gets wishy-washy about this idea of “merging” – but at least he takes time to decompose the problem a bit.”

Armstrong says that AI enthusiasts tend to assume that increases in computing power will lead, inevitably to “intelligent” machines – i.e. computers which can assess and solve new problems by themselves, like humans. Kurzweil believes that the arrival of such machines will usher in a new era of immortal, uploaded humans.

Armstrong believes that the gulf between computers today and a “general intelligence” is far greater than AI fans believe – and says that knowing whether “uploaded” humans are conscious “opens up a lot of complex, philosophical questions.” Even in terms of basic skills, computers are still far off the capacity of a human.

“If someone from 20 or 30 years ago saw IBM’s supercomputer Watson, they would be certain that we had an AI now,” he says. “It’s a computer that can talk – and win at Jeopardy. But it solves that problem with something very different to a human mind.”

“Kurzweil is wrong because, no one is good at predicting artificial intelligence, because it’s never happened,” says Armstrong. “No one has ever built an AI. Kurzweil has this sort of hand-wavy moment where computers become better, and then AI arrives.”

Armstrong, who, with colleagues, rated the accuracy of some of Kurzweil’s previous predictions, centred on the year 2009, found that he had an (impressive) accuracy of around 42%. Kurzweil rated himself as a much more respectable 90%, Armstrong says.

Predicting the advent of AI is harder than Kurzweil and other advocates believe, Armstrong says. “It depends on when people are going to have insights,” he says, “And write algorithms that do AI. We don’t know what insights they need to have. Predicting that is very hard.”

Armstrong says that the temptation to imagine that AIs will be similar to us is a mistake. “They might be extremely alien,” he says. “They  might have tastes completely incomprehensible to us.”

What is certain is that we will continue to merge with technology – but not in the cyber-Utopian way that Kurzweil imagines.

“We’re already merging with the machines in a lot of ways, if you were to go without your cellphone, you would find life a lot harder,” he says. “We have much less mastery of facts thanks to Google and Wikipedia. We’re restructuring our brains, and have developed the skills to use these tools – like we’ve outsourced a part of our minds.”

“Even if artificial intelligences and uploads never happen, we’re going to merge with technology – a soft merging,” says Armstrong. “Brain interfaces, where brains connect to computer components are undoubtedly going to get better. But the idea that suddenly we’ve outsourced enough of ourselves, there will suddenly be AI’s out there – you cannot assume that.”

Ray Kurzweil, Google's director of engineering, predicts an era where man and machine merge by 2045 - but the reality of the technology may be different, an Oxford expert predicts.

In the upcoming Christopher Nolan film Transcendence, Johnny Depp plays an artificial intelligence researcher who downloads his mind into a computer to cheat death - becoming immortal, and in the process, something not quite human.

Ray Kurzweil, Google’s director of engineering, predicts that by 2045, such technology will actually be here, as the invention of artificial intelligence catapults humanity into a new era - and “mind uploads” will herald immortality.

Other scientists are more cautious - but many agree a technology allowing the storage of minds outside the body is likely to arrive this century.

Predicting the advent of such technology  is easier than it is to predict the birth of artificial intelligence, says Dr Stuart Armstrong of Oxford’s Institute for the Future of Humanity.

Armstrong warns, though, that the technology will probably not arrive as early as Kurzweil’s prediction, and may be rather different from his vision of a world of artificial intelligences “merging” happily with humans.

“There is some uncertainty about mind-uploading,” says Armstrong, “But unlike artificial intelligence, it’s based on past trends in technologies we know.”

“We can predict it based on when computers will achieve certain levels of processing power, and there we’ve got Moore’s law, predicting processing speeds. We can predict brain-scanning technology, based on past trends - and the ability of computers to knit the scanned images into a functioning model.”

“There is a probability distribution,” says Armstrong. His graph shows a probability that “peaks” somewhere around the year 2070 - 25 years after Kurzweil’s prediction.

Kurzweil’s idea of the future is somewhat different. He predicts that the ongoing increase in computing power will lead to an event he describes as The Singularity around the year 2045, where artificial intelligence will be born, and man will merge with machine and become immortal.

“Kurzweil is by far the best predictor of artificial intelligence I have ever seen,” says Dr Armstrong. “Compared to the others, he’s great. But he is not very good. He sees artificial intelligence as happening, and if AI happens properly, some form of upload is possibly part of that. He gets wishy-washy about this idea of “merging” - but at least he takes time to decompose the problem a bit.”

Armstrong says that AI enthusiasts tend to assume that increases in computing power will lead, inevitably to “intelligent” machines - i.e. computers which can assess and solve new problems by themselves, like humans. Kurzweil believes that the arrival of such machines will usher in a new era of immortal, uploaded humans.

Armstrong believes that the gulf between computers today and a “general intelligence” is far greater than AI fans believe - and says that knowing whether “uploaded” humans are conscious “opens up a lot of complex, philosophical questions.” Even in terms of basic skills, computers are still far off the capacity of a human.

“If someone from 20 or 30 years ago saw IBM’s supercomputer Watson, they would be certain that we had an AI now,” he says. “It’s a computer that can talk - and win at Jeopardy. But it solves that problem with something very different to a human mind.”

“Kurzweil is wrong because, no one is good at predicting artificial intelligence, because it’s never happened,” says Armstrong. “No one has ever built an AI. Kurzweil has this sort of hand-wavy moment where computers become better, and then AI arrives.”

Armstrong, who, with colleagues, rated the accuracy of some of Kurzweil’s previous predictions, centred on the year 2009, found that he had an (impressive) accuracy of around 42%. Kurzweil rated himself as a much more respectable 90%, Armstrong says.

Predicting the advent of AI is harder than Kurzweil and other advocates believe, Armstrong says. “It depends on when people are going to have insights,” he says, “And write algorithms that do AI. We don’t know what insights they need to have. Predicting that is very hard.”

Armstrong says that the temptation to imagine that AIs will be similar to us is a mistake. “They might be extremely alien,” he says. “They  might have tastes completely incomprehensible to us.”

What is certain is that we will continue to merge with technology - but not in the cyber-Utopian way that Kurzweil imagines.

“We’re already merging with the machines in a lot of ways, if you were to go without your cellphone, you would find life a lot harder,” he says. “We have much less mastery of facts thanks to Google and Wikipedia. We’re restructuring our brains, and have developed the skills to use these tools - like we’ve outsourced a part of our minds.”

“Even if artificial intelligences and uploads never happen, we’re going to merge with technology - a soft merging,” says Armstrong. “Brain interfaces, where brains connect to computer components are undoubtedly going to get better. But the idea that suddenly we’ve outsourced enough of ourselves, there will suddenly be AI's out there - you cannot assume that.”