It’s a lot further than you think, but it has already started. The singularity is the moment AI will improve itself at a rate faster than we can. From here it will unleash incomprehensible brilliance that could devastate humankind. Or so the movies say.
For the time being, we are living in an ordinary world of ordinary problems that AI has the opportunity to fix or make worse. Firmly in the experimentation phase, innovation is ripe with unintended consequences. Mark Zuckerberg’s former motto, move fast and break things, has been elevated to sacred credo throughout tech. In Machines Behaving Badly: The Morality of AI, the professor and all-round guru Toby Walsh makes the case that for AI it would do better to move slow and for god’s sake be careful.
AI is not separate from but a mirror of the world of its creation. It inherits the character, flaws and biases. This means that Alexa can be racist and sexist. What’s more, AI continues to concentrate power in the hands of the powerful, enabling verticals of control and surveillance that would be impossible in the most efficient human organisation.
We are indeed edging closer to a machine-assisted dystopia: today Amazon warehouse staff have their hand movements tracked and can receive an automated notice of their termination without any human involvement. Uyghurs in Xinjiang are tracked via facial recognition software, while the likelihood that American prisoners will reoffend is calculated via AI analysis. Even when humans remain the ultimate arbiters, automation bias will tip the scales in favour of the machines’ recommendations. We simply trust too much in the magic of tech.
This informs the purpose of Machines Behaving Badly: Walsh demystifies the utopian promises of AI and attacks the dangerous but resilient fallacy that computers never lie.
The lie has heritage: from the postwar generation, it was promised that machines could help us shed our messy prejudices and subjectivity to arrive at a cleaner objective reality. Walsh’s point is that clusters of virtual instances running machine-learning algorithms will not achieve escape velocity from our foibles any more than sorting machines and punchcards did. Yet the stakes have been raised and they are out in the fragmented, under regulated wild of the market.
As a guru who could land a plum job in the private sector, the author remains in academia at UNSW. Yet he admits the cutting edge has graduated from universities and bedded down in the world’s Silicon Valleys. The context of invention is again important: here, agile is the dominant methodology. Not just a buzzword, it means delivering an imperfect product to market then incrementally improving it. Agile is baked into the commercial DNA of every startup and Big Tech outfit. Anything more cautious is viewed as clunky and old-fashioned and unlikely to attract venture capital that will go to the swift and brash.
Walsh also concedes, if a little bashfully, that AI inherits the lack of diversity in tech. This isn’t only race and gender, but the specific character of the nerds inventing it. The author’s partner can see them coming a mile off. That must be one of yours? she asks at an airport on their way to an AI conference. She’s right almost every time. The narrowness inherent in specialisation, with wild demand for their skills, the hype around their achievements and the power invested in their decision-making, all spell hubris.
Yet it’s not all bad. There is tremendous capacity for AI to help out with our thorniest problems, such as the climate crisis, medical decision-making and analysis. In the most interesting portion of the book, Walsh explains that by attempting to replicate our own intelligence we uncover and appreciate which aspects make our intelligence unique. An AI revolution could make us more, rather than less, human.
Whatever the case, Walsh states that AI could be our final invention. There are two ways to read this idea – the more optimistic, that we can appreciate our humanity while machines will do the inventing. The second need not be transcribed here.
Machines Behaving Badly: The Morality of AI by Toby Walsh is published by La Trobe University Press, $32. 99.
The Booklist is a weekly newsletter for book lovers from books editor Jason Steger. .