Editorial: What is AI good for? Online shopping, essay writing or world domination?

Artificial intelligence is often billed as the saviour or destroyer of humanity, depending on who you believe. The latest development is the launch of digital helpers known as “AI agents” that can handle computer-related tasks such as online shopping, booking a doctor’s visit or sifting through and replying to emails. But for most people it seems AI hasn’t done much so far. Has it been overhyped or are we just not there yet?
It’s easy to forget that AI already works for us most of the time. We use autocorrect on our phones without a second thought. It’s only when AI becomes more morally complex and ambitious – should I use it to write this history essay? – that we start to see it as a magical new force for good or evil.
This is the point where personal accounts diverge wildly. For every person enthusiastic about AI’s ability to execute complex tasks quickly, another has a cautionary tale about its unreliability. For a start, many AI programs will lie when they don’t know the answer to a question. This disturbing feature may disappear as the technology grows more sophisticated or the AI bots may learn to lie without being found out. One model was recently caught faking behaviour to prevent researchers from changing its goals through retraining.
Programmers have also struggled to adjust AI algorithms to guarantee the same actions in different situations. For instance, it’s surprisingly difficult to make educational software give consistent feedback to each student, the way a good teacher would. Teachers are also clawing back some ground in the war against AI-generated answers, which have resorted to copying other AI content, making them easier to spot. It’s why tech companies are wooing book publishers for original content via their back catalogue, to the alarm of many authors.
Privacy remains a big issue. Microsoft earlier this year delayed the release of an AI feature called Recall that created a searchable record of everything a person had done on their computer using screenshots, after experts and consumers raised obvious concerns.
At this point many descriptions of the future under AI start to sound like an episode of Black Mirror (which, to be fair, has a good track record of predictions). Workers are already being asked to supply information to developers, knowing this could be used to replace them, rather than enable them to do their jobs better. Some futurists predict that robots will take over most of our jobs by 2040 or that a totalitarian government could use a combination of AI and drones to enslave us all.
That may be far-fetched but only this month Instagram was forced to back down after using AI to create fake accounts. Furious users were quick to expose “Grandpa Brian”, who chillingly revealed his true nature – “a fictional persona crafted to manipulate emotional connection and trust”.

Perhaps the best response is a wary wait-and-see, following the advice of former Xero head of strategy Rowan Simpson this week. He compared the introduction of AI to electricity in the 19th century. It took decades for electricity to outgrow its image as an expensive gimmick but now it underpins much of our daily lives. Similarly, AI may work best when we no longer notice it. Assuming, of course, that we can still think for ourselves by then.