We’ve all seen the jokes by now. How we imagined a future where robots would do the dishes so that we could spend more time creating art, and instead, somehow, the robots are making all the art, leaving us with the chores.
Many ways in which AI pops up in everyday life are frustrating: bland emails, uncanny artwork, incorrect information, visual slop.
Yet in other ways, AI is providing solutions: it’s demonstrably helpful when it comes to cancer screening, climate forecasting, drug discovery, or finding ways to make systems more efficient.
Given the immense resource demands of training and operating large language models, it’s worth homing in on what we use AI for – and when to avoid it.
Avoid using AI where it struggles
If there’s a query that can be typed into a search engine rather than an AI service, use the search engine – it’s more likely to lead you to a correct answer.
AI is wrong roughly half the time, concluded a fact-checker for the technology magazine Wired in May after analysing studies on AI accuracy. One report found that AI responses from search engines were inaccurate 60 percent of the time. A BBC study found that AI assistants, like the ones that generate news summaries, were wrong 45 per cent of the time.
Distinguishing truth from falsehood isn’t the only task AI struggles with. “There are a number of things that AI is traditionally really, really bad at,” says Ulrich Speidel, a senior lecturer in computer science at the University of Auckland. “This includes what we call ‘planning under constraints’.”
For instance, ask AI to plan a road trip between Auckland and Wellington, driving no more than 200 extra kilometres off the route while visiting as many of a list of top sights as possible. It will most likely ignore your specifications and take you wildly off course, says Speidel. “It might drop you unceremoniously in Palmerston North – or it might get lost in Northland.”
AI also has difficulty with numerical analysis. If a scientific paper has been fabricated by AI, the tell is often in the data. A paper might sound extremely convincing, but made-up numbers are often easy to spot, says Speidel: the giveaway can be as obvious as the introduction and conclusion containing different numbers for the same values.
“And so, that can be fairly scary at times,” he says, “especially when we then start relying on it”.
Save AI use for where it really matters
It makes sense to use AI for what it’s good at: advanced pattern recognition, large-scale data analysis, and automation of menial tasks. AI can already spot signs of cancer in mammograms and find new ways of folding proteins to create new medications. In the future, it could help improve renewable-energy generation.
“If we use artificial intelligence wisely, it might get us closer to solutions,” says Speidel. “For example, looking at how we can make solar or wind turbines more efficient.”
Solar panels currently convert less than 25 percent of the sun’s energy into electricity. “If we could get that up to maybe 50 percent over time, then that might be a very, very, good use of AI,” says Speidel, “because it would allow us to deploy solar in a lot more places”. Not to mention that it might help AI cover its own costs energy-wise.
“We might be able to solve some of the world’s problems with it,” says Speidel. “But if all we use AI for is to generate deepfakes or make people redundant, then that’s probably not going to be the big salvation that everybody’s dreaming of.”
Advertise products and services devoid of AI
Being AI-free is increasingly a marketable position, with some companies pledging not to use AI to advertise their products.
The same goes for services, says Speidel. “It can also be a competitive advantage, saying, ‘Look, don’t leave your valuable projects to a machine, we’re real humans that can provide real solutions to real problems’.”
This is a significant edge that humans have over AI. “One of the things that we haven’t really seen with AI agents so far is what I would call true creativity in the sense that they’re actually solving problems,” says Speidel. “Every time I’ve tried to use AI to actually get it anywhere near a true problem – one that wasn’t sort of regurgitating something that had already been solved somewhere – it faltered.”
Meanwhile, community organising is ramping up, as people oppose unfettered AI development or attempt to clarify reasonable goals and purposes for it. Investigative journalist Karen Hao recently co-founded the website The AI Resist List to catalogue civil action against the scale-at-all-costs approach to AI development.
Ensure there’s clarity around costs and benefits of data centres
If clarity is needed on the best and worst uses of AI, the same goes for the environmental costs and economic benefits of data centres. This is an issue close to home.
Criticism of the proposed $3.5b hyperscale Southland AI data centre stems from the fact that its power and water use arrangements are unclear. “At the moment, a lot of these projects are surrounded by this opacity,” says Angus Dowell, a PhD student at the University of Auckland studying the links between big-tech digital infrastructure and governments.
Data centres should be clear about their impacts, he says. “If they are as vital as people say, then should we not have a clearer picture of the pros and cons? At the moment, a lot of it is speculation.”
There are benefits to New Zealand from the data centre: a proposed fibre-optic cable connecting Bluff with Australia would improve New Zealand’s network resiliency, says Dowell, and the centre itself brings medium-to-short-term investment to the country. “There will be jobs from that,” he says.
But it’s not clear how extensive the project’s benefits will be. Other larger data-centre developments serve as a warning, says Dowell. “We’ve had big-tech data centres here from Microsoft and Amazon Web Services, with very little evidence that those have benefited us over alternative local players. In fact, we’ve lost hundreds of millions in tax to Amazon and Microsoft who artificially reduce their New Zealand tax burden – effectively rerouting their profits offshore.”
Which begs the question, says Speidel: “If the benefits accrue elsewhere and the only thing that we’re getting paid for is cheap power, then is this really a worthwhile exercise for the country?”
The world is facing unprecedented environmental challenges. Planetary Solutions, an initiative of the Sustainability Hub at Waipapa Taumata Rau, University of Auckland, and Newsroom, explores these issues – and the practical ways we can all be part of the solution.
