Artificial intelligence — an aid to thought, not a replacement

Daily Maverick

13 hours ago

'The danger of outsourcing our thinking to machines is that we still have to live in the world they end up creating. That's too big a responsibility to just hand over.'
When ChatGPT briefly went offline last week, it felt, as journalist and writer Gus Silber put it, 'as if the sun had fallen from the sky'.
Speaking on a Jive Media Africa webinar on the subject of 'Machines are writing – do we trust the message?', Silber and other panellists tossed around concepts of 'Uberisation', 'forklifting' and 'outsourcing' to get to grips with AI technology and its ethical pitfalls.
Silber noted that in just a few years, AI had morphed from novelty to necessity and is now deeply woven into daily work across media, academia and science communication.
Its seductive convenience allows us to 'outsource thinking to a machine', said Silber, while noting both the potential and the perils of doing so.
Fellow panellists, science communicator and champion of language equity in science Sibusiso Biyela and Michelle Riedlinger, associate professor in the school of communications at the Queensland University of Technology, agreed, in a discussion peppered with metaphors, to highlight the divisions of labour in the partnership between technology and humans.
Introducing the webinar, Jive Media director Robert Inglis said that 'artificial intelligence, particularly generative AI, is reshaping both the practice of research and the craft of science communication. This impact is felt by researchers, by science communicators and by others working at the intersection of science, society and media and especially those who are grappling with how AI tools influence credibility, ethics and public trust.'
While many fret over the elimination of jobs and the technological encroachment on the preserve of what it means to be human, Silber readily calls himself a Utopian on the subject, believing 'it's ultimately going to be of good for humanity'.
Silber notes that the reach of AI, 'originally a niche technology, has expanded dramatically, driven by advances like fibre, broadband and always-on connectivity. Tools such as ChatGPT now serve as default knowledge engines, sometimes even surpassing Google.
Being able to 'outsource a lot of your thinking to, effectively, a machine,' he said, 'tempts users to let AI handle increasingly complex tasks'.
In academia and media, some rely heavily on AI-generated content, resulting in a sameness of voice: 'It sounds human, but it sounds human in a very kind of generic and samey way.' While AI offers powerful assistance in tasks like transcription – 'you can transcribe two hours' worth of interviews in five or ten minutes' – the risk is that its convenience leads to 'creative atrophy'. It's 'a real temptation, a kind of 'tyranny of ease', where you can just prompt the AI to write essays or theses. That scares me because it risks giving up your creative energy.'
Collaborative use
He nevertheless enthuses about the rise of multimodal AI and mentioned tools like Whisper, Notebook LMand Genspark AI, which are already revolutionising research, communication and creative industries. But he draws clear boundaries: 'I draw the line at outsourcing full creative processes to AI.' Instead, he advocates using AI collaboratively, augmenting human thought rather than replacing it.
'We're lucky to live in this creative technical renaissance. We can't go back to how things were before. My advice: explore these tools, break them, have fun and find ways to use them collaboratively. Let machines do the heavy lifting while we focus on human creativity.'
Anxieties, however, are pervasive, said Riedlinger. Her research shows that news audiences 'found familiar concerns: misinformation, copyright, elections, job displacement.' But people weren't rejecting AI outright; 85% wanted transparency; visible labels, a kind of 'nutritional label' for AI-generated content.'
She said there's a growing 'authenticity infrastructure' emerging, with companies like Adobe working on labelling multimodal content. Audiences want AI to support, not replace, human journalists and science communicators. 'The key is to keep humans in the loop, to ensure creativity, empathy and accountability remain central.'
To help navigate this, Riedlinger reached for metaphors.
First, she said, contrast 'forklifting versus weightlifting. Forklifting covers repetitive, heavy tasks – transcription, translation, drafting – where AI helps move things efficiently but under human guidance. Weightlifting represents skills that build strength: framing stories, interpreting data, learning audiences. These are areas we risk weakening if we outsource too much to AI.'
The second is the 'Uber metaphor'. 'You can make coffee yourself or order it through Uber. It's convenient, but hides labour behind the scenes: the barista, the driver, data centres. Generative AI feels equally magical but isn't free; there are hidden costs in energy use, data scraping and ethical concerns. Before outsourcing, we must consider these unseen consequences.
Hallucinations and bias
'In global studies, people increasingly recognise AI's limits: hallucinations, biases in gender, race, geography and class. Some see AI as a calculator, improving over time, but that's misleading. Calculators give fixed answers; generative AI doesn't.'
Reaching for yet another metaphor, she said 'it's more like a talking mirror from a fairy tale', generating fluent, tailoredand sometimes flattering responses, but blending truth and invention in a way that can flatten creativity and make unique ideas more generic.
'Authenticity, trust and disclosure are vital. We need consistent labels, audience controland clear public policies.'
This, said Riedlinger, will build trust over time. 'Science communicators must reflect on each task: Is this forklifting or weightlifting? Am I calling an Uber for something I should craft myself? Science communication deserves thoughtful tools and thoughtful users. We need to ensure that our publics have authentic interactions. '
The watchwords, when dealing with AI, are: 'Disclose. Collaborate. Stay in the loop as a human. Design for trust.'
Picking up on the trust, or mistrust, of the machine, Biyela said 'there's a lot of antagonism around AI, especially with articles not disclosing if they're AI-assisted. When audiences hear something was generated by AI, they often turn away. It becomes less of an achievement if it wasn't really done by a human.'
But, he said, 'audiences (and ourselves) need to understand AI's limitations and how it actually works. We call it artificial intelligence, but it's in no way intelligent. It's an automaton that looks like it's thinking, but it's not. It's a clever prediction model using computing power to make it seem like it's thinking for us. But it's not. The thinking is always being done by people. AI never does anything; it's always us. What it produces has been trained to give us what we want.'
Biyela emphasises that 'You're the human in the loop' and have to account for every line an LLM is asked to produce. 'If it summarises something you haven't seen, you have to check it. It makes the job easier, but it doesn't perform it.'
Caveats aside, Biyela says 'generative AI also offers potential in communicating science in underserved languages, like African languages.
Driving AI
In his conclusion, Inglis, too, reached for a metaphor to guide how science communicators and other professionals and students should engage with AI: 'We would never jump into a car without having learnt to drive the thing. Now you've got these tools at our disposal and we'll use them, but we've got to be aware of the dangers that using them for the wrong things can bring about in the world.'
In short, the panel agreed that in the partnership between AI and people, AI is good at the 'forklifting' work: sorting, calculating, transcribing, processing vast amounts of data quickly, but that humans still carry the mental load: setting priorities, interpreting meaning, understanding context, reading emotions, anticipating unintended consequencesand ultimately taking responsibility for decisions.
Inglis further reflected: 'Our work in science communication is to play a part in solving the complex challenges we face and to ensure we do so in ways that build a better future for society and for the planet.' He cited a recent study by Apple, which reveals just how bad large reasoning models are when it comes to deep reasoning, having been found to face a 'complete accuracy collapse beyond certain complexities'.
'This underlines the need for human operators to use these tools as an aid to thinking, not as a replacement for thinking. That grappling with complex ideas is exactly what we're doing with this webinar series – these kinds of answers can't be scraped from the web, they need to be generated and discovered through exploration, conversation, dialogue and skilful engagement.
'The danger of outsourcing our thinking to machines is that we still have to live in the world they end up creating. That's too big a responsibility to just hand over because it's easier than engaging with tough issues. It's lazy and at this time in the history of our planet, we can't afford to be lazy.' DM