NEWS FLASH: AI can’t do everything!

You’d be forgiven for thinking that we can all put our feet up now because AI can sort out anything important. It is categorically not the case.

I heard my friend’s daughter recently talking about wanting to be a hairdresser and we all agreed that it was a good option and was futureproofed from being taken over by AI. 

My second job at the moment is house and dog/cat sitting which certainly needs a human touch. I was even served a TikTok side hustles video that suggested offering my services picking up dog poo for wealthy but time-poor individuals! Maybe someone will programme a drone for that in the future, but I expect it’s low on Elon Musk’s list of priorities right now.

I’m ignoring the sparkly ‘AI Assistant’ offering to help me write this but then that’s just me. And it always will be.

When I’m feeling a bit down in the dumps about the state of the communications landscape, I remember the things that AI can’t do but us lowly humans at Accuracy Matters can. 

These include:

• Using context and nuance to unpick ambiguity: AI relies on the data, prompts and programming it’s given and therefore struggles to solve common editorial issues around clarity and ambiguity. Often, I would need to have a conversation with the author – either about the query itself or prior to editing in a comprehensive briefing meeting – in order to eliminate ambiguous phrasing.

   

• Creative problem-solving: Sometimes editors reach a point in a project where it’s better to deviate from the client’s brief and/or style guide, taking a different editorial approach to achieve a better result for the client. Sometimes that becomes apparent at the briefing stage, before the work has even started. And sometimes clients aren’t sure about what stage their material is at, so don’t know which type of editorial intervention to ask for. In all these scenarios, you would need a human editor – or at the very least, a specialist who can determine the best approach for your piece of work.

   

• Making ethical judgements: Choices about phrasing, tone, emphasis or even what material to include and what not to include can have an ethical component, which a human editor is currently best placed to arbitrate.

   

• Reviewing highly sensitive/confidential material: Apps and platforms are open to leaks and hacking, and it’s difficult to know what’s being done with the data we are feeding into them. I would be extremely cautious about running confidential material through any third-party software without thorough review and testing first (testing both the security and the efficacy of the process for that type of material).

   

• Understanding emotional context: Human emotions are complex and multifaceted – and AI doesn’t even come close to feeling them, although it can provide a very realistic simulation of feeling. Reading about the puzzle of whether AI has feelings takes me right back to my studies of philosophy of mind and philosophy of science (I knew it would come in useful at some point!).

   

• Spotting its own hallucinations or biases: Amazon scrapped an AI recruitment tool because the historical data it relied on favoured male applicants. AI chatbots confidently supplying apparently factual answers have been found to fabricate everything from legal cases and citations to company policies and summer reading lists.

At Accuracy Matters, we’re working hard to keep up with progress and recognise the advantages of building AI and automation into project workflows. However, I apply a healthy dose of scepticism and I’m always on the look-out for places where AI doesn’t work as well as humans – in many cases it is clearly the worse option.

Further reading
Hans Volker, IABAC blog, Limitations of AI: Current Challenges and Future Fixes
Akhil Bhardwaj, Nature, Artificial Intelligence and the limits of reason