The conversation about AI and the workplace is understandably dominated by the downsides – after decades of automation eliminating manufacturing jobs, people in the service sector are worried about being replaced by “robots”.

But every technological shift creates as well as destroys jobs. Artificial intelligence – at least in its current iteration, which uses large language datasets to create text, audio and video – is no different. What is, perhaps, surprising is the type of jobs it will create.

What kind of jobs are being created right now?

The most visible and obvious new roles are for those with the coding and development skills to help build AI models or adapt them for particular purposes. The UK government is actively pursuing AI startups and trying to offer support for them. Google owner Alphabet’s AI division, DeepMind, was founded in London and is still largely based here.

These roles aren’t all AI specialists – there are startups dedicated to customising AIs, such as ChatGPT, to be suitable for particular roles in specialist businesses, such as acting as customer service agents for specific industries, or even online sales assistants. Roles specialising in building and creating AI models are attracting the biggest salary premiums, but there are highly skilled and highly paid roles across coding and development.

Are new types of jobs being created already?

There are drastically different companies dipping their toes into the artificial intelligence revolution, though at this stage on a smaller scale than the new development roles. One that has attracted particular attention is Netflix’s advert for a product manager for its “machine-learning platform”, not least because the listed “compensation” is $300,000 to $900,000 (£240,000 to £710,000).

Against the backdrop of writers and actors striking partly over the impact of AI on their own roles, the job advert has sparked a significant backlash. It does not specifically mention automation of writing or acting roles but does include phrasing such as “powering innovation” and offering “personalisation to members” with the “highest possible impact”.

Not all of the new roles being created by AI offer a top-tier salary, however – the UK’s NewsQuest group advertised an “AI-powered reporter” role in April. The new recruit would “work with an AI system to help write your news articles”, ensuring “all content produced meets legal and ethical standards” while also being required to “meet and exceed page view targets”. The reward for all that was a salary of £22,000.

This type of role is based on the idea that, at present, using AI models and getting good results is a skilled task. Using very specific prompts yields better results than vague queries, and users need to have an eye for when an AI model seems to be giving factual information but is in fact “hallucinating” – making up untrue “pseudo facts”.

Should we all be looking to retrain for this type of role?

We don’t know, but it’s absolutely not a given. These early roles created by AI might not be the strongest long-term prospect. “In a nutshell, I would say it [the best prospect] is not programming, because coding will change,” says Dr Caitlin Bentley, a lecturer in AI education at King’s College London – suggesting that people coding AI may in time see some of their skills superseded.

And the new profession of “prompting”– giving AI very precise instructions to get better quality output – might also be shortlived, she suggests. “Prompting, I also think, is going to die soon because AI will get better. Instead, we need better interrogation and translation skills.”

So some roles created by AI are temporary?

This is definitely true – AI models rely on being “trained” on real-world data. For language, this data exists naturally, but for other uses (such as identifying traffic lights) AI needs specific instruction.

These roles will obviously not last for ever. Self-driving companies, for example, all employ “test drivers” for their driverless cars. But many roles serve a similar purpose, albeit far less visibly. A stark example came in reporting by the Bureau of Investigative Journalism last year showing that Amazon workers in UK warehouses had to sort items in a particular way so that AI could detect it.

However, workers in India and Costa Rica were then being paid to spend all day confirming ambiguous sorts to help train the AI, often for hours at a time with only seconds for each one – all with the ultimate aim that AI may then replace them.

“It’s a strange situation in which two humans are now involved in a job that one human used to do,” says someone familiar with the matter. “But they’re both being asked to behave like robots in order to help train robots that could eventually take both of their jobs.”

These “ghost in the machine” roles tend to be grim ones, focused either on repetitive tasks that AIs are still not sufficiently reliable at, or else on moderation of social media, for the same reasons.

Which roles will last?

Alison Gow, a former senior executive at Reach (publisher of the Mirror, the Express and many local titles), who has looked into AI in media companies, thinks the best prospects for its long-term use are where the technology is used to enhance or transform existing roles, taking on monotonous tasks.

“I’d expect to see AI editor roles being introduced, because someone is ultimately going to need to be responsible, and adding a lens of journalistic scrutiny to generated content,” she says.

Gow sets out the optimistic version of events for the news industry: “I expect we’ll see more automation of utility content such as weather reports and match reports in sport, and I’d hope we’ll also see AI used for greater accessibility – so transcription or translation services, or text to audio.”

She calls for “real-world workable benefits for a newsroom, not doing shiny stuff for the sake of it”.

I’m not technically minded – are there any upsides in this for me?

Legal and human resources within businesses might be rife for AI, it has been suggested. Gow is among those who take a different view, suggesting AI might require quite the opposite.

There will be many difficult issues to tackle in the courts, in parliament, with unions and other – on matters from copyright law to employment law and beyond – and businesses will need bigger teams to tackle them, she says.

“HR and legal departments will be expanding to incorporate more expertise or contracting to independent experts,” Gow concludes. “This is going to open up a whole new field of complaints.”

There is also, perhaps, hope for jobs requiring a human touch. In the same way that some companies brought their call centres back to the UK as a mark of premium quality, having actual human customer service agents might similarly become a mark of quality. This could lead to a higher premium being placed on so-called “soft skills” and service skills.

Whatever AI does to the workplace, it is not necessarily the people coding the technology who will be the winners.