Artificial intelligence is getting very good, very fast. Whether it’s music, text, code or imagery, the time when it was reliably possible to tell the difference between AI and human outputs is disappearing at an alarming rate.
Yet for all their wizardry, AIs can also be quite useless. They make things up and misunderstand instructions. They are brilliant as toys but incompetent as assistants.
All this makes it hard to know how to put AI into perspective. Is it the most important technological trend since the iPhone? Or since the industrial revolution? At this distance, it’s hard to say.
There are industry measures to assess the intelligence of AI models, known as benchmarks. These too show rapid improvement.
When Google released Gemini 3, its latest AI upgrade, last week, it broke records across the board.
But benchmarks are too narrow to be totally reliable guides to ability and potential.
This, says Marc Warner, is why you need to zoom out and look at the overall trend. When you do, he says, you see “a very strong exponential”.
An exponential trend is where growth doubles and keeps on doubling. At first progress seems slow, but, before long, the line on the chart is rising almost vertically.
“Nothing, nothing, nothing, everything,” as Dr Warner puts it.
It’s a pattern familiar from the COVID pandemic, where it caught out politicians and public health officials around the world.
Dr Warner says AI will be a more permanent reshaping of how everything operates
Now, says Dr Warner, who runs British tech company Faculty, it’s happening with AI – and he’s worried we don’t have a plan.
“We saw in COVID, if you don’t prepare for exponentials properly, they can really hurt you when things start to get very serious,” he says.
Could AI be as disruptive as COVID? It depends if its growth keeps going, Dr Warner says, and if AI is good at as many things as it appears to be.
“But if those were true, this would be way bigger than COVID,” he says.
“COVID was a temporary shift…This will be a more permanent reshaping of how everything operates.”
When will AI top out?
Dr Warner, who trained as a physicist before moving from academia into tech, has been here before.
In March 2020, Faculty was modelling patient data for the NHS. He saw the virus doubling at a faster rate than the government appeared to appreciate.
He got in touch with his brother Ben, who was working as a data scientist in Downing Street.
Dr Warner alerted the government to the exponential growth trajectory of COVID. Pic: PA
Late on the evening of 12 March, Ben and Marc explained the situation to the prime minister’s chief of staff, Dominic Cummings. Ten days later the country was in lockdown.
“Marc Warner is one of the smartest and most ethical people I have ever met in my life,” Mr Cummings later told MPs. “I think that without him thousands of people would be dead.”
Eventually, Dr Warner says, AI will stop getting better, simply because the tech companies run out of energy to train their power-hungry models.
“The amount of energy you’d need to train these models would be more than exists on the whole planet,” he says.
“So this has to top out at some point.”
‘Doubling every seven months’
However, the data he’s looking at – a comparison of humans and AI models on software development tasks by AI research firm Metr – suggests AI is doubling capacity every seven months and that it’s likely “there’s at least another five years of this”.
So, by his estimate, there are a lot of changes to come.
Pic: Metr.org
Some will be hugely positive. But inevitably there will be downsides.
“As with any new technology, there will come a bunch of disruptions… That’s why I feel like it’s so important to actually think seriously about if this holds true, what it would mean,” says Dr Warner.
“We will be able to manage it, but we’ll only be able to do that if we actually have a real plan.”
Does he think the government has one?
“In the short term, I think the government’s actually doing a decent job,” he says. “They announced this sovereign AI fund, I think that’s a good thing.”
He also praises the UK’s AI Security Institute, which investigates technical risks.
Read more from Sky News:
AI satirist on Sky correspondent’s rap battle with PM
There’s a new Bobbi on the beat – and they’re powered by AI
👉 Tap here to follow the Sky News Daily wherever you get your podcasts 👈
Just as in the pandemic, huge uncertainty surrounds even the best forecasts of AI.
Metr’s data only measures software development, and only assesses if AI has a 50% chance of succeeding at a task, so it’s hard to generalise from.
Then there’s the economic uncertainty. There may be a speculative bubble around AI, Dr Warner says, but that doesn’t make the underlying technology any less impactful.
“It feels to me something like we’ve gone from the first [aeroplane] flight to something like Concorde in a seven-year period. And that is a very big deal.”
Especially when there might be a lot more to come.
