The Good News Episode

The average GP detects 8 cancer cases a year. The average cancer patient sees a doctor 3 to 5 times before being recognised as at-risk. Two-thirds of all cancer deaths come from cancers that aren’t part of any screening programme — there’s no routine test that would find them.

“C the Signs” is an AI that reads the whole record at once: previous test results, prescriptions, age, family history, even postcode. In one published case, a GP saw a patient with diarrhoea and abdominal pain. Standard guidelines wouldn’t have recommended a cancer referral. The AI flagged a faecal test. The result was an early-stage colorectal cancer diagnosis — and a patient who lived because of it.

Across practices using the tool, cancer detection rates rose from 58.7% to 66%.

In the largest NHS study of its kind — 175,000 women — Google’s AI served as a second pair of artificial eyes alongside a human radiologist.

It found more cancers. It found more invasive cancers. It caught 25% of “interval cancers” — the ones that develop between routine scans that came back clean. Detection rates rose from 7.54 to 9.33 per 1,000 women.

But the headline number is the time. A radiologist takes, on average, 2.08 days to clear a scan. The AI does it in 17.7 minutes. In a country with a 29% shortfall of clinical radiologists, that’s not replacement — it’s the difference between a system that runs and a system that collapses.

There’s a condition called focal cortical dysplasia — tiny malformations in a child’s brain that cause severe, medication-resistant seizures. They’re treatable with surgery, but only if you can find them.

Modern MRI and PET scans produce so much high-resolution data that human radiologists miss these lesions in more than 60% of traditional cases. An AI tool now spots them with up to 94% accuracy.

In a recent trial, 12 of 17 children went on to targeted surgery. 11 of them came out seizure-free.

The UK has 1.2 million people on mental health waitlists. The ieso digital programme is a Cognitive Behavioural Therapy service delivered by an AI conversational agent — trained on more than 750,000 hours of typed therapy — alongside human support.

In a landmark study, 82% of users experienced a clinically meaningful reduction in anxiety symptoms. Over half saw improvements within two weeks. The outcomes matched traditional human-led therapy. The cost in clinician hours was up to eight times lower.

This isn’t AI replacing therapists. It’s one therapist suddenly being able to help eight times more people.

Sutton Council in London installed sensors on the everyday objects in vulnerable seniors’ homes — the fridge door, the kettle, the bedroom lamp. The AI learns the rhythm of a household.

When that rhythm breaks — a meal skipped, a morning cup of tea unmade — it sends a quiet alert to a family member or carer. Often, before anything has actually gone wrong.

The technology is invisible. The dignity is the point.

In Australia, a 79-year-old woman named Peta Rolls is trialling a voice assistant called Aida. Aida doesn’t sound like a robot. It asks her what she’s doing today. If she says “going shopping,” it responds in context. If she mentions she’s not feeling well, it asks follow-up questions.

If she says her chest hurts, it calls an ambulance.

The promise here isn’t a robot companion. It’s a quiet safety net for the millions of elderly people who live alone, and whose worst day might otherwise go unnoticed for hours.

AI-optimised wind farms produce 3 to 5% more energy from the same turbines. Across the global fleet, that’s enough to power roughly one million additional homes without breaking ground on a single new structure.

Meanwhile, the UK’s national energy system is saving £30 million a year and cutting 300,000 tonnes of CO₂ emissions — purely from better AI-powered solar forecasting. Knowing exactly how much sun will hit the panels tomorrow means the grid stops paying gas plants to stand by “just in case.”

When new infrastructure is needed, AI can evaluate millions of possible routes for overhead power lines — cutting planning study time by 93%.

Smart grids act as the brain of a city. They predict the weather. They learn the rhythm of when people cook and shower and turn the lights on. They route excess afternoon solar into electric vehicles charging overnight. They tune the building HVAC to who’s actually in the building.

In some districts, AI-managed networks have reduced overall energy consumption by up to 30%. No new generation. No new wiring. Just better choices, made faster than humans can make them.

In Bundaberg, Australia, a bioenergy project uses AI to analyse the composition of agricultural waste in real time — and squeezes 35% more energy out of the same input.

Getting British homes to net-zero requires replacing tens of millions of gas boilers with heat pumps. The cost has been the wall everyone hits.

Recent AI-designed installations have cut the cost to the consumer by approximately 75% and halved installation times. The Manchester Prize is funding AI drones that thermal-scan entire neighbourhoods overnight — pinpointing the worst-insulated homes for targeted upgrades. Another project is using AI to design smart external panels that turn an old building’s brickwork into a radiator.

None of this looks like “AI.” It looks like a cheaper boiler.

By 2050, large-scale solar is projected to supply roughly a quarter of Australia’s electricity. Keeping the panels clean and working is a massive logistical problem — dust, bird droppings, microscopic electrical hotspots — that has traditionally required weeks of human inspection across kilometres of remote terrain.

The new approach is autonomous robots. LiDAR. Thermal cameras. Sensors that build 3D maps. They drive themselves across the farm and report which panels are dirty, which wiring is damaged, which inverters are about to fail.

Humans still do the fixing. They just don’t have to find the broken thing anymore.

Stretch 4, from a company called Hello Robot, costs $29,950. It is a telescoping arm on an omnidirectional wheeled base. No face. No legs. No hands that look like hands.

Its CEO, Aaron Edsinger, argues that the viral videos of humanoid robots doing parkour are missing the actual humans the robots are supposed to be serving. The point of a robot in a home is not to look human. It’s to help a human.

Our homes are already adapted for wheels. Ramps. Smooth floors. Doorways. Legs are an expensive, dangerous solution to a problem we’d already solved.

This is the unacknowledged risk of bipedal humanoid robots in eldercare. If one glitches, or runs out of battery, or hits a software error mid-step — it collapses. Onto whatever is beneath it. A patient. A pet. A child.

Stretch 4 is built so this can’t happen. It’s low. It’s light. Its centre of gravity sits on a wheeled base. Hit the emergency stop and it simply freezes in place.

The safest robot in your house is not the one that looks like you. It’s the one that can’t fall on you.

Vision-only home robots are functionally blind to the chaos of an actual home — a wandering elderly dog, a sprinting toddler, a power cord on the floor.

Stretch 4 uses the kind of sensor stack you’d find on a self-driving car. Two hemispherical 3D LiDARs. Three high-resolution cameras. A ring of six laser-line sensors specifically to detect small floor hazards — drop-offs, cables, rugs.

They deliberately over-built the sensing budget. Because the alternative is a robot that hurts the person it was meant to help.

Henry Evans had a stroke in his forties. Quadriplegia. Non-verbal. He now sits on the board of Hello Robot.

He uses Stretch 4 as an extension of his own body. A computer interface, a precise arm, a stable base. With it, he fetches his own water. He closes his own blinds. He manipulates his own environment for the first time in two decades.

In his own words, paraphrased: a paralysed person doesn’t need a robot with legs and dozens of joints mimicking human movement. They need an affordable, stable machine that grants them confidence, independence, and a life with more possibility.

That’s the whole episode in one sentence.