Five real stories about what AI is already capable of doing - critical summary review

There is an old problem in biology that went unsolved for fifty years. Proteins are molecules that run almost everything in the human body: digestion, immunity, the response to viruses. To understand how they work, scientists need to know the exact three-dimensional shape they fold into... and figuring that out used to take months, sometimes years, for each individual protein. In twenty twenty, a program called AlphaFold, built by Google's DeepMind lab, solved that problem in a matter of minutes.

The achievement was extraordinary enough that, in October of twenty twenty-four, two of its creators, Demis Hassabis and John Jumper, were awarded the Nobel Prize in Chemistry. The system has now mapped more than two hundred million protein structures and is freely accessible to over three million researchers in a hundred and ninety countries. Thirty percent of the research done with that data is focused on human disease. It's as if a tool had unlocked a library that had been sealed shut for half a century.

That's the first thing worth understanding

about what artificial intelligence is already doing in the world: it's solving scientific problems that the human brain, working alone, couldn't crack at scale. Not as a replacement for scientists, but as infrastructure. Like a microscope that nobody had thought to build yet.

Medicine is moving at the same pace. In twenty twenty-four, researchers at University College London developed a blood test capable of predicting Parkinson's disease up to seven years before the first symptoms appear. It uses artificial intelligence to detect patterns in biomarkers that the human eye simply can't distinguish. Other studies have extended that same approach to early detection of pancreatic cancer... one of the deadliest forms of the disease precisely because it typically shows up too late. The work is still advancing, but the direction is clear: AI is giving doctors a longer runway.

The second story is closer to everyday life and considerably more complicated.

Artificial intelligence already writes, composes music, generates video, and illustrates stories. Tools like Sora, Veo, and Runway have made it possible to produce synchronized audio-visual content without a full production team. Illustrators use programs like Midjourney to generate rough drafts as a starting point. Musicians blend algorithm-generated compositions with real recordings. This is happening every day.

But there's a side of this that's hard to ignore. Most of these systems were trained on work created by human artists, typically without permission and without compensation. In twenty twenty-four, more than ten thousand creators — including novelist Kazuo Ishiguro, actress Julianne Moore, and Radiohead's Thom Yorke — signed an open letter condemning the unauthorized use of creative works to train AI models. The legal debate is still wide open. What is clear is that human creativity and automated creativity are already competing in the same market, and the rules of that game are still being written.

The third story is about money... and about what happens when a technology crosses from one era into another.

Jensen Huang founded Nvidia in nineteen ninety-three to make graphics chips for video games. That was the business for decades. When artificial intelligence started demanding massive computational power, Nvidia's chips turned out to be exactly what research labs needed. Huang saw that before most people did. In July of twenty twenty-five, Nvidia surpassed four trillion dollars in market value — the first company in history to reach that number. Huang's personal fortune reached one hundred and fifty-four billion dollars.

Alexandr Wang has a different kind of story. He founded Scale AI at nineteen, in twenty sixteen, after dropping out of MIT. The company does work that looks invisible from the outside: it organizes and labels data so that AI models can learn from it. Think of it as teaching a machine to tell a car apart from a bicycle... but across billions of examples. At twenty-eight years old, Wang has a net worth estimated at three point two billion dollars and was named Chief AI Officer at Meta after Zuckerberg invested fourteen point three billion dollars in Scale AI.

Pieter Levels is the most accessible of the three. A Dutch developer who, since twenty twelve, has been testing ideas online and cutting what doesn't work. Starting in twenty twenty-two, he began using generative AI to build digital products in much shorter development cycles. Today he earns around three point five million dollars a year, operating entirely alone, with profit margins above ninety percent. His cost structure comes down to servers and API subscriptions.

These three cases are not a blueprint. They're a portrait of how the same wave hits people at very different positions: a decades-long executive, a twenty-eight-year-old startup founder, and an independent developer working from a laptop.

The fourth story is about productivity...

and it's not quite what most people picture. According to the AI Jobs Barometer published by PwC in twenty twenty-four, sectors that adopted artificial intelligence saw productivity growth four point eight times higher than their previous rate. The point isn't working more hours. It's spending less time on repetitive processing and more time on the things that require human judgment: strategy, relationships, decisions that depend on context.

The difficult side of that equation is real. Jobs built around pattern recognition are among the most vulnerable to automation: document review, data entry, standardized customer service. Accenture's Future Skills report estimates that ninety-seven million new roles could emerge by twenty twenty-five at the intersection of people and technology. But the transition isn't automatic. It requires training, adaptation, and time... resources that aren't equally distributed.

The fifth story is about the planet's infrastructure.

DeepMind's GNoME system identified two point two million new chemical structures in twenty twenty-three, many of them with potential applications in more efficient batteries, cheaper solar panels, and low-power electronics. Pilot projects in China and Germany are already testing some of these materials at scale. At the same time, Google's GraphCast model has outperformed traditional weather forecasting methods across several key metrics. That matters for energy management, logistics, and the response to climate-related disasters.

The potential is real. The question that remains is whether the pace of development will outrun the ability to regulate it and distribute its benefits broadly... or whether the gains will concentrate in the hands of a few before anyone figures out the rules.

What to do with this information

Scenario one: you work in any field that runs on data, documents, or communication. Right now, the ask isn't to become a programmer. It's to understand which parts of your work are most exposed to automation and which are hardest to replace. Skills like synthesis, judgment, and relationship-building gain value when raw processing becomes cheap.

Scenario two: you have a product or service idea but assumed you'd need a large team to make it real. Pieter Levels is an example that shorter testing cycles and cheaper infrastructure already exist. That's not a guarantee of success. It's a real change in the cost of entry.

Scenario three: you follow financial markets or are thinking about where to put capital. Nvidia is the most studied case, but the run-up in AI infrastructure stocks came with significant volatility. In January of twenty twenty-five, the company lost nearly six hundred billion dollars in market value in a single day. The promise is real. So is the risk.

Scenario four: you're a creator of any kind — artist, musician, writer, designer. The legal debate around copyright and the use of creative work to train AI models has no definitive answer in any country yet. Keeping up with legislation and understanding the terms of service of the tools you use is, at this moment, as important as learning to use them.

The world isn't waiting for consensus. It's moving. And the most useful question might not be "what will artificial intelligence do?" but "what do I do with what it's already doing?"