TL;DR: AI has already done things that would seem like science fiction if they hadn’t actually happened. From inventing its own language to discovering new planets and identifying diseases from a voice recording, these ten facts reveal how far AI has come and how strange the reality is compared to the headlines.
Most of what gets written about AI is either breathlessly optimistic or catastrophically worried. What gets lost in both versions is something simpler: the real story is already extraordinary.
Here are ten things AI has actually done: verified, sourced and genuinely astonishing.
1. AI Beat the World’s Best Board Game Player at a Game With More Moves Than Atoms in the Universe
In March 2016, Google DeepMind’s AlphaGo defeated Lee Sedol, an 18-time world champion at the ancient board game Go, four games to one. This wasn’t supposed to happen for another decade.
Go has more possible board configurations than there are atoms in the observable universe. Chess, by comparison, is a parlour game. The complexity of Go had convinced most experts that human intuition could never be replicated by a machine. AlphaGo didn’t just win, it played Move 37 in Game 2, a move so unexpected that experienced commentators thought it was a blunder. It was a work of strategic genius.
Lee Sedol later retired from professional Go, stating that AI was an entity that “cannot be defeated.”
2. The First AI Chatbot Was Built in 1966 and People Fell in Love With It
ELIZA was created at MIT by computer scientist Joseph Weizenbaum. It mimicked a therapist by reflecting statements back as questions. If you typed “I’m feeling sad,” ELIZA might respond: “Why do you say you are feeling sad?” That was essentially the whole trick.
ELIZA understood nothing. It had no model of the world, no memory, no comprehension. Yet people formed genuine emotional connections to it. They asked to speak with it privately. They confided things they wouldn’t tell real therapists. Weizenbaum’s own secretary asked him to leave the room so she could speak with ELIZA alone.
Weizenbaum was so disturbed by this reaction that he spent the rest of his career writing about the dangers of humans anthropomorphising machines. His 1976 book Computer Power and Human Reason remains one of the most important warnings ever written about AI and it was inspired by a chatbot that was little more than a find-and-replace algorithm.
3. An AI-Assisted Novel Passed the First Round of a Major Japanese Literary Prize
In 2016, a novel partly written by an AI passed the first-round screening of the Nikkei Hoshi Shinichi Literary Award, one of Japan’s most respected science fiction competitions. The judges who screened it had no idea it was partially AI-generated.
The novel didn’t win but it made the shortlist. The judges later noted that the prose was competent and the structure sound. The weaknesses, they said, were in character development which is arguably the most human part of storytelling.
4. AI Can Detect Parkinson’s Disease From a Voice Recording
Parkinson’s disease typically causes subtle changes in a person’s voice, micro-tremors, pitch variations, breathiness, that are often inaudible to the human ear and imperceptible even to trained clinicians in early stages.
Researchers have developed AI systems that can detect early signs of Parkinson’s from voice recordings alone, with accuracy exceeding 85% in some studies. The system can identify people who have the disease but haven’t yet developed noticeable symptoms, potentially years before a traditional diagnosis.
This matters because Parkinson’s treatments are far more effective when started early. An AI that can flag risk from a phone call or a voice note could change the course of the disease for millions of people.
5. GPT-4 Scored in the Top 10% of the US Bar Exam
The bar exam is one of the most demanding professional tests in the world. It requires mastery of legal reasoning, case analysis and the ability to apply complex rules to novel situations, exactly the kind of higher-order thinking that was thought to be beyond AI.
OpenAI’s GPT-4 scored in approximately the 90th percentile, comfortably within the range that would qualify a human as a licensed attorney in most US states. It also scored in the top percentiles on the LSAT, the GRE, multiple AP exams and the USMLE (the medical licensing exam used to qualify doctors).
To be clear, GPT-4 cannot practise law as it lacks the judgment, accountability and professional responsibility that the role requires. But the test results forced a serious reassessment of what “uniquely human” cognitive tasks actually are.
6. Google’s AI Translation System Invented Its Own Secret Language
In 2016, Google’s neural machine translation system which had been trained to translate between specific language pairs was given a test. Researchers asked it to translate between two languages it had never been directly trained on together.
It succeeded. But more surprisingly, when researchers examined the internal representations the system was using, they found it had developed what they described as an “interlingua”, an internal language of its own that no human had designed or programmed. The AI had invented a more efficient internal representation of meaning to do its job better.
Nobody told it to do this. It emerged spontaneously. The Google researchers were candid about being unsure exactly how it worked.
7. AI Found New Planets That Human Astronomers Had Missed
NASA’s Kepler space telescope gathered so much data about distant stars that the team of human astronomers assigned to it simply couldn’t analyse it all. There were too many star systems, too many light curves, too many subtle variations to check manually.
A team at Google trained a neural network on Kepler’s data and it identified two previously unknown exoplanets hiding in the noise. One of them, Kepler-90i, completed the discovery of the first known solar system with eight planets besides our own. Human analysts had looked at the same data and not spotted them.
8. Every Face on thispersondoesnotexist.com Is Completely Fake
Visit thispersondoesnotexist.com and you’ll see a hyper-realistic human face, a face with pores, hair strands, catch-lights in the eyes and all the imperfections of a real photograph. That person has never existed. The image was generated entirely by a type of AI called a Generative Adversarial Network (GAN).
Refresh the page and another face appears. Refresh again. Each one is unique, plausible and entirely artificial. The AI has generated millions of these faces. The technology has profound implications for identity verification, misinformation and our basic ability to trust what we see.
9. AI Discovered a Potential Antibiotic That Scientists Had Never Considered
In 2020, MIT researchers used AI to screen over 100 million chemical compounds in a matter of days, looking for molecules that could kill bacteria in ways fundamentally different from existing antibiotics. This matters enormously because antibiotic resistance is one of the most serious threats in global medicine: bacteria are evolving faster than we can create new drugs to kill them.
The AI identified a molecule called halicin, a compound that had previously been studied as a potential diabetes drug and then shelved. It turned out to be remarkably effective against drug-resistant bacteria, including some strains that had defeated every known antibiotic.
The same screening process, done by human researchers with traditional methods, would have taken centuries.
10. The Word “Robot” Was Invented in 1920 by a Playwright Who Imagined Machines Rebelling Against Humans
The word robot was coined by Czech playwright Karel Čapek in his 1920 play R.U.R. (Rossum’s Universal Robots). It comes from the Czech word robota, meaning drudgery or forced labour. In the play, artificial workers are created to serve humans and eventually rebel.
Čapek’s robots were biological, not mechanical, closer to what we’d now call clones than machines. But the concept he introduced, intelligent artificial beings turning against their creators, became the template for virtually every AI anxiety story that followed. Every Terminator, every malevolent computer, every cautionary tale about machines that no longer need us traces its creative DNA back to a play written more than a century ago.
The fear of AI, it turns out, is almost as old as the dream of it.
What These Facts Actually Tell Us
Taken together, these ten facts share a common thread: AI repeatedly does things earlier than expected, in ways nobody predicted, using approaches that humans didn’t consciously design. The AI that invented an interlingua wasn’t told to. The antibiotic-discovering AI wasn’t given halicin as a candidate. AlphaGo’s Move 37 surprised even its creators.
This pattern, emergence, surprise, capability ahead of schedule, is what makes the field genuinely hard to forecast. And it’s why understanding AI, rather than simply fearing or worshipping it, has never been more important.
Frequently Asked Questions
Are these AI achievements actually real or exaggerated? Every fact in this article is sourced and verifiable. The links throughout point to original research papers, official announcements, or peer-reviewed publications. AI achievements are often presented sensationally, these ones don’t need embellishment.
Does AI passing the bar exam mean it can replace lawyers? No. Passing a multiple-choice and essay exam tests knowledge and reasoning under controlled conditions. Practising law requires judgment, client relationships, ethical responsibility and accountability that no current AI system can provide. The exam result is impressive; it doesn’t change what AI is actually suited to do.
Is AI detecting disease better than doctors? In specific, narrowly defined tasks like identifying certain patterns in imaging data or detecting vocal biomarkers, some AI systems outperform individual clinicians. This doesn’t mean AI is a better doctor. It means AI is a useful tool that can augment clinical judgment, catch things that might otherwise be missed and process data at scale that no human team can match.
What’s the most important of these facts? The antibiotic discovery (Fact 9) is arguably the most consequential. Antibiotic resistance is a slow-moving global health crisis. An AI tool that can screen billions of compounds in days, finding candidates that would take human researchers centuries to identify could save millions of lives.
Want to understand the bigger picture? Read our complete guide: What Is Artificial Intelligence? →