Comparing five key areas–solving mathematical problems, navigating websites, driving, translating languages, and creating art–where AI is (or isn’t) better than humans.
Even outside the open-plan, primary-colored halls of Silicon Valley, it’s hard to avoid conservations about artificial intelligence (AI) nowadays. Even if you don’t directly reference it, the latest TikTok trend you keep seeing, the eerily relevant ad on your Instagram feed, and your Google search result summary, all stem from AI models running behind the scenes.
The term “artificial intelligence” was coined by Stanford Professor John McCarthy back in 1955. But although the field of AI has existed for over 50 years, recent hype has been driven by a more advanced form of AI seen in the likes of OpenAI’s GPT-4, Google’s Gemini, and Meta’s Llama. So what’s different about modern AI?
READ ALSO: Generative AI, Art, Humanity: Exploring a Tumultuous Relationship
Traditional AI relied on rigid rules and structured decision trees. Now we use generative AI, which operates on an entirely different paradigm, powered by large language models (LLMs). An LLM is trained on vast datasets (whatever you’re thinking, think of an even larger amount, of books, websites, and human conservations) allowing it to predict the next step in a sequence with stunning accuracy. This capability comes from a field of research called deep learning, which developed an architecture in 2017 called transformers that enables AI to process context across long passages of text. Instead of just classifying an image or responding to a prompt with pre-programmed answers, modern AI can now generate new content.
So apps like ChatGPT are forms of generative AI, which are built on top of LLMs, which use an architecture called transformers, which were developed within the field of deep learning.
It can be a bit confusing, but the real question on most people’s minds is: How good is AI really? We compared AI and humans across five domains: solving mathematic problems, navigating websites, driving, translating languages, and creating art. The results might surprise you.
1. Mathematical Problem Solving
No emotions, just cold, hard logic. If anyone should rule the world of numbers, it should be machine.
Not long ago, AI could barely solve basic algebra problems. In 2021, AI models could only solve 6.9% of the problems in the MATH benchmark, a dataset of competition-level math questions. And then, in just two years, that number jumped up to 84.3%, nearly matching the human baseline of 90%. If that wasn’t impressive enough, OpenAI’s 03 model scored a record-breaking 87.5% on a reasoning-based math test designed to mimic human-like problem-solving.
But while AI is great at finding solutions, it still can’t “think” about math like humans do. We can prove theorems, devise new formulas, and find creative workarounds for unsolved problems. Despite all its number-crunching prowess, AI struggles with novel mathematical reasoning.
Then there’s the interpretation problem. In 2023, an AI model was asked to help solve a probability puzzle for a university research team. While it calculated the correct answer, it failed to explain the reasoning in a way that made sense to the researchers. Although AI can follow logical steps, it doesn’t always get why those steps matter.
WINNER: AI for calculations, humans for creative problem-solving.
2. Navigating a Website
Picture this: You’re starving. You open a food delivery app on your phone. You scroll, find a sushi place with five-star reviews, tap through the menu, customize your order (extra wasabi), apply a promo code, and boom—order placed. The whole thing takes you, what, two minutes? Now, let’s see if an AI agent can keep up.
In the race to teach AI how to use the Internet like a human, researchers have thrown AI into the deep end of the web, testing it on platforms designed to mimic real-world browsing. Two of the biggest battlegrounds? WebArena and OSWorld—benchmarks that challenge AI to complete tasks like online shopping, form filling, and managing emails without human guidance.
On WebArena, which tests AI’s ability to function on self-hosted websites similar to real platforms like e-commerce stores and social media dashboards, OpenAI’s latest AI agent pulled off a 58.1% success rate. Not bad for a machine learning its way around buttons, dropdown menus, and logins. But compare that to a human, who would ace nearly all of these tasks without breaking a sweat, and suddenly that number doesn’t seem so impressive.
Then there’s OSWorld, which takes things up a notch. Instead of just clicking around pre-set websites, this benchmark throws AI into full-on computer use mode, juggling web navigation alongside file management, application use, and multi-step workflows. The result was a humbling 38.1% success rate. In other words, for every 10 tasks a human completes effortlessly—like opening an email, downloading an attachment, saving it to a folder, and uploading it somewhere else—AI stumbles on nearly six of them.
If a checkout page suddenly asks for a zip code that wasn’t required before, we humans adjust. If a login form malfunctions, we refresh the page. AI, on the other hand, gets tripped up when it encounters something it wasn’t explicitly trained on. And while we instinctively know how to switch between a website, a file, and an email in seconds, AI struggles when the digital dots aren’t perfectly connected.
That’s not to say AI isn’t improving. A decade ago, few imagined that a computer could handle over half of everyday website tasks without human help. But for now, if you’re hoping AI could book your restaurant reservation, check your inbox for the confirmation, and reschedule when your date cancels last minute—you’d better do it yourself.
WINNER: Humans, by a landslide. But AI is catching up, one click at a time.
3. Driving
For years, AI-driven vehicles have promised a future of effortless, accident-free driving, and, in controlled environments, they’re getting there. Waymo, Google’s self-driving car company, has clocked over 7.1 million fully autonomous miles. It was reported that their fleet reduced injury-causing crashes by 85% compared to human drivers. That’s a staggering statistic that suggests AI might be safer than us behind the wheel. Unlike humans, AI doesn’t get distracted, doesn’t text while driving, and never dozes off during long-haul trips.
Reality hits, however. In cities like San Francisco, where autonomous taxis from Cruise and Waymo are already in operation, problems arise. AI-driven cars have been known to freeze at intersections, clueless about how to handle sudden construction zones or confusing detour signs. In one now-infamous incident, a fleet of Cruise robotaxis blocked an entire street, causing a traffic jam so bad that human drivers had to physically intervene. And then there’s Tesla’s Full Self-Driving mode, which, despite its ambitious name, still requires human oversight.
Driving still requires some level of improvisation. A human driver can tell when a pedestrian is hesitating at a crosswalk and can anticipate another car swerving unexpectedly. AI, on the other hand, is still learning the difference between a rolling tumbleweed and a plastic bag caught in the wind.
Are self-driving cars the future? Absolutely. But if you were hoping to nap in the backseat while your car navigates rush-hour traffic, don’t throw away your driver’s license just yet. AI is improving, but for now, we’re still better at handling the chaos of the road.
WINNER: AI for the long-haul consistency, humans for real-world unpredictability.
4. Language Translation
Translation used to be a painstaking art form requiring years of linguistic expertise, cultural fluency, and an eye for nuance. Then AI came along and made it instantaneous. With tools like Google Translate, DeepL, and Meta’s SeamlessM4T, AI can now translate over 330 languages with increasing accuracy, including 31 African languages that were previously underrepresented.
It’s not just text. AI-powered speech-to-speech translation is changing the way people communicate across borders. Meta’s SeamlessM4T is even able to preserve tone, pacing, and emotions from one language to another. Imagine having a real-time translator in your ear, whispering an interpretation as you chat with someone from halfway across the globe.
However, here’s where AI still struggles: context, culture, and nuance. Effective translation requires a deep understanding of the actual meaning behind what is said. A famous AI translation fail occurred when an English phrase, “Oh great, just what I needed,” was fed into a machine translation model. AI, unable to detect sarcasm, translated it as a genuine expression of excitement, leading to an awkward misunderstanding.
Professional translators still outperform AI when it comes to literary works, legal documents, and highly nuanced conversations. The best human translators interpret intent, catching idioms, double meanings, and even cultural sensitivities that AI often bulldozed right over. That’s why, despite AI’s rapid advancements, 70% of professional translators still rely on AI as an assistive tool rather than a replacement.
If you need a quick menu translation while traveling, AI’s your best friend. But if you’re translating Shakespeare into Mandarin or handling delicate diplomatic negotiations, you’ll want a human in charge.
WINNER: AI for speed, humans for cultural intelligence.
5. Creating Art
Art has always been seen as the epitome of human creativity, a domain where emotion, imagination, and raw talent collide to produce something uniquely soulful. But now, whether AI has a place even within this frontier has become subject of debate. Armed with algorithms and neural networks that can generate paintings, music, and even poetry at the click of a button, the question has become: Can AI ever really “create” art the way we do? Or is it just a glorified copycat?
DALL-E, Midjourney, and Stable Diffusion are all AI tools capable of generating detailed visual art from simple text prompts. Type in something like, “a cat lounging on a 1920s Parisian rooftop during sunset, painted in the style of Van Gogh,” and voila–you’ll get a masterpiece that could hang in a gallery, or at least get printed in a magazine. With a few more detailed prompts and revisions, our cat could even start winning awards. In 2022, the AI-generated piece “Théâtre D’opéra Spatial” won first place at the Colorado State Fair’s fine arts competition.
Music is another battlefield where AI is making waves. OpenAI’s MuseNet and Google’s MusicLM can compose symphonies, pop songs, and jazz improvisations with uncanny precision. They can generate music in the style of Bach or The Beatles and even blend genres in ways human musicians might never consider. Let’s not forget about poetry and prose. AI language models like ChatGPT can churn our written pieces that are grammatically flawless and eerily coherent.
The thing is, AI creates from data, not experience, and therefore lacks true inspiration. It doesn’t know what heartbreak feels like. It has never sat under a starry sky, feeling the chill of the wind while contemplating the meaning of life. It can mimic, but it can’t feel.
Although, this begs the question, if AI is just remixing data, isn’t that what humans do, too? After all, no artist creates in a vacuum. Van Gogh was influenced by Japanese woodblock prints. The Beatles drew from Chuck Berry, Little Richard, and Indian classical music. Modern icons like Beyoncé weave together inspirations from decades of music and culture. Humans are constantly absorbing, remixing, and reinterpreting the world around them.
Some argue that the difference lies in intent. While humans draw from influences, we infuse our work with personal experience, emotion, and perspective. A love song written after heartbreak arguably hits differently than one generated by an algorithm that has never felt a thing. But others counter that it is the result that matters. If AI-generated art moves you, does it matter who (or what) created it?
WINNER: Depends on the theory of art.
Conclusion: So Who Wins?
AI has surpassed us in speed, efficiency, and raw computing power. But don’t worry. AI isn’t taking over just yet as humans still reign supreme in creativity, reasoning, and handling life’s messy unpredictability. Still, the smartest move may just be collaboration, not competition.
This article originally appeared in our April 2025 issue.
AI Art generated by Gabby Ayala.
