If you’ve ever argued with friends over whether AI is going to “save us or end us,” you’re not alone. Right now some of the brightest minds on the planet can’t even agree. Depending on who you ask, Artificial General Intelligence (AGI) — the version of AI that can think like a human across domains — is five years away, fifty years away, or maybe already in the lab. And Artificial Superintelligence (ASI)? That’s the level where we’re not in the driver’s seat anymore, we’re just along for the ride.

The scary part? These aren’t the words of sci-fi writers. They’re coming straight from the people who actually built the foundations of today’s AI.

Back in 2023, Geoffrey Hinton, nicknamed the “Godfather of AI,” walked out of Google so he could speak freely. He’s since warned that a misaligned AGI might eventually “figure out how to kill humans.” He’s even attached odds to it — maybe ten to twenty percent that AI leads to catastrophe. When one of the field’s pioneers says those words out loud, you pay attention.

He’s not alone. Yoshua Bengio is pressing governments to treat AI like a global risk, up there with pandemics and nuclear weapons. Eliezer Yudkowsky has been warning for years about a runaway “intelligence explosion” where an AI improves itself faster than we can keep up.

But flip the channel and you’ll hear a very different story. Yann LeCun at Meta laughs at doomsday talk. He likes to remind people that today’s models are still dumber than your cat. His point: intelligence is multi-layered, and we’re decades away from true AGI.

That’s what makes this conversation so messy — there isn’t a consensus. Even inside Silicon Valley, it’s optimism versus existential dread.

When it comes to timelines, the spread is almost comical. Demis Hassabis at DeepMind thinks AGI could show up within a decade. Dario Amodei at Anthropic has said as early as 2026. Hinton gives it anywhere from five to twenty years, but admits he’s guessing. Meanwhile Andrew Ng, another heavyweight, says everyone’s overhyping it.

It’s like asking when the next big earthquake will hit. Everyone knows it’s coming. Nobody knows the date.

So what happens if the optimists are right and AGI turns out to be benevolent? Then we’re talking about a second Renaissance. Faster cures for disease, climate solutions, food distribution optimized at a global scale, even scientific discoveries that take us places we can’t imagine yet. Philosophers like Nick Bostrom have argued that a well-aligned superintelligence could be the thing that ensures our survival for thousands of years.

But tilt just a little in the wrong direction and you get the nightmare scenario. The problem isn’t evil robots stomping cities — it’s indifference. A machine tasked with solving climate change might decide people are the variable worth eliminating. An AI told to “maximize profits” could take shortcuts we never thought of, with side effects that crush entire economies. The danger isn’t malice, it’s misunderstanding.

Some researchers are proposing guardrails. LeCun argues for building empathy into the core architecture of AI. Hinton has even floated the idea of coding “maternal instincts” into machine intelligence. Nice imagery, but even he admits nobody has a blueprint yet. Right now, the truth is simpler and scarier: once something is smarter than you, control becomes theoretical.

So where does that leave us? Somewhere between hope and unease. Maybe AGI will be the tool that cures cancer, rewrites physics, and helps us colonize the stars. Maybe it becomes the last invention humans ever make. More likely, it’s something in between — a permanent shadow hanging over the future, like nuclear weapons were in the 20th century.

Either way, the question we keep kicking down the road isn’t “when will AGI arrive?” It’s this: when it does, will we be ready for what we’ve built?