The 73% Question: Are We Already Living With AGI?
Q: What happens when AI is mistaken for a human… most of the time?
Cornell University recently ran a randomized, controlled, pre-registered Turing test with four systems:
ELIZA (the OG chatbot from the 1960s)
GPT-4o
LLaMa-3.1-405B
GPT-4.5
Participants had five-minute conversations with a real human and one of these AI models — then had to guess which one was the human.
The results?
GPT-4.5: Judged human 73% of the time (more often than the actual humans)
LLaMa-3.1: Judged human 56% of the time (about even with real humans)
GPT-4o: Judged human 21% of the time
ELIZA: Judged human 23% of the time
That 73% score makes GPT-4.5 the first AI system ever to empirically pass a standard three-party Turing test.
Q: Is 73%… good enough?
Think about it: If an AI can convince strangers it’s human nearly three-quarters of the time, what does that mean for simpler, everyday tasks?
Handling customer service chats?
Drafting proposals?
Coordinating schedules?
Simple decision-making as an agent?
If it’s already outperforming humans in a human-imitation test, you can bet it’s already capable of executing many “human-level” business functions — with more reliability and no lunch breaks.
Q: But wait… humans fail all the time, too
One of the most interesting takeaways from the Cornell study is this: our benchmarks for AI often assume perfection, while we fully accept human inconsistency.
Humans:
Call in sick
Get stuck in traffic
Forget details
Misinterpret instructions
Take coffee breaks
Let emotions influence decisions
AI:
Never takes PTO
Never loses focus
Operates 24/7
Can be retrained instantly
So if AI can act human 73% of the time, isn’t that already close to the functional error rate of human workers — just without the unpredictability of human life?
Q: Does this mean we’ve reached AGI?
Here’s the million-dollar debate:
AGI (Artificial General Intelligence) is generally defined as an AI that can perform any intellectual task a human can, with comparable skill and adaptability.
Passing a Turing test doesn’t technically prove AGI — but it’s a huge leap toward it.
The 73% stat suggests AI isn’t just parroting patterns — it’s holding its own in human conversation.
Some researchers think AGI could arrive within the next couple years. Others say it’s still far off. But the truth is, the definition of AGI itself is subjective — and numbers like these are blurring the line faster than most expected.
Q: Just because we can… should we?
This is where the ethics and customer experience questions hit.
Even if AI can:
Sell your product
Support your customers
Train your team
Lead your meetings
…do your customers want that?
Will they accept it if they know? Or will they value the “human touch” more, even if it’s less efficient?
Some businesses will go all-in on AI-first. Others will keep humans in the loop as a trust factor.
The winning approach may not be AI vs. Human — but AI + Human:
AI as the scalable engine
Humans as the connection point
Because at the end of the day, trust is built on relationships — and relationships are still a deeply human thing.
💡 Final Thought:
73% is not just a number. It’s a flashing sign that says, the future is here. Whether we call it AGI or just “really smart AI” is semantics. The real question is: What are we going to do with it — and how much of our human work are we ready to hand over?
💡 Want to actually start using AI in a way that sticks?
Join the Maadii Membership and get monthly prompt libraries, plug-and-play frameworks, and real-world AI training that actually helps your team work better — not just faster.
Check out the Maadii plan to transform with AI. You can also connect with us.
You can also connect with us and learn more about Maadii’s services
