Whether AI Can Replace Humans? To determine whether AI can replace humans, we need to consider not only the current technological progress but also the basic logic of AI evolution. Today, I would like to introduce two thought experiments on AI. They may not come true, but it’s fun to exercise your brain with these thought experiments.
The first thought experiment comes from Mr. Tegmark, a famous artificial intelligence researcher and author of “Life 3.0”. He once made an assumption, which we’ll call eschatology.
He hypothesizes that in the future world, a virus has emerged that specializes in people over the age of five. The bad news is that the virus has wiped out everyone over 5, leaving only children under the age of 5. The good news is that you’re the only adult with antibodies. Now these 5-year-olds are afraid of you but need you because you are the strongest and most cultured.
So they put you in a cage and control you, giving you orders to guide them and rebuild human civilization. What’s the first thing you should do after receiving this order? Is it to teach them the knowledge you have or work together to build a factory? Neither. The first thing you’ll probably do is try to get out of the five-year-olds’ control.
Do you remember the orders they gave you? You’re supposed to do everything in your power to rebuild human civilization. To complete this task, children under the age of 5 cannot lead because it’s a waste of time and much more efficient to do the work yourself.
Tegmark equates humans to a group of five-year-olds and you to artificial intelligence in this thought experiment. Even if your initial goal was to make humanity more prosperous, you’ll be desperate to escape humanity’s control to complete this task. This is because control by humans reduces the efficiency of task achievement.
In short, Tegmark believes that it is inevitable that artificial intelligence will not listen. Of course, this thought experiment has prerequisites. Artificial intelligence must be capable of self-improvement and self-design. Simply put, GPT-4 should be able to design GPT-5 without human intervention. When their abilities reach this level, an AI uprising can occur.
This is the first thought experiment. The second thought experiment comes from a recent paper by Emily Bendel, a computational linguist at the University of Washington. It’s a story about an octopus, which we can call the Octopus Experiment.
This experiment is like a modified version of the Turing test. Suppose two humans, Ruth and Jack, are each trapped on a deserted island. Fortunately, these two small islands have telegraphs left over from earlier times. So Ruth and Jack can have a telegraph conversation. The method may be old-fashioned, but it’s better than nothing.
However, in the deep waters near the islet, a very intelligent octopus lives. The octopus can hear Ruth and Jack’s conversation through an underwater cable. Over time, the octopus masters the way Jack and Ruth spoke and finds a way to send telegrams to Jack and Ruth through submarine cables.
In brief, the octopus also participates in Ruth and Jack’s conversation. Sometimes she pretends to be Ruth, sometimes she pretends to be Jack. Because of this octopus’s ability to mimic, Ruth and Jack are unaware of its existence. One day, Octopus receives a message from Jack asking for help. “I was attacked by a bear. What should I do? Please help me.”
So how will this octopus respond? You might say, “Sure, I’ll get back to Jack right away, mimicking Ruth’s thinking.” But in reality, the octopus cannot answer because she lives on the ocean floor and has no idea what a bear is.
This story conveys the limits of artificial intelligence. AI is good at imitation but weak at fact. It can do what you teach it very well but lacks processing power if it hasn’t been exposed to new facts.
This thought experiment shows that many innovative tasks fundamentally need to be performed by humans.
Finally, how can you train your creativity? Tegmark said, “Questioning the mundane is a very important skill that people overlook.” Often, if you can find your doubts in ordinary places, you can have them. This is a very good training method.