The Turing Test is so Last Century: Introducing the Barista Test for Artificial General Intelligence
Most people are familiar with the Turing Test, but if you are not, it goes something like this. You have a conversation with two participants using a computer keyboard and screen, and one of the two participants is a machine. If you cannot tell which participant is the machine, then it has passed the Turing Test. This test was named after Alan Turing, who was the motivation for the Oscar-winning film “The Imitation Game.”
In this age of ChatGPT, the Turing test seems quaint. Nobody would be fooled into thinking a computer is a person, regardless of how clever it might be, because it is after all just a computer. So, what’s next? The Coffee Test has been proposed, and is attributed to Steve Wozniak, one of the founders of Apple Computers. According to Steve, this test would require a robot, not a computer screen. The robot would need to locate the kitchen and brew a pot of coffee in a random house that it had never seen before. While the Turing Test is considered a test for artificial intelligence or AI, the Coffee Test is considered a test for artificial general intelligence, or AGI, which is sometimes defined as the ability of a machine to perform any task that a human can perform.
While the Coffee Test is a good start, it does not go far enough. First off, who in their right mind would let a robot into their house after seeing The Terminator? Perhaps if it was made from mimetic polyalloy and had a face like a movie star, but in all reality, robots are going to look pretty much like robots for the foreseeable future. So, the first qualification to the Coffee Test is that the robot has to talk its way into the house. I suspect the conversation would initially go something like this:
Ding-dong (that is the doorbell, not the robot)
Homeowner: “What do you want?”
Robot: “I am a robot, and I would like to make a cup of coffee.”
Slam (that is the door, not the robot).
So, the robot is going to have tough time, unless it gets lucky and knocks on the door of a science fiction fanatic or scrap metal recycler. However, after a few tries the robot might get its CPU in the game, and the conversation would go something like this:
Ding-dong (doorbell again)
Homeowner: “What do you want?”
Robot: “I am a robot, and I would like to buy a cup of coffee for $100.”
Homeowner: “Make it $150 and you’ve got a deal.”
Robot: “Do you take bitcoin?”
But this is still pretty quaint — everyone would just assume that the robot was remotely controlled by a human, and that Borat was behind it all. A better test is the Barista Test. The robot has to be completely autonomous, that is, not connected to the Internet or any other network. The robot also has to convince the homeowner to let it come in and make a cup of coffee. After all, human interaction is a task that humans do all the time, and AGI isn’t really AGI if it can’t perform that task. But, you might say, even I couldn’t talk my way into someone’s house and convince them to let me make them a cup of coffee, but to that I say, have you ever tried?
Finally, the robot can’t just make any cup of coffee, it has to make whatever type of coffee the resident of the house wants. Yes, even a double soy latte with cinnamon sprinkles, assuming that they have soy, cinnamon and a working espresso machine with one of those foam attachments that is not all clogged up and gross like mine is after the kids use it. While the homeowner might still think that the robot is remotely controlled, that would not change the outcome of the test. After all, that is the objective of AGI, to be indistinguishable from a human. So, extra credit for getting the homeowner to think it’s an elaborate hoax with a remote human operator. And extra extra credit for cleaning the foam attachment.
The ability to create a robot that can function autonomously and pass the Barista Test will probably depend on a “neuromorphic” computer chip — an integrated circuit that can replicate the function of action selection by a human brain. A regular digital computer lacks the processing power to handle all of the variables that would be associated with navigating a complex environment like the inside of a strange house and responding to whatever questions or complications that the people in the house might present. For example, in order to create a ChatGPT user interface, thousands of servers are needed, which is too big for a robot that can walk through the front door of a house. That does not even account for the fact that ChatGPT often lies and hallucinates, and it would not do to have the robot hallucinate that the car is an espresso machine and to lie that the engine oil is butter coffee. In contrast, a neuromorphic computer chip that mimics the ability of the human brain to make decisions could potentially fit in a robot, if it existed, and it might not even lie or hallucinate. There are many neuromorphic chips out there, not in the grocery aisle next to the Ruffles, but in university labs where scientists do things like try to make computers act like part of a human brain, such as the one bubbling in the vat next to the Jacob’s Ladder. However, as of today, there are no neuromorphic chips for acting like the part of the human brain that selects which grind of coffee to make. As the inventor of such a neuromorphic chip, I believe it is feasible, but you will have to wait for my next article to hear more about it.
Without a neuromorphic chip, any robot would need to be connected to a network and would stop working properly if the network connection crashed. While that might be OK for some things, and might even pass the Coffee Test, the Barista Test is a better example of what it would take for a robot to be commercially successful. Would you hire a barista that stopped working in the middle of making your double soy latte with cinnamon sprinkles? I didn’t think so.
