Reverse Turing Test

Reverse Turing Test

title says all

·

2 min read

hupili_one\_robot\_doing\_a\_lab\_test\_to\_distinguish\_another\_bot\_fr\_05a7bc31-88ab-449c-bcab-d3da24bc4968.png

There is one thing I'm always fascincated about but did not detail in the post On Future of AI: Reverse Turing Test.

Recently, Large Language Models has shown that machine is capable of imitating human in many aspects, for example computer programming.

The imitation game is on.

The concept of Turing Test is widely known: if a human can not distinguish machine from human via Chat interface, the machine is said to pass Turing Test.

Think of one day when machine takes over the world. They will build their own system and perimter to guard the access. If there is a region that only allows machines to enter, how can machines tell if the user is a machine or human?

They may apply a "Reverse Turing Test".

The first feature to distinguish human from machine is information exchange rate. Notice the key here is Chat interface. When machines imitate human, they can easily slow down and output word by word like a human thinking + speaking at the same time. However, human naturally speaks much much slower... If RTT is designed by machine, they wouldn't purposefully slow down the conversation, unless for amusement.

One possible way to bring human on par to the game is the ECG technology. Once we can use our brain to interact with machine directly, we are not limited by the format and rate of "language". That may be the "LLM-moment" for machine. They suddenly realise there is part of human brain that works perfectly logical/ structured/ fast.

Once the language barrier is lifted, another distinguishing feature left is the speed to do simple calculations, like multiplying two 20 digits number. Machine does well at those tasks but human brain is weak at them. However, once human becomes cyber enhanced, the gap is closed. Think of one day that everyone has an ALU and huge external memory attached to brain. This is very like today when ChatGPT can call plugins to enhance its domain specific ability.

So the key question becomes, if we allow both machine and human to access an array of facilities that do perfectly in domain specific tasks, what is the key distinguishing factor between the two?

It stays the open question. We can not wait to see RTT in action and the first human to pass RTT.

Did you find this article valuable?

Support HU, Pili by becoming a sponsor. Any amount is appreciated!