Dec 162022
 

Remembering how ChatGPT failed to solve the riddle involving squirrels, a little later I asked him the following:

Bob was twice as old as his sister Alice when he was six. Today, Bob is 30 years old. How old is Alice?

When Bob was six years old, his sister Alice was half his age, so she was 6 / 2 = <<6/2=3>>3 years old.
Since then, Bob has aged by 30 – 6 = <<30-6=24>>24 years, while Alice has aged by 24 / 2 = <<24/2=12>>12 years.
Therefore, Alice’s current age is 3 + 12 = <<3+12=15>>15 years old. Answer: \boxed{15}.

Intriguing. For starters, note the weird typography: ChatGPT was trying to use math, enclosing some equations in double angle brackets (using less-than and greater-than signs as substitutes) but used LaTeX notation for its final result. Which, of course, is wrong: ChatGPT fell for the same trap that tricks some humans.

But why, I wondered? Could I present a simpler form of this riddle that ChatGPT could solve correctly? Why, let us try:

Bob was three years older than his sister Alice when he was six. Today, Bob is 30 years old. How old is Alice?

When Bob was six years old, his sister Alice was 6 – 3 = <<6-3=3>>3 years old.
Since then, Bob has aged by 30 – 6 = <<30-6=24>>24 years, while Alice has aged by 24 – 3 = <<24-3=21>>21 years.
Therefore, Alice’s current age is 3 + 21 = <<3+21=24>>24 years old. Answer: \boxed{24}.

Once again an incorrect answer but I think I can spot the pattern. Whatever the mathematical operation is, ChatGPT applies it to both the quantities in question and their difference (or ratio).

I think this is a very clear example of how a trained AI automaton that has no independent ability to reason can go astray. Incidentally, this is exactly how humans who learn by rote but without full comprehension often fail.

Before we read too much into these failures, however (which, I presume, are easily corrected with more guided training) let us notice the fact that in both cases, ChatGPT correctly understood the question and applied the concept of time in its answers. Which is quite remarkable on its own right, mistakes notwithstanding.

 Posted by at 5:55 pm