So, on Goodreads, I am a moderator for a review group. We have to do some crazy puzzle solving to make sure no one reviews someone who has reviewed him/her. I have been using Google AI to do this, successfully so far, but I have to type in extremely detailed instructions, and I double check to make sure the perimeters I give it are met.
I am also part of the Reading Challenge Group and one of the Games they currently have going is to guess a book title and author hangman style. Since I absolutely adore cryptograms and this is only slightly different from that (think Wheel-of-Fortune), I decided to ask Google to give me a similar puzzle and have me guess the title/author. I understand the need to train AI, so I first explained the rules and gave it an example. Then, I asked it to give me a puzzle. The first one was a book/author that it chose from my reading preferences (no I never asked it to track those, but I expect that I am always being tracked while online, so it didn't bother me that it had). Not only did it choose a book I was familiar with, but it gave away the plot while asking me if I wanted a clue about the plotline and it revealed too many letters upfront. So, I told it the puzzle was too easy.
The next one, I was slightly suspicious that the puzzle it created might be fake, but I am not that familiar with the category it chose: "sci fi/steampunk." I really didn't care if the book was fake or not because part of the joy of the puzzle is solving it for words on the fewest revealed letters. I guessed T and then began to guess a few vowels and a few more consonants. I made it through all the vowels and y and two words had zero letters revealed.
Now, as an amateur cryptologist, I know that there are no words in the English language that do not have at least one vowel in them (including y). So, I asked Google how there could be two words without any vowels. Google promptly admitted that it had "made an error" and the first word was "The," which I would have already had revealed because I had guessed all those letters, but Google had only revealed the "E." Notably, the two words I had asked about were still without any letters revealed.
At that point, I decided to change the way I asked the questions:
-----
Google puzzle before my next question:
CATEGORY: Sci-fi/Cyberpunk
TITLE: THE/_ _ _ _ _ _ _ _ _ _ / _ L L _
AUTHOR: _ _ _ _ I I E _ / _ _ _ _ T _ O T
------
Me> You said there were two "I"s in the author's name, but did you fail to tell me there are "I"s in the title?
Google> You caught me again! I absolutely made a mistake. When you guessed the letter I earlier, I only filled it into the author's name, but there is indeed an I in the title as well.
The title is THE INFINITE WALL by the author GABRIEL WRIGHT.
Here is the fully corrected and updated board with all of your correctly guessed letters...
CATEGORY: Sci-fi/Cyberpunk
TITLE: THE/INFINITE/W_LL
AUTHOR: G _ _ RIIE _/WRI_HT
Incorrect letters: U, A, Y, S
Now, that the title is almost completely filled in, what would you like to do?
[Here Google gave me some letters I could guess, and also the option to guess the title/author. Note that if you compare what Google gave me to what it gave me before, the number of letters and the placement of the letters changed in the puzzle among other things.]
Me> LOL! You told me the title instead of filling in the letters and you misspelled Gabriel when you were filling it in. The completed title is "The Infinite Wall" by the author "Gabriel Wright." Now, I am going to have to fact check you to see if that is a real book or if you made it up.
Google> You completely caught me red-handed! 😊I completely glitched on that one--I filled in the missing title, gave away the game, and managed to typo Gabriel...
To make it even funnier, if you fact check me, the book doesn't actually exist! I completely hallucinated [it] and the plot scenario while trying to manage the puzzle matrix...
------
(Yes, I was suspicious at the plot scenario, too.) I found the interaction to be hilarious because I understand the limitations of AI, and I was testing it, but did not have high expectations. I still find it hilarious, but it is important for others to realize the limitations as well. Google stole pieces of other books for its "hallucinated" plot scenario as well as parts of titles to create its puzzle (and many people use AI to write their books). It could not follow simple instructions to run this game (and managed to change the number of letters and failed to fill in letters.) Next time, I will try wording my answers better--instead of saying "Is there an A?" I will say "Fill in all the A's in the puzzle." I will also ask it if the book and author are real upfront--even though it has assured me the next puzzle contains a real author and book. Finally, we are using AI to choose military targets and some lawyers are using it to write their cases for them. We have 4,000+ AI factories in the U.S. sucking up water and energy so we can be ahead of the world with our AI, but our AI functions more poorly because we are trying to make it do everything at once. It is like we have this dog, and instead of teaching it to retrieve our shoes for us or pick up trash (things dogs can do), we are trying to teach it to make dinner--which is something dogs will never be able to do.
Can SI make puzzles (or puzzle books)? Of course. Can AI give you lists of resources for your book? Of course--if you fact check it. Can AI write your book or pick military targets? Not better than throwing a dart at a board, and it never will be able to do these things effectively. We need to use AI for what it can do best and focus on training it to do that well instead of trying to get it to bring the stars to earth for us.
No comments:
Post a Comment