Akinator and Artificial Intelligence

My eldest daughter refused to get out of bed. She took an afternoon nap and when it was dinner time, she made a lot of grumbling noises when I asked her to come out. My usual tactic is to sit on her legs so she could not escape me while I tickled her feet. This time, I had a more humane idea. I stood beside her bed and said I wanted to show her something. She kept her eyes shut. I then asked her to think of a any character who is real or fictional. Her eyes opened. She peered suspiciously at me, and then glanced at the phone in my hand. She knew I was up to something. “Uhmm, okay.” I told her not to tell me who she had in mind, but I already knew who she was thinking about. I started reading out the questions that are now appearing on my phone. She eventually sat up and asked to see what I was looking at. I showed it to her. It is Akinator.

A systematic approach in eliminating the wrong answers and gathering new information

Akinator is an online game (also available as a mobile app) that challenges the user to think about any character that exists or is fictional, then it asks a series of questions and attempts to guess the character. I cannot remember how I came across it. Someone probably posted it on Facebook or Twitter, and I tapped on it out of curiosity. I am familiar with programming, therefore my immediate thoughts were along the line of how this application is written. It must contain a huge database of characters with keywords associated to them. Then with a huge series of if-then-else statements and algorithms, does the software plow through questions that it tries to identify if are related to the character you have in mind, make some elimination and deductions, and zooms into the very likely answer. For me, I tried using Sesame Street’s Big Bird. Akinator asked me a bunch of questions which got me understanding its deduction process. I answered all his questions truthfully, and he asked if I was thinking about Nemo. I laughed. After telling Akinator that he (it) was wrong, I allowed more questions to be asked. At some point, the question was if the character is from Sesame Street. I knew it hit the mark. Then it asked if it was blue. Then if it was yellow. Akinator was probably going to settle on either Big Bird or Bert, but due to an earlier question if my character has feathers, the guess rested on Big Bird. I thought that was too easy a character to guess. And Akinator guessed it wrong at the first try.
Scarlett Johansson as Black Widow. Image from https://en.wikipedia.org/wiki/Black_Widow_(Natasha_Romanova)
Some time in the evening, I quizzed my second daughter with Akinator too. She thought about Black Widow as the character (which I immediately guessed because she is such a Black Widow fan!), and I thought that Akinator will first begin to find out if the character was fictitious and very popular right now. Very early, it asked if the character was a Marvel character, and I knew the answer would be very quickly guessed. Akinator may seem like something so completely new or even futuristic, but it has been around for a long time – since 2007 to be more exact. Players get to answer from “Yes”, “No”, “Probably”, “Probably not” and “Don’t know”. After Akinator asks about 25 questions while narrowing down the possibility of your character, it then starts to make its guesses. By the third time it has guessed and if it guessed incorrectly, it will ask you to type in what the character it is you really have in your mind. It remembers the questions you had answered, makes the connection with the character you have typed, and now it has actually learned a new character that it had previously not known about. It further enforces whether the information about this character when it next encounter other players that also happen to think about this character.

Not something new

Artificial intelligence seems so smart that it must be only a very recent thing, right? Actually if you were old enough to have been on BBSes, there was a Door utility called Chat With Lisa. The utility picks up words that you have typed out and makes some (rather rough and not entirely accurate) assumption about what you are trying to say, and attempts to answer you as though you are chatting with a real person on the other side of your modem connection. Fast forward to today, you have Siri and Google Assistant. When you start to wonder about how advanced technology has become, you should also wonder how actually simple human logic might actually be such that a machine can also do the same thing. This is not to belittle the wonders of the human mind, but it should get us thinking how erroneous we may be in assuming our own superiority without comparing it with anything else that can easily do the same thing as us. Sure, machines cannot feel or have empathy (it can however emulate or simulate it, but never really doing it), and the human body when compared to any machine of this day, is far more complex, but we really need to understand that the thought making process may not be as complicated as we think. From this observation, the main ingredients in making good deductions is the amount of information or data that is available, and a systematic way of eliminating wrong answers to come to a likely correct conclusion. Hence, if we want to make great deductions ourselves, in a very high level principal, these are the two main assets we need to set ourselves to improve on and to master.

You can only make accurate deductions based on what you know

Tomoe. Image from http://kamisamahajimemashita.wikia.com/wiki/Tomoe
What my eldest had in mind when she was finally about to get out of her bed was Tomoe of the Kamisama Hajimemashita manga. As expected, Akinator at some point suspected that my daughter’s character was a Japanese manga or anime character. Targeted questions were obviously trying to deduce if the character was from Dragon Ball Z. I chuckled, seeing that Akinator’s only chance of guessing correctly is if the developer or the game administrator had known about this much more obscure character than the usual Japanese manga ones. Because I have grown tired of hearing about Tomoe from my daughter, I already knew from the start who she had in mind. I started to say that I would be really impressed if Akinator would be able to correctly guess– Akinator then made its first attempt: Tomoe (Kamisama Hajimemashita). My daughter’s and my jaw dropped. Akinator then also stated in its answer that so far there had been over 16,000 users who thought about Tomoe. 16,000 against what I would expect of millions of users that had ever used Akinator, is a very tiny number of people. Yet at the same time, if knowledge of an obscure incident can be stored in referable memory, it can certainly come in very handy down the road, when one may least expect to use it. But by just asking the right questions and correctly making deductions, one may solve very unanticipated problems that were not initially prepared for. For fun and research, try Akinator for yourself.

Leave a Reply

Your email address will not be published. Required fields are marked *