How does Akinator actually do the thing?

636 views

How does Akinator actually do the thing?

In: Technology

7 Answers

Anonymous 0 Comments

By having a very large database with things/people/animals and their properties.

Then it just ask about random properties and if you say that your answer has that property, it eliminates all the answers that don’t have them. Repeat it a few times and you’re left with a final answer.

Anonymous 0 Comments

So, Akinator has a database of questions it can ask, which is connects with its list of possible figures. Over time, it learns to associate certain questions with certain figures. On the desktop version, you can (Or at least used to be able to, I’m not sure now) see what answers it expected you to give once it has guessed your character. It also used to ask for your age to help narrow down potential suspects. Though I’m not sure if the mobile any newer versions still have some of those features.

Anonymous 0 Comments

It’s just process of elimination. You’ll notice that it quite often asked the same/simular question multiple times in different ways i.e ‘Is your character a female?’, ‘Is your character feminine?’, ‘Is your character a woman?’.

It’s just incrementally eliminating possibilities from all of it’s lists of traits based on the context of previous answers until there are only a hand full left, and it will pick one.

You’ll also notice that if it’s wrong and you continue, it almost always gets the next one right, because that small pool is then further eliminated from.

And that list of traits to characters grows every time someone beats him and inputs their character. Obviously you can still beat him, but you have to get pretty obscure these days.

Anonymous 0 Comments

It’s….game of 20 questions. A very well understood game of divisive choice making. It would be pretty easy to compile a list of popular fictional characters and attach attributed to those characters that can be categorized and queried on. You’re talking a dataset that’s probably only in the millions.

Thing is…your dataset doesn’t need to be complete at day 1. It just needs to be reasonably substantial to make the guesses have a good shot at being correct. There’s no real penalty for being wrong.

Anonymous 0 Comments

It simply matches your answers against its massive database of characters. If you think about it, a game of 20 questions with only two answers, yes and no, will yield 2^20 possibilities, will yield over a million combinations of yeses and nos.

Anonymous 0 Comments

It’s literally a game of [Guess Who?](https://en.wikipedia.org/wiki/Guess_Who%3F) except with millions of subjects and hundreds of data points about those subjects instead of just “Does your person have a bum chin?” or “Does your person wear glasses?”

Akinator simply asks you the questions that apply to the largest number of subjects, and when you answer no it “flips down” all of those subjects like you would in a game of guess who. It keeps doing this until it has a very low number of subjects left and then guesses.

Anonymous 0 Comments

Let’s say you pick some super random nobody obscure background character that nobody has ever heard of. You answer all the questions and finally it gives up and asks who you were thinking of. It saves all your responses and now it knows that person. The next time it may not guess your character because it may ask different questions. Each time though it builds more of a database on that character and eventually it will get it right.