Berta and I were at the toy store this week, looking at, um, toys. We saw this crazy thing that I have to mention.
It’s a small orb-like electronic device. On the front is a screen made of small red lights and four buttons. The buttons are labelled Unknown, No, Yes, and Sometimes.
When you turn it on, it starts asking you questions. Usually around 20. You begin by thinking of a thing - anything - and answering the questions as appropriate. The device has an unnervingly high success rate.
Here is an example question list for the animal “Anteater”:
- Is it an animal? - Yes
- Is it larger than a microwave oven? - Yes
- Is it dangerous? - No
- Does it have claws? - Yes
- Is it brown? -Yes
- Can you buy it at a store? - No
- Does it live in groups? - Unknown
- Is it an herbivore? - No
- Can it run fast? - No
- Is it small? - Yes
- Is it a predator? - Yes
- Does it lay eggs? - No
- Does it have a long tail? - Yes
- Is it involved in movies? - No
- Can it growl? - No
- Is it a carnivore? - Yes
- Can it climb? - No
- Does it have short fur? - Yes
- Does it bring joy to people? - No
- Is it smooth? - No
- It's an anteater? - Drat! It guessed it again!
An anteater isn’t all that obscure, but it seems to know a lot. Some other things that it got fairly easily were Human, Zinc, Pomegranite, Toaster, and Building.
Incidentally, I had just done anteater one time before I transcribed all of this, and it asked me a completely different set of questions. After checking out the 20Q web site, it’s apparent that this trinket uses a neural net to figure out what you’re thinking. Neural nets are very cool, much more interesting than the simple decision trees that I had originally thought might be used.
Does using a neural net reduce the number of possible items, though? You figure that a decision tree can result in at most 2 to the 20th items - that’s just over a million things. If you factor in unique results for each of the four options (instead of just yes or no, but it would certainly never be like this), then you get just over a trillion items. I’m going to guess that the neural net can probably guess somewhere less than the million items of the b-tree, but that’s just intuition and not math knowledge.
Play and contribute to the online version of the game, and see how uncanny the results are.
