Artificial intelligence can have many technological applications and help humans in a wide variety of tasks, but making jokes was not one of its main objectives.
Still, a website has developed a neural network that is learning to be funny with memes.
Humour can be a great symptom of intelligence when we learn a language the most complicated level and that shows us that we master the language is to be able to make jokes with ease. The same goes for artificial intelligence systems if they are able to make jokes it is that they understand human beings better than we thought.
ImgFlip It is a website that allows you to quickly create a meme, that joke format between image and text that has become one of the kings of the internet. This platform now has a neural network that can make memes for us although he is still learning.
Neural networks, whatever their purpose, require exhaustive learning based on examples to become increasingly intelligent and autonomous. It is a process similar to the human brain that uses the trial error method to, throughout its life, be more expert in a task.
We also use observation and imitation to learn and these methods are what this neural network is based on. The web explains that its artificial intelligence was trained using public images generated by users of the Imgflip meme generator for the 48 most popular meme templates.
They are memes that we have all seen more than once on the internet and that are the most popular and easy to include a text. Each image expresses a feeling and history, there is the photo of the boyfriend looking at another girl with his girlfriend in front, we also have the photo of Yoda or Fry from Futurama.
As all are usually linked to a certain structure that repeats itself, the neural network has already been learned and you only have to add the context to vary the joke. If we press the same photo several times we will see how the text changes, but the structure is similar.
To get a really funny meme you have to be patient or help the system by including some predictive text that gives you a guideline to follow. With practice, it will become more and more fun, although simply by seeing the somewhat absurd memes that you can also have a good time.
In this article those responsible for the neural network explain in detail how work has been done to generate this neural network. From the first data collection process with 100 million months to the same 48 images.
“The first 12 characters (of the code) are the template ID of the meme “In this way, the neural network understands what image it is and thus continues to fill in the code that helps the network know where the letters should be placed, how much the text can occupy in the image and the context and languages that are usually used.
For now, we can be calm, but in the future, we may see a comic robot performing a monologue in a theatre that he has prepared himself.