Artificial Intelligence amalgamated with Machine Learning is making great progress and it has taken a human’s place in many aspects. But there is good news for some humans, it seems that AI and comedy doesn’t go very well.
In past many years, researchers and scientists have tried to give AI bots some personality which also included humour. But they faced many challenges as the base of AI learning, that is the Neural network, specialises in imitation. Through impersonation, the AI learns many things and so it is made to strictly follow rules. This method has even led to an ethics council generation as it was discovered that the AI bots could be taught wrong things by feeding it fake news. So, when we look at how the AI gains knowledge, we come to know that making it funny is a little difficult. Feeding it a few jokes and making it repeat would be simpler but it would not fulfil the point of artificial intelligence.
To tackle this problem, He He and her team of scientists and researchers sought out the method of surprise. Using the local-global surprisal principle they were able to generate puns which were funny to an extent. In this method they would use neural networks and let it do its job and just edit the output a little.
Human Intelligence is necessary ingredient with the neural nets to create humour. In humans we can track surprise by pupil dilation or facial recognition and reactions of the body, in AI bots we can track surprise through probabilities. The probability of a word to appear least in a sentence will create surprise.
In local-global surprisal principle, we have a pair of homophones. One of the homophones is used to generate a non-humorous sentence. We then exchange one homophone with the other homophone and get a word which when changed creates surprise and the contrast is what will make it funny.
With neural networks this principle works beautifully, we feed neural networks with a pair of homophones, for example: hair and hare. It forms a normal sentence with the first word and then swaps the second word for the first. This insinuates surprise and to make the sentence a pun with a little more logic it swaps another word apart from the homophones.
There were many successful results with this method but there is very less data on this principle to work upon. Also, this principle enabled AI bots to generate puns, but puns that were monotonic in structure and formation and hence making it less humorous after a while. They also detected grammatical errors made by the bots during pun creation.
While the research is still ongoing, He wants to use this technology to create an AI bot that can write short stories. With Neural network helping in creating dull string of sentences and using human intelligence to edit those output to make the stories more interesting, she aims to make the AI bot learn to create short stories about things the humans could never think of.
There have been many failed attempts at making the Artificially intelligent machines to learn how to be funny, but this seems to be a breakthrough.
Good news for the comedians is that when the AI machines were pitted against humans for humor battle, humans won over AI bots. Getting news of advancement in Artificial Intelligence everyday and making humans better with AI had many people fearing for their jobs but personality is one aspect where we need to use more of the human intelligence with the neural networks to make it work. Trying to build an AI bot resembling humans may be difficult because, the closer we get to human-like AI machines, the more it is prone to err as humans are far from perfect.