While AI has made strides when it comes to comprehending and responding to jokes, there is still a long way for the technology to catch up with human humor. AT MIT, researchers show that AI systems can understand quite rudimentary humor (puns & wordplay) with a 60% hit rate. However, AI Models still do not find it easy to grasp more sophisticated ways of using humour such as irony or sarcasm. Nevertheless, within these obstacles have been the inclusion of humor detection in many companies’ chatbots, like OpenAI moving to build or create a more entertaining experience for users.
Let’s take GPT-3 for instance, the model behind many other famous AI systems that gets trained on both recognizing a joke and the proper response to it. OpenAI once published a study of their own, reporting that out of all pairs where users told GPT-3 a joke, 55% of the time the model was able to provide an appropriate funny reply. And while that number is impressive, it’s still lagging behind the nearly-90% of times humans respond favourably to basically the same jokes.
AI funny depends on a database full of iokes and structures behind human language. An example is IBM’s Watson which trained with 2000+ jokes and then tested if the system could tell whether a joke was funny or not. It took a lot of tuning to get Watson to understand humor in things like ads or conversations, but even then, it failed when faced with more subtle forms of humor. When Watson answered a user telling a joke, it was mostly logical, which is humour based on patterns learnt from prior data (i.e., nothing real here).
Humor is one of the ways AI can create illusions of when it sounds human — but in reality, even if you write full paragraphs revealing your humanity here, machine learning will respond out-of-the-box in a customer service setting. AI systems capable of “chatting with AI” and having some fun friendly banter with customers are utilized by companies to enhance customer experience. The startup ChatGPT, for example, has developed a human-like chatbot that elicits friendly and witty responses from the customer service representative leading to 15% more user interaction compared to any traditional bot.
AI isn’t going to be as funny as a human yet, but it is getting better all the time. Talk to ai says its goal is expanded immediacy for human-AI experiences, even when the AI is getting more sophisticated with its jokes. AI systems analyze hilarious stuff, identify words connected with humor and make suitable jokes to enrich the conversation. AI may not fully get every punchline yet but it is certainly getting good at having a light-hearted banter.
Indeed, some users say that AI systems such as talk to ai can cause a smile or laugh due to its capability of performing basic jokes or exchanging banter. Based on user feedback, humor led to a 30% increase in customer retention with services that integrate humor-oriented AI interactions. This incremental development in AI humor demonstrates the potential of this technology to engage the users at a more natural (and thus playful) level, which is encouraging notwithstanding its incomplete mastery over possibly the most difficult of all human emotions — humour.