The Humor Algorithm: Can AI Learn to Understand Sarcasm?
Interacting with AI bots is becoming more interesting. Instead of providing bland responses, apps can now add humor to conversations. They can grasp users’ tone and respond with witty comments. But how far will machines go in understanding sarcasm? Will they be able to crack jokes that make us burst into tears—from laughing?
Searching for Funny: GPT-4o, Grok, and Other Programs Display Sense of Humor
Humor training is a big focus area for tech companies that aim to build bots that recognize sarcasm and jokes like humans. Algorithms are trained on large amounts of data consisting of texts, audio, and emotional content to understand contexts and distinguish between being funny and being serious.
Recently, ChatGPT unveiled its new, most powerful language model, the GPT-4o. The “o” in the name stands for “omnipotent” to indicate the model’s capabilities to handle multiple forms of content: text, images, audio, and video. Language models differ from each other in their capabilities depending on engineering and training structure.
Among the improvements that the GPT-4o brings are faster responses, better image and video recognition, an increased number of languages, and the ability to joke. In a demo video, when asked to be sarcastic, ChatGPT responded in a tone that many users found to be similar to Scarlett Johansson’s voice in the movie “Her”. The bot’s answer was quite funny, especially the last remark “Let’s get the party started, or whatever.”
GPT’s sense of humor is challenged by other bots, including Elon Musk’s Grok. Developed by xAI, Grok is a generative AI chatbot whose first version was introduced in November 2023 with a bold introduction:
Grok is designed to answer questions with a bit of wit and has a rebellious streak, so please don’t use it if you hate humor.
The program has real-time access to X, formerly Twitter, due to which it provides up-to-date information and can crack jokes on a broad range of topics, including politics, sports, and crypto. If you want Grok to make up a joke for you, you need to be a subscriber of the X Premium. If you want to see some examples, there are always posts shared by users. Here’s one where Grok jokes about Bitcoin and cold wallets.
The creation of Grok has a history behind it. Elon Musk, who co-founded OpenAI in 2015, doesn’t approve of the company’s current policy. According to Musk, OpenAI is a maximum-profit company with closed-source code that doesn’t serve the benefit of humanity. This March, Musk even filed a lawsuit against OpenAI for not being true to its initial mission.
In contrast, xAI claims to take another strategy. The company made Grok open source in March. Being open-source means that its code is available to everyone to see, understand how the bot works, and use the code for new apps. That being said, while ChatGPT has free versions available for users, Grock only offers paid versions.
Researchers Teach AI Algorithms to Detect Sarcasm With the Help of Sitcoms
Sarcasm detection has been a part of AI research and development since the 2010s, coinciding with the rise of machine learning. 2024 marks new progress in this aspect. Researchers from the Netherlands have built a tool that recognizes 75% of sarcasm in sitcoms. The program has been trained on the text, audio, and emotional content of words spoken in American sitcoms, including "Friends" and "Big Bang Theory."
According to an article by The Guardian, scientists at the University of Groningen’s Speech Tech Lab announced they intend to improve the accuracy of AI detection further. For this purpose, they consider adding visual hints to the AI’s training process, such as eyebrow movements and smirks. The researchers stated that the tool is unlikely to achieve 100% accuracy because it’s something that even humans can’t. They also mentioned the challenge of figuring out if the bots' answers are genuine or sarcastic.
Do We Need AI-Generated Humor
Already today, AI bots can create memes, be sarcastic, and understand jokes between the lines. Sometimes they succeed in making us laugh, and sometimes they don’t hit the mark. Good or bad, seeing AI making jokes adds engagement to human-bot interaction. Scriptwriters, comedians, and content creators can use AI tools as assistants to create funny material.
Computers joke depending on the way they are programmed and the data they are trained on. Randomly coming up with new lines, they present risks of being sexist, racist, overall unethical, or not funny. The human sense of humor and the ability to manage the programs are what can make bots naturally and genuinely entertaining.
But even when imperfect, watching bots learn and master a sense of humor is interesting in itself. Like in the famous scenes from Terminator, when the cyborg played by Arnold Schwarzenegger, learns to smile, he looks pretty awkward and funny. It feels like unsuccessful jokes or unnatural mimicry make the machines even more charming. If you don’t know what we are talking about or want to see the Terminator smile once more, below is a YouTube video with his smile scenes.
Anahit Avetisyan
Author
Web3 writer and crypto HODLer with a keen interest in market trends and recent technologies.
