Can artificial intelligence effectively generate humor?

Artificial intelligence (AI) has been a topic of great interest and curiosity in recent years. With its rapid advancement and integration into our lives, many have wondered if AI can truly possess a sense of humor. The BBC recently conducted an experiment involving a professional comedian and jokes written by AI, shedding light on just how much machines understand the nuances of human humor.

Karen Hobbs, an experienced comedian, took on the challenge of performing a set that was entirely written by ChatGPT, an AI platform developed by OpenAI. This performance was quite daring, as Hobbs followed three comedians delivering their own human material. Over the years, AI has become more mainstream and has found its way into various domains. However, its rise has not been without controversy and concerns about the potential threat it poses to human creativity.

Generative AI, such as ChatGPT, operates by processing extensive amounts of text data, allowing it to generate responses statistically based on patterns and relationships within language. This does not necessarily equate to true creativity but rather synthesizes existing information into unique combinations. The philosophical question of whether this can be classified as true creativity is yet to be definitively answered.

Alison Powell, an associate professor of communications, brings an interesting perspective to the debate. Drawing from her background in improv comedy, she highlights the importance of understanding how jokes work. While AI may be seen as derivative, there are risks for comedians, such as data theft and regurgitation, as generative AI tools often train on internet content without permission.

Another concern lies in the potential for AI to progressively improve its comedic abilities, posing a threat to aspiring comedians who rely on innovation and originality. The controversy surrounding OpenAI's alleged use of copyrighted material emphasizes the need to protect the creative work of comedians and prevent unauthorized use or replication.

Les Carr, a professor of web science, expresses concerns regarding the potential stealing of content by AI platforms from comedians who heavily rely on social media platforms to showcase their talent. Protecting intellectual property becomes crucial when AI starts replicating and sharing content that has already resonated with audiences, without clear sources or attributions.

During the experiment, Hobbs discovered an odd issue with ChatGPT: it defaulted to writing jokes in the voice of a male comedian, relying on crude stereotypes of millennial women. The resulting jokes leaned on clichés and lacked satisfying punchlines. This highlights the limitations of AI models when it comes to delivering well-structured and impactful comedic narratives. Unlike human comedians, AI tools lack the ability to adapt in real-time and read and respond to the audience.

However, researchers are exploring ways to give AI a better understanding of the world, allowing it to adapt and engage with audiences. Ongoing efforts aim to enhance AI's audio capabilities, social awareness, and comedic timing. Current limitations aside, experts like Michael Ryan predict a future where truly funny AI comedy sets will become a reality.

It is important to distinguish between the capability of AI to generate jokes and the authentic delivery and vulnerability that audiences expect from human comedians. While AI-generated jokes may statistically perform well, replicating the complete experience of on-stage comedy mastery remains a challenge.

Some may argue that digitizing comedy is a futile and ethically questionable endeavor. Developing generative AI models requires vast resources and energy, potentially overshadowing more productive investments in supporting young comedians and nurturing cultural production.

As Hobbs took to the stage, uncertainties loomed as the audience included many who had never experienced live comedy before. Drew Gorenz, a PhD student, conducted an interesting experiment comparing human-generated jokes to those created by AI. The AI jokes performed surprisingly well among readers, but specific prompts were necessary for satisfactory results. While AI possesses potential in humor generation, it lacks human authenticity and adaptability.

For now, comedians can breathe a sigh of relief, knowing that the unique blend of vulnerability, timing, and stage presence they bring to the craft cannot be easily replicated by AI. The human experience of bombing or triumphing in front of an audience remains quintessentially human, and AI models have yet to truly grasp this elusive quality. So, for now, comedians can continue to hone their skills and deliver the laughter that only human connections can bring.

Earlier SSP reported that X secretly uses your tweets to train its Grok AI model.