徐泽洲：领导干部要做“看齐”的表率

What makes something funny?

Comedy is a lot more than just a punchline - it's a blend of surprise, wordplay, unexpected twists, great timing, and emotional impact. It can be absurd or subtle, something everyone gets, or just a few people do. People laugh for various reasons: because they feel discomfort, surprise, delight, because they relate to something, or just because they feel closer to others.

For AI technology, these human experiences are difficult to copy. To make a good joke you need more than just a sequence of words, you also need to understand your audience. As studies point out, humor works best within shared cultural backgrounds. It depends on grasping not just what is said, but also what is implied.

So, can AI be funny, or even funnier than us? Short answer: sometimes.

AI’s comedic abilities today

Some of the latest large language models, like GPT-4, are fully capable of generating jokes that, at least by certain standards, are quite funny. That’s because these models have been trained on large datasets that include various kinds of comedy and entertainment. In fact, some studies found that AI-generated jokes were occasionally rated as funnier than those written by humans - participants often couldn’t tell the difference, and many favoured the structure, pacing, or originality of AI jokes.

Still, there's a difference between generating something that looks like humor and truly understanding why it’s funny. It’s easy to observe that AI's attempts at comedy often lack soul. While it has no trouble with mimicing patterns and structures (like knock-knock jokes), it struggles with nuance, context, and emotional depth.

So far, the ongoing debate around whether AI understands humor or is simply using patterns to perform a statistical trick hasn’t led to a clear answer. Although, sources online break this down: most large language models are pattern matchers. That means they don’t know if something is funny, they only predict what words are likely to come next based on huge datasets. AI humor usually involves borrowing from already existing tropes and punchlines.

Whether this imitation is enough or not is a matter of personal preference and perspective. Ultimately, it’s up to you to decide where you differentiate between simulation and genuine insight.

The limits of AI humor

As mentioned before, the key challenge for AI is cultural context, as humor often relies on shared experiences, inside jokes, and references that are only logical in certain communities. What lands in a small town might not be funny in a big city, or even in a neighboring town.

Because artificial intelligence doesn’t have lived experience, it can easily miss the subtle cues that trigger stronger reactions to jokes. It’s often pointed out how AI jokes can feel off - like they’re told by someone who doesn’t really grasp why they’re funny. The structure might be right, but the soul’s not there to be found. The reason for that lies in the lack of emotional grounding that AI simply can’t reach.

A new kind of comedian?

Despite its obvious limitations, AI is slowly shaping its place in comedy. Some models are even being trained specifically to write stand-up routines or sitcom scripts. A few comedians have already collaborated with AI systems for comedy shows, blending machine-generated one-liners with spontaneous human improvisation.

As is often the case with technology, this points to a possibility of AI not necessarily replacing comedians, but rather assisting and strengthening them. Writers can leverage artificial intelligence to ideate on jokes, get rid of creative block, or \recycle old material with fresh methods.

From an entertainment perspective, AI is already used for writing ad copy, memes, and social media content - all areas where quick, clever humor pays off.

Ethical questions

The fun fades when AI-generated humor steers into sensitive or controversial territory, which is not unlikely to occur since, unlike humans, AI can’t pass true judgment, doesn’t possess empathy, and doesn’t understand cultural nuance. However, AI models learn patterns from large datasets shaped by human culture in all its complexity - so there is still a risk of them reflecting and amplifying flawed biases.

Even with ethical safeguards and filters, there’s a possibility that AI could generate offensive, harmful, or tone-deaf jokes. When engaging with AI, it’s important to take into account the dangers of inappropriate humor arising from flawed training data, inadequate contextual understanding, or reinforcement of cultural clichés.

The future of AI comedy

The bottom line is, AI is capable of being funny, just not in every way that is obvious to humans. It operates on patterns of humor and is capable of generating good jokes, and even making people laugh. But genuine comedy, one that taps into lived experience, social cues, and emotions - that remains uniquely human territory.

On the other hand, as AI models evolve with the help of more nuanced training data, they might be starting to grasp subtler forms of comedy. They don’t understand humor the way humans do, but maybe they don’t need to do that at all. Sometimes, if a joke makes you laugh, the source of it doesn’t really matter.