Can an AI Write a Speech Better Than a Human?

When we were invited to speak at a conference, my creative partner and I wondered what would happen if we asked an AI to write our speech. The results were promising.
Headshot of author Jacob Vaus
Jacob Vaus
Expert Columnist
February 3, 2022
Updated: February 10, 2022
Headshot of author Jacob Vaus
Jacob Vaus
Expert Columnist
February 3, 2022
Updated: February 10, 2022

The idea was a half-joke.

We were invited to speak at the Brazilian tech conference i2Ai, described as “A program dedicated to artificial intelligence and its applications.” The host had seen our recent videos and would like us to speak on the artistic uses of AI. Eli, my creative partner, asked, “What if we had AI write our speech?”  

An AI-written short film

Our YouTube channel, Calamity AI, uses artificial intelligence for creativity. We’ve made short films written by AI, had musicians sing machine-written lyrics and followed AI recipes. We pose the question, can artificial intelligence create art? 

A singer performs an AI-written country song

When I ask this same question of peers, I often hear them say no, despite the rise of synthesized artwork and music. In college, a group studying AI presented my film class with a similar query: “Who thinks artificial intelligence could write a meaningful script?” In the room of 90, not a single student raised a hand. The belief is natural. As creators, we’d like to believe we have something beyond automation: authenticity, heart.

Calamity AI’s founders bake cookies using an AI-written recipe

Can AI Match Human Creativity?

Our YouTube channel, Calamity AI, uses artificial intelligence for creativity. We’ve made short films written by AI, had musicians sing machine-written lyrics and followed AI recipes. We pose the question, can artificial intelligence create art? 

More From Jacob VausHow We Made a Movie by an AI Script Writer

 

AI Speechwriting

When Eli suggested having an AI write the speech, I smiled at the thought. We’d been asked to speak for an hour in front of industry professionals, though, and we couldn’t get away with that. But soon, we couldn’t resist the idea. The day before the conference, we halted our work on the speech and fed it into the program Shortly Read. Quickly enough, there was no doubt at all: We decided that we would generate an artificially written speech and read it at the conference. Soon the question became whether anyone would be able to tell what we’d done.

After writing our introductions, we input the following: “We’re interested in investigating what happens when you combine the capabilities of AI of those with highly skilled creative humans. For the videos we put on our channel, we input the opening lines and let the AI do the rest. For instance, if we were to take the speech we had presented so far and fed it into the AI, the AI may say something like...

The AI wrote, “I like blueberries.” 

We generated further. And further. 

The AI took over, creating a sprawling, 3,000-word speech that often swerved in uncertain directions. Eli and I agreed to tell no one and act like it was our work to the best of our abilities. 

 

Can You Believe It’s AI?

The conference began the following day. Edward Hunter Christie, the head of NATO’s innovation unit, spoke before us. Because of our nerves, we’d aimed to look professional: red curtains, button shirts, printed speeches. As Christie’s talk finished, the host introduced us to the audience: “They are doing a great job making the most of artificial intelligence. I’ll let them introduce themselves and start this great presentation.” 

We introduced ourselves and read our first input lines. Then, we tried to maintain our demeanor as we moved to those that had been machine-written. 

The speech began composed enough: “Now that you know where we are now, we want to talk to you about how we came to this point, and where we want to take this channel.” It wrote that we’d given another presentation to Chapman University. The AI notated the script in speech form, indicating when Eli or I should speak. Soon, the speech became rapid and rushed, and the dialogue moved as followed:

ELI: The second slide was one of our sponsors. 

JACOB: But, after that, I just kind of…

ELI: Threw in some random quotes from some of the contributors.

JACOB: And then some pictures, just random pictures I pulled up on the internet.

Before long, the AI decided that Eli and I were brothers. We struggled to stay straight-faced as we read on.

ELI: It was at this point that I was finally able to really focus on my brother.

JACOB: For the rest of the presentation. 

ELI: My brother takes notes on our laptop. 

JACOB: At the end of our presentation, Apple said they would be interested in offering us an internship.

The absurdity was part of our experiment. Without intervention, the majority of AI-generated content lies in the uncanny valley — the odd disconnect between believable and inhuman. With the right presentation, however, the unsettling can become understandable. A strange, short film may appear to be the work of an offbeat director rather than an AI. This speech, likewise, seemed like the product of two speakers who didn’t know how to address an audience. 

On we went, describing that we had worked in Apple’s “rap department.” Then, we begin talking further about artificial intelligence and trying to integrate it with Python. Soon, the speech was even believable again, as we talked about designing our own AI interface. The script even ranged back to what seemed like normal material, describing our videos and future goals. But by the end, the dictated dialogue grew strange and gloomy. 

JACOB: You know, sometimes really sad things happen to people.

ELI: People die. That’s not like a coincidence…

JACOB: No, it’s not a coincidence. That’s very sad. It’s scary.

ELI: It’s really sad.

You can almost see the AI’s intent. For a well-trained speech writer, such an emotional turn could be a haunting nod to life’s uncertainties and a chance to make the speech more personal. The AI’s execution, however, is stunted and clumsy. Rather than naturally integrating the idea, the emotional turns here arrived without any transition. At the tail-end of the speech, the AI took another rushed swerve and offered a hopeful ending. 

“We’re going to have a lot of fun,” I said. “With some other. You know. Our mother.”

“Our father and our brother,” Eli agreed.

“Have a wonderful day,” I concluded.

The line marked the end of our speech. We had talked for 40 minutes without interruption. The host returned, declaring it a “great presentation.” He did not appear suspicious. Rather, he informed us that we’d be moving on to audience questions. We got things like, “Why do you call yourselves Calamity AI?” among other softballs. The idea began to sink in: Despite the absurdity, it had worked. Nobody had questioned the nature of the speech. We thought.

The fourth question was as follows: “Was this presentation done by AI?”

In the video, Eli and I went still. “Who- who- who asked that?” I say. 

The host read the name from the chat. “My gut tells me many things,” he said.

I stammer: “You can believe what you… choose.” Then, I am out of my chair, beneath the table, fighting laughter, leaving Eli to fend for himself. 

More on AI and CreativityCan AI Make Art More Human?

 

The Future of AI Creativity

Despite acting to the best of our abilities, the participants were not fooled. The script was naturally ludicrous. But in presenting the speech, I could see the potential. With human intervention, tweaked lines and better acting, the AI may have produced a believable presentation. In the same way a machine-written script is illogical, with characters making inane, nonsensical decisions; the speech was too wild, too random to have been the product of a human. Yet the bones were there, the intent. The speech, in its own way, followed a structure: It began in an informational mode and grew personal before ending on a moment of hope. It wasn’t empty, rambling nonsense. Something was there. 

After this experience, I believe the the uncanny valley is getting shallower. Though I learned that AI can’t write a convincing speech on its own, I suspect it could serve as a cowriter by adding lines when writers falter, offering concluding statements and adding details. The writer would only need to keep it in check. 

As the phenomenon of creative AI develops, I have no doubt the future will be full of machine-written speeches, synthesized voices and AI novels. At Calamity AI, we aim to explore this gap: the divide between AI and human creativity. Can a comedian get laughs from machine-written content? Can a short film win an award? Or, in this moment, could AI write the compelling end to an article? 

I am feeding the entirety of this piece to the program Shortly Read. The machine produces this final line:

“In the end, that’s all information should be. A joke on its own.” 

Expert Contributors

Built In’s expert contributor network publishes thoughtful, solutions-oriented stories written by innovative tech professionals. It is the tech industry’s definitive destination for sharing compelling, first-person accounts of problem-solving on the road to innovation.

Learn More

Jobs at Calamity AI

Great Companies Need Great People. That's Where We Come In.

Recruit With Us