The Opinions - A.I. Isn’t Coming for Moe the Bartender. Not Yet, Anyway.

Episode Date: February 4, 2025

In this episode, the actor Hank Azaria, known in part for his numerous roles on “The Simpsons,” confronts how A.I. is already shaking up the vocal acting world. As he explains the human touches that shape his characters, he also offers hope for a future in which there is still a need for performers like himself. Is it inevitable that artificial intelligence will soon put him and his fellow creatives out of a job?Read Hank Azaria’s essay and watch him perform his most famous “Simpsons” characters at nytimes.com.Thoughts? Email us at [e-mail]. Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com for information about our collection and use of personal data for advertising.

This is The Opinions, a show that brings you a mix of voices from New York Times opinion. You've heard the news. Here's what to make of it. My name is Hank Azaria. I'm an actor and voiceover artist and producer. I'm most known for my voices on The Simpsons, such as... Motor bartender. Comic book guy, of course. Police chief wig him. Snake.

Professor Fring, of course, one of my personal favorites. Cleodish. The slack-jawed yokel often makes himself known. The sea captain, y'er. The old sea captain. Duffman, purveyor of Duff Bear, oh, yeah. I was in a long campoli, the French accent, the Claude, the Scover guy. It goes on and on that could publish a tear doing this for a long time. I didn't really think about AI seriously in terms of voice acting

until about a year or two ago. When it started getting pretty serious, pretty obvious, that it was making a real run at sounding like humans can sound. Obviously, the Scarlett Johansson thing was a big deal that brought into everybody's consciousness. Actor Scarlett Johansson says she was approached by tech company OpenAI to be the voice of ChatGPT. The actress said, no thanks, but when the company released a voice assistant named Sky, it sure sounded a lot like Johansson. And then, since then, I've had not one but two. companies pitch me tech wanting me to get on board with the early versions of it to either protect

my voice, my name, image likeness, and sound, or license it in a way that can be used in AI with my permission. In other words, oh, we want, Mo the bartender to be involved in this. Can we use your voice? You don't need to do it for us, but we have your permission to recreate it. I love my job, and I consider myself the luckiest person in show business. If not on the planet, that I actually get paid a lot of money to do these voices on a show that's lasted or approaching 40 years. And I don't want to be put out of a job on that show or any other. And that is a little bit frightening.

You know, AI can mimic sounds better and better. I think within five years, they'll be able to mimic them pretty perfectly. But I think there's a humanness that the AI can't do right now, at least vocally, and may never be able to do that involves the character's motivation, certain emotions, subtleties of physicality, facially or otherwise, it just can't quite capture

all those little minutia that add up to a human being. For example, let's take Moe. So this is what he sounds like, but there's inherent aggression in everything, Moes. He's angry. Every, oh, in all moments, there's hate underneath everything that he says. Chief Wiggum is inherently stupid.

He doesn't understand almost anything. And there's a child-like innocence to him. And I know, a curiosity, a kind of wide-eyed happiness, just to be alive as long as there's a donut in his future, which there always is. A snake is really sneaky. He's sizing you up always and trying to lull you to a sense of calm so that he can rob you.

those qualities, they're subtleties that are almost more the character than the sound of it itself. And I'm not sure the computer can or ever will be able to do it. What a hubris statement that was. Cut to me out of a job in the cold while computers are laughing at me in Mo's voice. Oh, you thought we couldn't do it, huh? How you like living on the street there, pal? So the creation of Mo and Chief Wiggin, they're good examples of what I'm talking about here as far as the human element in this kind of work goes.

In the case of Mo, it started out as I loved Al Pacino as a kid and a teenager. So I did a young Al Pacino impression, Godfather Al, dog day afternoon, Al. And, you know, I'm dying here. Everybody's coming down on me here, that out of. And I actually auditioned for The Simpsons. I was doing a play in L.A. I was playing a drug dealer. And I thought the Al Pacino voice would be good for that.

So I auditioned with this voice here. And they said they wanted it to be gravelly. So you take Long Al Pacino and you make him gravelly and you get motor bartender. So a lot of times, like I got a bass voice I'm working with. Then you give a little gravel on top. Lately I've been imitating Bruce Springsteen a lot. I got a Springsteen tribute band, another hero of mine, since I was a teenager. And actually, Mo is kind of halfway between Bruce and young Al Pacino.

Right in the middle there. Mo the bartender. Agadour, Spartacus from the Burk Age, is my grandmother. We're Sephardic Jews, Spanish-speaking Jews. And she sounded like this. And not only is a good example, because she was very loving and maternal and feminine. So it isn't just the timber this voice. It's also the affection and the maternal instincts that come through with this voice.

And then there's a physicality to the characters that exist. If you're doing a scene where you're chopping wood, it helps actually, I don't think Moe's ever chopped wood, but if he was, he sounds like, yeah, all right, well, get that going for you as soon as I get this wood chop. Man, it's hard chopping this wood. So you kind of got to do it. It's hard to fake that quality in your voice unless you're exerting. Not only putting subtleties of emotion and motivation in,

but yes, the physical exertion, the way emotion changes your voice. From what I've heard so far, AI can get pretty close, but pretty close, you know, only in horseshoes, really. Does that count? What AI can do is bring analysis and ideas. I believe Google has this app now. You can put in any piece of writing you like.

My wife put in a poem she wrote some years ago. And it will create a 15-minute podcast, a male AI voice and a female AI voice, chatting about the poem. It was utterly convincing. It was very smart, incisive commentary. The voices were completely believable, including ums and pauses. It sounded extremely lifelike. Now, there wasn't a tremendous range of emotion.

When I tried to make it try to be funny, it really fell short. But what it had to say was quite interesting. So it's like chatGBT, right? It gives you some really good ideas. And, you know, if you don't like what the chat GBT came up with on the first try, just hit a button and it'll endlessly try variations on it. You're eventually going to get an idea or two you like. It's a tool like any other.

And, you know, similar to the vocal aspect, It still needs a human to bring it home, but it's an incredible aid along the way. Forgive me, my writer, friends. You know, the way TV's done is there's a showrunner, a head writer, and there's a writing staff. I don't think you could replace the showrunners. I don't think you could replace the people with the vision

and the real eye on the prize of creatively what we're going for here. But I'm pretty sure a headwriter at this point doesn't need a staff. I'm pretty sure, based on chatGBT prompts and then just zushing scripts that the chatGBT generates, as long as he's rewriting it, or she, or they, I think that can be done. And I can't imagine it's too long from now that a studio is going to keep paying for what they know the computer can do for free. I look at the AI visually, vocally, from a writing standpoint, it's very exciting. I enjoy it. I think it's mind-blowingly fascinating and fun

to see the art and the songs. I just came upon the AI-generated What if Led Zeppelin 2 was recorded in the 50s in Rockabilly version? It's awesome. Way way down inside, honey, you need it. I'm going to give you my love.

I'm going to give you every inch of my love. I love. I love Led Zepp. And, you know, I just find that really fun. I don't want to be replaced as a vocal performer in animation or on camera by the same token. How fascinating that that can happen. I hope humans and AI can work together and collaborate.

I mean, that's my hope, right? Like, I do a, yeah, what's up, Doc? I love Bugs Bunny. I do a sort of passable Bugs Bunny. What's up, Doc? That's not too exact. but the computer could make it much more exact, especially if it's using my performance.

And I'm probably a good enough performer, and I know Bugs Bunny so well that I could probably, I could probably do it, and we could tweak it to get it pretty much exactly like Mel Blank sounded. But you still need someone like me, not just creating the voice to begin with, but also then knowing how to fix it once you hear it back.

People are going to listen to and enjoy and watch what they like. And if it's passable and if it's good, they're going to like it and they're going to listen to it. And they're not going to care whether AI generated it or a human generated or some combination of the two. Right now, that just basic human response is going in my favor. Because I'm pretty confident that what AI generates by itself as motor bartender or anything else isn't going to cut it. But if it does start to cut it, people are going to listen to it, and they're going to be grateful that it's so readily available. I mean, like what happened, you know, to the music industry.

I cried a tear because the record industry reinvented itself. I got to listen to all the music for free all of a sudden. So I don't think people are going to feel much differently about any of this. If you like this show, follow it on Spotify, Apple, or wherever you get your podcasts. This show is produced by Derek Arthur, Sophia Alvarez Boyd, Bishaka, Fibylette, Christina Samuelski, and Jillian Weinberger. It's edited by Kari Pitkin, Alison Bruzek, and Annie Rose Strasser. Engineering, mixing, and original music by Isaac Jones, Sonia Herrero, Pat McCusker, Carol Saburo, and Afim Shapiro.

Additional music by Amin Sahota. The fact check team is Kate Sinclair, Mary Marge Locker, and Michelle Harris. Audience Strategy by Shannon Busta, Christina Samuelski, and Adrian Rivera. The executive producer of Times Opinion Audio is Annie Rose Dresser.