Will Robots Replace Improv Comedians?

Technology + Improv?

As someone who is both a web developer and a lifelong improviser, you'd think I'd be open to the intersection of technology and improv. But, I am not. I'm a bit of a stick in the mud when it comes to innovations - not because I don't find technology convenient and intriguing - but more because I am an old fashioned free spirit at heart and the idea of robots taking over the world is a genuine fear that I have. That being said, I'm also fascinated by that fear. My favorite book is Player Piano, by Kurt Vonnegut, which explores a "future" society where machines have taken all of the jobs over from humans, thus leaving everyone without a life purpose. Ahhh!

I met Kory Mathewson a few years ago in Portland OR, when we were both performing in the Stumptown Improv Festival (side note: one of my all time fave festivals!) I certainly don't know Kory very well but from our interactions that weekend and over social media the past few years, I'd describe him as: A really smart and talented improviser who is also a genius computer engineer person. I guess I'll have to ask him if that is right or not. But - he is always working on a crazy technology project. And his latest deep dive? An improvising Robot. 

Uh, Oh.

Kory and his project partner, Piotr Mirowski, have been working for the last 2 or so years on the AI for an improvising robot. (I don't even know if I just used AI in the right way so we are off to a great start.) Needless to say, as a professional improviser - and given my fears on robots taking away all of our jobs and ya know, our overall purpose for living - when Kory told me he was working on an AI/Improv project, I was more than skeptical.

My initial thoughts, in fact, were:  

  1. I am going to hate this.
  2. This could never work. 
  3. AI can’t read emotion, understand intonation, sarcasm etc.
  4. AI won't be able to find “game” or “what’s funny”
  5. Can AI have a sense of humor? Or an understanding of humor?
  6. And what about physicality and characters?!
  7. Did I mention I hate this?

That being said, I was curious.  I chatted with Kory, and read through their Scientific Article* titled "Improv Comedy As a Turing Test."  Much of their work thus far is explained in Scientific jargon for Scientific folks (not my speciality) but, as a non-Scientific folk, I was nevertheless intrigued and conventrated hard enough to get the gist.

NERD ALERT: *I asked Kory if "scientific article" is the actual term. He said it is "a Scientific Article, but more specifically, it is a peer-reviewed conference publication." Okay.

Here is part of the Abstract straight from their "peer-reviewed conference publication"


The best improvisational theatre actors can make any scene partner, of any skill level or ability, appear talented and proficient in the art form, and thus "make them shine." To challenge this improvisational paradigm, we built an artificial intelligence (AI) trained to perform live shows alongside human actors for human audiences.


Kory and Piotr have actually performed live shows with the robot (30 performances to a combined audience of almost 3000 people) to test this thing out.  That's a lot of shows that this robot has done all over the world - and I'm a little jealous. I've done my bit of research and now I'm on a mission to find out...

Will A Robot Take My Job?

Let's find out.

Q: Does the robot have a sense of humor?

A: So, no. The AI (known as A.L.Ex aka Artificial Language Experiment aka the Chatbot) does not have a sense of humor. Much of the laughs in the show come from the human justifying the insane thing that A.L.Ex just said. So, if you know the short form game, Actor's Nightmare, or have ever seen an improviser do a show with a random person from the audience, the concept is very similar. 

  • If that's the goal - then this makes a lot of sense (and I don't have to worry about a robot taking over my job so soon.) It's more of a gimmick than anything else - but can give freedom to a solo improviser on the road who wants to do a 2-"person" show.  
  • The problem there is that the human is doing all of the heavy lifting in the scene. If improv is supposed to be about support, both parties doing their best to make each other look good, than Mr. Chatbot here ain't pulling his weight. 

POINT: Raero


Q: Does ALEX always sound "robotic?" What is its voice like?

A: So if you thought improv wasn't (gender) binary enough, get this:  A.L.Ex understands if he is being labeled male of female in the scene, based on name, and then randomly chooses one of 60 different voices to go along with its character.  So, let's say you label your robot scene partner, "Rachel" in a scene (what a lovely name), A.L.Ex could take on the voice of an American female. But, if you call the robot "Richard" - perhaps an Australian male's voice will be heard. 

  • Obviously this isn't ideal (what about a non gender binary artform!) but honestly, I am so impressed it can do even that much! I assumed everything would sound like Siri. 
  • Voice can communicate a lot in an improv scene. Having different voices and dialects adds immensely to a scene. Already the Chatbot beats me on that front. I'm terrible at accents!

POINT: Robot


Q: How much training does it have? Has it taken all the levels at Second City or what?

A: Since a lot of the humor is coming from the human's justification, it sounds like Chatbot isn't all that funny. I've trained at many places including ImprovBoston, UCB, The PIT, and Reckless Theatre. (Take that!) The Chatbot's training is in its programming. The more context that's built in, the more it can do well in scenes. Overall, I should expect that it will work naively - and that I will shine as the better improviser. It's much like working with a less skilled improviser - which sometimes can be a real workout but can be a lot of fun as well.  


POINT: Raero


Q: Timely question! How would it handle “Social Justice Improv?"

A: Politics bleeds into comedy every day - but nothing had really prepared me for the changes we've experienced on stage and in the improv classroom in the last year. Sex, Race, Gender... when "tough" topics arise on stage, it's hard to improvise in the moment and be smart doing it. Especially when we are all trained to "say yes!" to whatever is in front of us. I've learned a ton in the past year or so, and I know my teammates have as well.

When these on-stage moments happen, the main things we are trying to accomplish are - how do we comment on a "tough" topic, stay above the line, make sure the audience knows we personally are not ignorant (even if our character is) - all while trying to make the scene funny. I mean, that's a lot to juggle.  I wonder how A.L.Ex would handle this, as humans themselves have enough trouble. I asked Kory and he said that how the robots respond in those moments depend on 1) their masters' opinions (because they programmed it) and 2) how they are “trained” to filter their output. Basically, they can stoke the fire and see how the robot responds - and then reward or punish these systems when they do correct/incorrect things. Uhhh... that's kind of the same way we have to train humans, actually!? It's quite hard to teach improvisers how to do feminist, non-misogynistic improv when they don't even know how to identify that it's happening in the first place! I've had these very lessons and conversations myself! Damn! Maybe Chatbot ain't so different than us afterall!




Q: How good is the robot at initiations? Because I'm hilarious!

A: The Article mentions “sentiment analysis” (say what?) so I asked Kory about it. Sentiment Analysis is a basic way that the AI can determine emotion/sentiment coming from the Human. This is so freaking interesting. Here is a basic example:  If an AI were reading Twitter - it would match the written text as well as the emoji someone used, and then make its decision on sentiment based on that. For example, if someone wrote. “Wow. Really Perfect day so far :/ “ - it would understand enough to know the person wasn’t having the absolute most perfect day.

Anyway, if you want to do positive, agreeable improv, then they can program the robot to 1.) Make positive statements and 2.) Say something that will make the HUMAN respond in a positive way. That latter one, to me, is the kicker. It’s basically programed to create really smart initiations that will get a specific response from its scene partner.

For example: 

“Look, Brenda… don’t try to cheer me up.”
That initiation will illicit a very specific response from the improv partner. (They’ll probably try to cheer you up!) 

“I know you’re going to freak out, but…”
That initiation informs your scene partner to, in fact, freak out.

As an improv instructor, I can tell you - it’s very hard to get improvisers to make initiations that are that clear in intension! In this way, the Chatbot may be better at initiations than, say, the average improviser! Initiations vary by skill level, of course. 


POINT: Robot
(I personally think i'm pretty good at initiations but I'll let it take this one...)


Q: Emotions are really important in improv. Can the robot emote?

A: Nope!

POINT: Raero


SCORE: Raero: 3     Robot: 2


Phew - that was a close one.

Ok so a Robot isn't going to take my job— yet. 

I'm going to be honest, ya'll. I came into this HATING the very idea (did I mention that?) and although I still don't love it, I am fascinated by it. I definitely at least want to try improvising with A.L.Ex at some point. I mean I have to, right? (Of course, this is how it all starts...)

One last thing from their article that I wanted to mention was the following: “The Audience developed emotional attachment to the AI.”  It's creepy but this makes sense to me. I can ‘feel’ for almost anything with the right circumstances— but I'm a very emotional person. I mean, I bawled during the movie, “HER.”

But going off of that: Kory and Piotr "chose humanoid robotics because the more realistic an embodiment is the more comfortable humans often are with it; though comfort sharply drops when creatures have human-like qualities but are distinctly non-human.” In English this means, audiences are comfortable with a little R2D2 looking thing - or even a talking toaster. They can have empathy for those things. A Flat computer screen doing the talking doesn't work as well even though it uses the exact same program. That being said, when the programmers try to make something look too human (body, face, legs, the whole 9 yards) but it talks and acts like a robot? People are not comfortable with that. 

People can be so fickle, right?

At the end of the day, Piotr reminded me that audiences go to the theater and live comedy (or movies for that matter) to see and experience human connection. And that's a relief because that's certainly something I can do that A.L.Ex cannot. So... I guess A.L.Ex won't be taking over teaching my classes any time soon (plug! New Hip Hop Improv Classes open for registration now!)  But, I won't be mean to A.L.Ex if I see it at DCM. I won't pre-judge it. I may even ask it to do a quick set at The PIT sometime.

I know I'll be funnier.


Interested in more about this AI project? Check out https://humanmachine.live