Why didn’t Alexa make the Wheel of Freestyle on The Tonight Show with Jimmy Fallon?
🥁 ….. Because she couldn’t spit re-prompts
Did you get it? Maybe that was a bit corny
We all know Jimmy Fallon and The Tonight Show. He has a segment called “Wheel of Freestyle” where, you guessed it, guests will freestyle rap randomly generated words. But because of COVID19, in-person guests have had to take a hiatus which led to a newly invented segment called “Hey, robot!” A few days ago, he and his guest Ricky Gervais played this game where they had to ‘trick’ Alexa into saying a word blindly selected from a jar.
Round 1: Jimmy draws the word ‘yoga.’ Ricky gives it a shot — can he make Alexa say it?
Ricky: “Meditative exercise from the Far East.”
Alexa: “You have ‘Guided Meditation — Meditation of the Day for Calm’ enabled. Do you want to open it?”
Let’s stop for a second and talk about how Alexa used explicit confirmation. She was <65% confident that the intent was Jimmy wanted to open ‘Guided Meditation’ skill, which he has previously enabled. In this prompt, Alexa picked up on the word ‘Meditative’ and asked if Jimmy want it to open the skill.
Ok, so Jimmy and Ricky were not impressed. But now it’s Jimmy’s turn, so how did he do?
Jimmy: “Alexa, downward dog is a form of ….?”
Alexa: “Sorry, I don’t know that. But I do have a skill called Bark Like a Dog. Want to try it?”
You can see here that Jimmy is frustrated and he says, ‘no!’
Despite the impatience of Jimmy and Ricky, we have to give praise to Alexa. She recognized it as a ‘task failure’, which that means there were no data or search results that could be found because of too many search criteria and user restrictions. In this case, Alexa added an alternative suggestion ‘I do have a skill called Bark Like a Dog. Want to try it?’ But it didn’t help, because it wasn’t the response Jimmy and Ricky wanted ‘Yoga’, which led to even more user frustration.
Alright, back to Ricky.
Ricky: “What exercise rhymes with ‘boga’?”
Alexa: “Sorry, I’m not sure.”
This was an interesting ‘no match’ response. It’s a very obvious question where a human can easily think of the word ‘Yoga’ which rhymes with ‘Boga’. I tested the exact same question on my Alexa Device and got the same no match response. If we break the question down, Alexa first needs to know the names of all exercises and then sort them to detect assonance or words that rhyme with other words.
Maybe the conversation design and dev team at Amazon Alexa can add this to the algorithm?
It’s Jimmy’s turn once again, but will Alexa fail once again?
Jimmy: “Alexa , what is a low-impact exercise?”
Alexa: “Low is from Duluth, Minnesota, and it formed in Duluth, Minnesota, United States.”
This response was a ‘misrecognition’ response, Jimmy used the word ‘low-impact’ which I believe is a word In-Vocabulary (IV) because I asked my device the same question and Alexa was able to give me a correct response to that question. In this case, however, the environment factor made Alexa recognize it as something else ‘Low is from Duluth…’.
At this point, you can see the frustration levels increasing in their facial expressions. The next one luckily provides a little comical relief…
Ricky: “Alexa, Buddhist exercise.”
Alexa: “Exercise added to your shopping list.”
This was clearly hilarious to watch and the context made it even funnier. Alexa’s response was an Implicit Confirmation where it was >64% sure that Ricky wanted to add Exercise to his shopping list. Fail!
They both cackle at the response and decide to give it another try.
Jimmy: “Alexa, exercise that involves meditation.”
Alexa: “You have ‘Guided Meditation — Meditation of the Day for Calm’ enabled. Do you want to open it?
Jimmy: “NO!”
Again, Alexa makes the same mistake, going for an Explicit Confirmation, thinking Jimmy wants to open the Guided Meditation skill.
One last turn to trick the robot.
Ricky: “Who is smarter than the average bear?”
Alexa: “According to another Amazon customer, smarter than the average bear — all it really means is that the person being talked about is smarter than average. But you may be asking where the bear came from. That would be Yogi Bear, a vintage cartoon character from the 20th century. Did that answer your question?
At this point, Ricky tries something different and asks Alexa a question. He tried to trick it to just say the word ‘yoga’ or in this case ‘yogi’, which Alexa did and won Ricky points for thinking outside of the box!
Jimmy and Ricky got excited, Jimmy says, “Yes! That counts, that counts!” Jimmy claps with laughter. (But really, I think they just wanted to be done with this word.)
Round 2: The word is ‘turtleneck.’ This goes surprisingly MUCH better than the first round.
Jimmy: “Alexa, what shirt would Steve Jobs wear?”
Alexa : “Here’s something I found on the web. According to Forbes.com Steve Jobs, on the other hand, was known for his iconic black long-sleeve mock turtleneck and jeans.”
Ohhh, nailed it!
After all the wrong responses, Alexa got it on the first try!
Let’s break it down here, the question has the entities ‘shirt’ = ‘Clothe’ and ‘Steve Jobs’ = ‘Celebrity Person’ and search the web and came back with a response that included the word ‘Turtleneck’.
Lucky or not, I think Alexa’s ability to fallback on a quick web search for entities that are not in-vocabulary is a well-thought-out design choice. In this case, Jimmy was a happy user!
Round 3: Last round and the word is ‘wagon.’ Who will be the winner?
Ricky: “Vehicle, a stagecoach is an example of this vehicle.”
Alexa: “The Chevy convertible driven by Pink is a random vehicle.”
Yeah, not close at all.
Alexa once again misrecognizes the user’s utterance in this implicit confirmation response. If the conversation continued, a well-designed response would have apologized for the misrecognition. Since it used an implicit confirmation instead of an explicit conformation, where Alexa confirms that ‘Are you asking for examples of stagecoach vehicles?’.
Apparently, Alexa is the queen of obscure answers, which makes Jimmy and Ricky laugh at that one. So how will Jimmy fare?
Jimmy: “Alexa, what are Darius Rucker’s hits?”
Alexa: “Shuffling songs by Darius Rucker on Amazon Music.”
We can all guess he was aiming for the hit song ‘Wagon Wheel’ but Alexa didn’t quite catch that.
Back over to Ricky.
Ricky: “What is a dog doing when it’s moving its tail back and forth?”
Alexa: “Hmm. I’m not sure.” (Honestly, same.)
At this point, the two comedians are just done trying to trick the robot. But just for fun, Ricky asks one more question.
Ricky: “Alexa, a hen is a female bird. What is a male bird called?”
Alexa: “This might answer your question. A male Aves is called a cock.”
Ok, so that definitely wasn’t the word ‘wagon’ but judging by Ricky’s reaction, Alexa did respond with the new word he intended.
So, why did Ricky decide to deviate from the game and make up his own word? He was probably so irritated with Alexa’s conversation design. When humans don’t like where the conversation is going, we are likely to change the subject. And that’s exactly what Ricky did. He interacted with the machine as if it were a human!
This just goes to show how important it is to have a well-designed conversation in order for a human to feel comfortable and at ease with the machine — whether it’s an Alexa or a Messenger chatbot. We know when something is human and when it is a machine, so the ultimate goal is to make that conversation feel as natural as possible.
At the end of the day, a happy user is what will make or break your success.
Designing an effective and cohesive conversation isn’t simply throwing things together in the hopes that it works. Check out Conversation Academy’s Conversation Designer certification and see how you can improve your user-bot conversations that translate to better experience for your users!
Have a voicebot or chatbot you think could use some upgrades? Hop on a discovery call with me, we’ll take a look at your bot to see what’s under the hood and talk about how to improve and land your more conversions.
