RICK AND MORTY
Using AI to determine which Beth is the clone
Training a classifier and visualising personalities
SPOILER ALERT: If you are not up to date with the final episode of season 4 of Rick and Morty you should avoid reading on.
Ever since the end of season 3 of Rick and Morty, fans have wondered whether or not Rick had cloned his daughter Beth and if so, if the Beth that remained with the family was real or the clone.
Lucky for us fans, part of this mystery was put to bed in the season 4 finale (Star Mort Rickturn of the Jerri), where it was revealed that Rick had indeed cloned Beth, made it impossible for anyone to know which was the clone and then wiped his memory of ever doing so.
So whilst one mystery was put to rest, the other remained. And where there is a hint of impossibility, the data scientists will be. And so, the problem:
Which Beth is the clone and can AI help solve the mystery?
From here on I will be referring to Beth from seasons 1–3 as “Real Beth”, the Beth that is fighting the Galactic Federation in space as “Space Beth” and the Beth that remains with her family as “Family Beth”.
The process I used to solve this problem is relatively simple but tackles this question in two different angles.
The Process:
- Get all the lines of dialogue of each of the three Beths
- Run their lines through a personality insights algorithm (IBM’s Watson)
- Visualise their personalities and draw conclusions
- Build a classifier trained on Family Beth and Space Beth
- Predict which one Real Beth is most likely to be
The above process can all be found on Github here. For the highlights, see below.
Visualising their Personalities
Lucky for me, IBM have created a much smarted algorithm for personality insights than I could. They call it Watson and it has a service called Personality Insights that is accessible via API.
Once having built a Watson service, I gave it all the dialogue of each of the three Beths. It outputs a dictionary that contains several characteristic traits and which percentile the person being tested ranks in. By taking the percentile for each trait for each character, I was able to visualise their personalities side-by-side.
See below how IBM’s Watson classified their personalities:
Above is the output of Watson’s “Big 5” personality traits for each of the three Beths. Amazingly enough, we do see a consistency in the fact that Real Beth and Space Beth have similar levels of each trait and that Family Beth does not. Family Beth is in very different percentiles of Conscientiousness, Extraversion and Emotional Range. She also has observably more Agreeableness.
The same can be noted when we look into the “needs” of each character according to Watson. See below:
Similarly to what we saw in the Big 5 traits, Real Beth and Space Beth often are in similar percentiles whereas Family Beth often is way off. We can see that Real and Space Beths need more curiosity, harmony, practicality, stability and structure than Family Beth. They also need much less love than Family Beth.
Given that the personality of Space Beth is significantly more similar to that of Real Beth, we can make an initial conclusion that Space Beth is more like Real Beth and hence Family Beth is the clone.
Building a classifier to get probabilities
I wanted to validate the above conclusion by building my own classifier that would determine which Beth is most like Real Beth.
To do so, I extracted features from all the lines of dialogue of all three Beths with TfidfVectorizer. This allowed me to input all of their lines into a LogisticRegression model to predict which Beth is most to likely to have said that line.
The model was trained on all the lines of dialogue each Beth said in the season 4 finale, with the target value being the name of which character said such line (family_beth or space_beth).
With features extracted and my model trained I could generate predictions of which Beth the Real Beth from season 1–3 is most likely to be. I input the extracted features from Real Beth’s lines of dialogue into my model to get probabilities of which Beth said each line.
By taking the average probability for each of the Beths, we see that Real Beth is:
- 39.8% likely to be Family Beth
- 60.1% likely to be Space Beth
My classifier predicted that Real Beth is 20.3 percentage points more likely to be Space Beth than Family Beth. And thus Family Beth is most likely to be the clone. A conclusion that matches that observed in the personality insights above.
Conclusion
To help solve an unknown mystery, I harnessed the power of AI to see the reality that was never revealed to us. I took two separate approaches that gave one uniform answer: Family Beth is the clone because Space Beth is considerably more similar to Real Beth.
This is not a trivial conclusion. Anyone that watched the episode would note how Space Beth seemed to have developed as a character much more than Family Beth had. Her experiences since season 3 had been so dramatically different to that of Family Beth that there was a complete transformation of her character. She dressed differently, acted differently and was different.
Yet even with all those different experiences, two separate algorithms found her to be more similar to Real Beth. That whilst the facade on the outside may have seemed to be wildly different, deep down she truly hadn’t skewed far from her season 3 self. Proving that regardless of how different her journey was during the events of season 4, she is still more “real” than Family Beth.
And so there you have it. Space Beth is real and Family Beth is the clone.
