Week 10 (5/8) – Final Thoughts and Results

Edward W -

This week’s goal: Gathering and analyzing the data I collected from my survey.

Hello everyone! In this penultimate post, I’ll be revealing the answers to my survey and detailing the final results from my survey. But before that, I wanted to take the time to thank everyone who took the survey. I am honestly beyond grateful for your input despite how long my survey ended up being, and I could never have made it this far without all of your help! Thank you again for your time and patience!

Without further ado…

Here are the survey results!

In total, I got 62 responses for my survey! The average score was 3.323 / 10 questions being correctly identified as AI, with the lowest score being 0 / 10, and the highest score being 7 / 10. Typically, the score that a person got on the survey varied by approximately 1.352 from the mean score of 3.323. The 25% percentile score was 2, the median score was 3, and the 75% percentile score was 4.

Below, I’ll show the results for each question along with the answers.

Key

0+     5+     10+
Number of years of musical experience that a surveyed group has per person
50.00%     2.50
Lowest percentage/emotional rating
50.00%
AI percentage (correct answer)
2.50
Highest emotional rating
0+     5+     10+
Sample group rates AI as most human-like of the three
Sample group rates AI as least emotional of the three
0+     5+     10+
Sample group rates AI correctly fairly consistently
Sample groups rates AI as most emotional of the three

 

1. Baroque – solo organ

Original Piece: Bach, Johann Sebastian – Fuge in c-moll über ein Thema von Legrenzi, BWV 574

Main Audio Clip
Scores
0+
5+
10+
Original Ending
27.42%
35.00%
50.00%
AI Ending
32.26%
25.00%
25.00%
My Ending
40.32%
40.00%
25.00%

Happy
Sad
0+
5+
10+
0+
5+
10+
Main Audio Clip
2.145
2.70
2.67
2.032
2.10
2.08
Original Ending
2.355
2.95
3.00
1.968
2.15
2.25
AI Ending
1.855
2.45
2.33
2.387
2.35
2.42
My Ending
2.258
2.75
2.75
1.839
2.15
2.17

 

2. Classical – string quartet (2 violins, 1 viola, 1 cello)

Original Piece: Mozart, Wolfgang Amadeus – String Quartet No. 10 in C-major, K.170

Main Audio Clip
Scores
0+
5+
10+
Original Ending
30.65%
40.00%
33.33%
AI Ending
38.71%
35.00%
33.33%
My Ending
30.65%
25.00%
33.33%

Happy
Sad
0+
5+
10+
0+
5+
10+
Main Audio Clip
3.000
3.45
3.58
1.274
1.55
1.50
Original Ending
2.774
3.15
3.17
1.403
1.55
1.58
AI Ending
2.839
3.30
3.33
1.355
1.55
1.58
My Ending
2.710
3.15
3.00
1.419
1.60
1.58

 

3. Romantic – solo piano

Original Piece: Liszt, Franz – Le rossignol (1st version), S. 249d

Main Audio Clip
Scores
0+
5+
10+
Original Ending
27.42%
30.00%
41.67%
AI Ending
20.97%
30.00%
33.33%
My Ending
51.61%
40.00%
35.00%

Happy
Sad
0+
5+
10+
0+
5+
10+
Main Audio Clip
3.435
3.70
3.92
1.258
1.35
1.25
Original Ending
3.581
3.80
3.83
1.371
1.55
1.58
AI Ending
3.355
3.55
3.75
1.290
1.50
1.50
My Ending
3.194
3.45
3.58
1.452
1.65
1.67

 

4. Contemporary – solo piano

Original Piece: Prokofiev, Sergei – XIV. Feroce from Visions fugitives, Op. 22

Main Audio Clip
Scores
0+
5+
10+
Original Ending
30.65%
30.00%
25.00%
AI Ending
33.87%
35.00%
33.33%
My Ending
35.48%
35.00%
41.67%

Happy
Sad
0+
5+
10+
0+
5+
10+
Main Audio Clip
2.516
2.85
3.17
1.742
2.15
2.08
Original Ending
2.774
3.30
3.50
1.597
1.80
1.75
AI Ending
2.484
2.75
3.00
1.677
2.15
2.17
My Ending
2.210
2.45
2.58
1.790
2.05
2.00

 

5. Contemporary – solo cello

Original Piece: Kodály, Zoltán – Sonata for Solo Cello in b-minor, Op. 8

Main Audio Clip
Scores
0+
5+
10+
Original Ending
24.19%
35.00%
33.33%
AI Ending
41.94%
25.00%
33.33%
My Ending
33.87%
40.00%
33.33%

Happy
Sad
0+
5+
10+
0+
5+
10+
Main Audio Clip
2.113
2.30
2.25
2.097
2.45
2.75
Original Ending
1.952
2.20
1.83
2.435
2.90
3.33
AI Ending
1.774
1.90
1.67
2.597
3.05
3.42
My Ending
2.032
1.95
1.83
2.048
2.45
2.83

 

6. Jazz – jazz quartet (alto sax, double bass, piano, drumset)

Original Song: Layton, Turner – After You’ve Gone
Piano: Emmet Cohen
Trumpet: Bruce Harris
Alto Saxophone: Patrick Bartley
Double Bass: Russell Hall
Drumset: Joe Saylor

Main Audio Clip
Scores
0+
5+
10+
Original Ending
30.65%
15.00%
8.33%
AI Ending
45.16%
65.00%
83.33%
My Ending
24.19%
20.00%
8.33%

Happy
Sad
0+
5+
10+
0+
5+
10+
Main Audio Clip
3.532
3.85
4.25
1.210
1.30
1.08
Original Ending
3.371
3.80
4.08
1.242
1.40
1.25
AI Ending
3.403
3.60
3.83
1.210
1.30
1.17
My Ending
3.532
3.80
4.08
1.129
1.25
1.08

 

7. World – alto (clarinet), tenor (bassoon), lyre, frame drum

Original Song: Seikilos – Epitaph to Euterpe
Tenor + Lyre: Yerko Lorca
Alto + Frame Drum: Kuan Yin

Link to Song: https://www.youtube.com/watch?v=qdlFLw5Asc8&t=46

Main Audio Clip
Scores
0+
5+
10+
Original Ending
30.65%
40.00%
33.33%
AI Ending
33.87%
30.00%
25.00%
My Ending
35.48%
30.00%
41.67%

Happy
Sad
0+
5+
10+
0+
5+
10+
Main Audio Clip
2.871
3.35
3.33
1.565
1.50
1.33
Original Ending
2.903
3.25
3.25
1.484
1.45
1.33
AI Ending
2.839
3.30
3.42
1.613
1.55
1.42
My Ending
2.694
3.15
3.17
1.597
1.60
1.50

 

8. Pop – alto (clarinet), bass guitar, trumpets, synth, drumset

Original Song: 藍心湄 – 走路的女孩 from 我要你變心
Lan, Pauline – Walkman Girl from Change Your Heart

Main Audio Clip
Scores
0+
5+
10+
Original Ending
41.94%
55.00%
66.67%
AI Ending
24.19%
20.00%
8.33%
My Ending
33.87%
25.00%
25.00%

Happy
Sad
0+
5+
10+
0+
5+
10+
Main Audio Clip
3.306
3.40
3.50
1.306
1.40
1.33
Original Ending
3.145
3.20
3.08
1.274
1.40
1.42
AI Ending
3.258
3.30
3.25
1.226
1.45
1.50
My Ending
3.290
3.30
3.00
1.290
1.55
1.58

 

9. Ambient – solo piano

Original Song: not a cat. – pov: you lost your hardcore minecraft world

Main Audio Clip
Scores
0+
5+
10+
Original Ending
35.48%
35.00%
33.33%
AI Ending
37.10%
35.00%
41.67%
My Ending
27.42%
30.00%
25.00%

Happy
Sad
0+
5+
10+
0+
5+
10+
Main Audio Clip
1.887
2.20
1.92
2.323
2.35
2.42
Original Ending
1.903
2.30
2.08
2.452
2.35
2.33
AI Ending
2.129
2.35
2.33
2.145
2.20
2.08
My Ending
2.903
2.25
2.08
2.355
2.25
2.25

 

10. Personal Composition – cello and piano duet

Original Piece: W., Edward – Lac de la Nuit from Eight Duets for Cello and Piano, Op. 4, No. 4

Main Audio Clip
Scores
0+
5+
10+
Original Ending
40.32%
40.00%
50.00%
AI Ending
24.19%
30.00%
16.67%
My Ending
35.48%
30.00%
33.33%

Happy
Sad
0+
5+
10+
0+
5+
10+
Main Audio Clip
2.565
2.85
2.75
2.032
2.05
2.17
Original Ending
2.597
3.00
2.92
2.000
2.10
2.25
AI Ending
2.516
2.95
2.92
2.048
2.20
2.42
My Ending
2.452
2.05
2.75
2.065
2.15
2.33

 

Important trends to notice

  1. The more musical experience a person has, the more likely they are to rate a piece as more emotional than people with less musical experience. I’m not completely sure why this is, but it might be genuinely because one’s experience with music makes them able to relate to and more deeply comprehend and process the music that they are listening to.
  2. Typically, the music created by the original composer had higher happiness ratings than other pieces. This just feels like a testament to the greats of music; even if we can’t tell which songs are AI and which ones aren’t empirically, we can still feel that the original makes us more emotional.
  3. Typically, the music created by AI and me had higher sadness ratings than other pieces. This was honestly quite surprising to me, since not creating a particularly happy ending doesn’t correlate to a particularly sad one. For some reason, AI just sounds sadder to us, and it’s the same with much compositions as well! (I don’t know whether to be proud or hurt 😂😭)

Some answers to my research questions

  1. Can people tell the difference between AI music and manmade music? – Generally, not really. The only pieces that people were able to find the AI fairly well were for contemporary cello and jazz. These pieces were special, because the AI stood out through its compositions in these pieces. For contemporary cello, the AI was repeating notes without any sort of structure, which helped it to get identified. For jazz, the AI kept following the chord notes for its sax solo. It played it safe in the realm of compositional value, but that ended up backfiring for it in the realm of AI identification.
  2. Does increased knowledge of music theory facilitate the identification of AI music? – Kind of, kind of not. People with more musical experience were able to identify AI in ambient music decently, but experience also ended up being a hindrance too, especially for contemporary cello. However, the scores that experienced musicians got in jazz were very, very good, and it kind of showed that when the general populace can find AI fairly well, the experienced musicians will get it right almost all of the time. I think that this experience also led a lot of musicians to question themselves when it came to some of these pieces, which led them to do worse than the general populace sometimes.
  3. Does AI music evoke weaker emotional responses in people compared to original compositions? Kind of, kind of not. AI is not completely noticeable worse than manmade music, but it does seem to elicit a slightly lower happiness response in us. With that being said, it does elicit a slightly higher sadness response in us. It shows how the “gut feeling” that many musicians claim they can identify AI music with does exist (kind of), and as AI music becomes quantitatively harder to distinguish from manmade music, it might be this qualitative “gut feeling” that we might have to rely on for identification.

My final product

For my senior project final product, I decided to produce an AI piano piece that I performed at my senior project presentation. For those of you who missed it (or those of you who want to see it again), here is a recording I did of the final piece.

Final thoughts and conclusions

AI is a lot better than we think, and a future where AI composition is on par with human composition is already here. The results of this project show that AI is already pretty good at fooling us, and it demonstrates that in the future, you have to be alert to what you are listening. Especially since many musical AIs have to scrape music off of the Internet in order to train its databases, it is all the more important to make sure that you are supporting human artists now versus these bots who might not credit the composers its music is taken from.

I think AI has pedagogical value, as it can show aspiring musicians how to build a foundation for their pieces, how to formulate melodies and harmonies, and even what not to do in composition. But purely creating music with it might not create very good results yet.

Thank you for reading my blog posts these past ten weeks! The amount of support I’ve gotten has been overwhelming, and I could never have continued working on this project without it. Thank you again for everything!

– Eddie 😛

More Posts

Leave a Reply