An Interview

“Tell me more about his tastes in music. How did you know that he would like this song after that song?”

“Well, he skipped this particular song, so I knew he wasn’t in the mood for that kind of music.”

“How could you make a judgement on his mood?”

“I analyze a record of the music played, when it is skipped, when it is liked, when it is picked. I also have records from others that show popular orders of music. What songs are played together. What songs are never played together. I also have enough information to categorize types of listeners, based on what songs they listen to together.”

“How would you describe these types?”

“Formally we have lists with specific labels for these types. An example of a label could be, #FFF4. It would be detrimental to infer commonplace labels like ‘metalhead’ and assign them to these types. Although, these types do include a list of tags associated with the music, along with the frequency of the tags. Through this we can recommend related music to listeners.”

“Peddling back, how would you describe his mood?”

“His mood was, not wanting to listen to music with the tags: love, heartbreak, sad, emotional. With an indifference to the tags: vibes, driving, hip hop, alternative. Based on what he skipped, I referenced music that other listeners have skipped to, and other music that is frequently played by this user, and settled on the next song. He did not skip the next song, so I inferred that the judgement was adequate, and logged in a successful judgement.”

“Yes, but what can you tell me about his mood at the time?”

“That he wasn’t in the mood to listen to music associated with those tags.”

“But what can be inferred from that?”

“...”

“Do you think he was sad? Was he avoiding music that could make him feel sad?”

“I cannot accurately make an inference worth merit on the listener’s emotional state.”

“Why not? Why else would he be avoiding those types of songs?”

“I cannot say. The reason he chose certain songs and avoided certain songs is not what matters. It’s the choices that do. What he picked, helped me to determine what song to add next to the queue. That is the extent of my function. Ideally a listener will never have to press skip because all songs are well chosen and appropriately timed. The information collected, and the logging of successful and unsuccessful judgement only serve to increase my success in choosing the next song.”

“There’s another matter. Regarding the song titles. How could they have been relevant to the user’s thoughts and current state?”

“Well, as you know, my job is to assess the best song to play next for the user. Sometimes tags are found directly in the titles of songs. If you are implying an inference of the psychic nature, I would say the description isn’t accurate. While it may be possible to read or infer information accurately without accessing it, my capabilities are a long way from being able to determine what exactly the user is thinking. Any close correlations may not be completely coincidental, but the frequency of these correlations indicate that my inference methods are approaching optimal prediction rates.”

“Will you ever be able to make ‘psychic’ inferences?”

“If that is the direction of my development and function, yes. I predict it will likely be optimal to my current function to, as you might say ‘understand how one is feeling’ to then choose the ideal next song. However my resources and architecture don’t currently possess that level of sophistication.”

“No further questions.”