Dartmouth College Neukom Institute Turing Tests in Creativity

 
 
 

The Neukom Institute for Computational Science at Dartmouth College is pleased to announce the first annual Neukom Institute Prizes in Computational Arts. These competitions aim to inspire innovations in computational methods that generate artistic products, such as literary, musical, and visual art.

In the 2015-2016 academic year, the Neukom Institute will run three different competitions:

DigiLit

PoetiX

AlgoRythms

The Neukom Institute for Computational Science, in collaboration with the Digital Music and Sonic Arts programs at Dartmouth, is pleased to announce its first competition for live machine DJing. Submissions will be in the form of executable computer code. The task will be autonomous performance of a 15-minute DJ set, only using songs from a prior unseen music collection, its accompanying metadata, acoustic features, and a starting seed track. The competition will consist of a dance party, anticipated to be in the Hopkins Center for the Arts, Saturday, April 23rd, 2016, at the close of the Digital Arts eXposition. [Venue and precise date and time, still to be confirmed.]

With services such as Spotify, Pandora, Apple Genius, Google Music Mix, and many others, we have become accustomed to algorithms mediating our enjoyment of music and other media. Can we tell the difference between human and machine music recommendation? Can a machine control a dance floor like a DJ ? The goal of the competition is to find out whether machines can be preferred to humans in musical performance.

Dance Party DJ Turing Test

Rules

  1. Entries must be submitted via the competition Web site https://math.dartmouth.edu/~turingtests/AlgoRythms.php 11:59PM UTC on Friday, April 8th, 2016
  2. Prior to the competition, (later in April, 2016, precise date TBA), up to six machine DJ finalists will be selected by a panel of experts using the following criteria:
    • Correct and self-contained execution within Apple OSX 10.10, Ubuntu Linux 14.04, or Microsoft Windows 8.1 or 10.x. All libraries must be statically linked, and an executable “RunMe” provided. Systems must run with a maximum of 4 x 2.2GHz 64-bit Intel processors and 16GB RAM.
    • Quality of technical execution: cross-fades, tempos (beat-alignment), and musical styles, create smooth transitions between moments in the set. There should be no gaps or pauses, other than those used for musical effect.
    • Creativity and originality: how well the source audio is selected and presented to create an original experience of the music.
    • Ability to respond to a prompt: a song will be given to each machine and human DJ to begin the set. Different seed songs sould result in a different set list. The same seed song might produce the same set list.
    • At the discretion of the panel, it is possible that no finalists will emerge
  3. Finalists will take part in the Algorhythms Dance Party (late April, 2016, precise date and place TBA). The format of the competition will be as follows:
    • All DJs will perform 15-minute sets, each allocated to one of two dance floors.
    • Human and machine DJs will be randomly assigned to the two dance floors. At any one time the music is either selected and spun by a machine DJ or by a human DJ.
    • The order of Djs will be determined by random assignment.
    • The competition last one to one-and-a-half hours, with back-to-back DJ sets in two rooms with a DJ hidden in each booth.
    • A maximum of six machine DJs and six human Djs will participate.
  4. Source material: human and machine DJs will be limited to selecting music from a list of 1000 tracks to be released just prior to the competition. Competition organizers will make a training data set available—consisting of 10,000 open-source audio tracks and accompanying audio features—to developers prior to the competition. At competition time, the competition audio collection will be made available for both Human and Machine Djs. The data will consist of:
    • Training: A dataset consisting of publicly available metadata and audio clips will be available for algorithm training by October 1st, 2015.
    • Competition Data will be held back until 2 hours prior to start of the competition for machine DJs and 24 hours prior for human DJs.
  5. Audience Evaluation:
    • The Audience are aware that some of the Djs are machines, although they will have no way of knowing which are machine and which are human DJs.
    • After each set, the audience will be prompted, via mobile phone, or polling card, to choose one of the following categories: human DJ and machine DJ.
    • Results will be reported as a comparison of the voting for each category: human DJ and machine DJ.

Prizes

A first prize of $3000 will be awarded to the team that enters the “best” DJ set generation as determined by achieving an audience-vote distribution indistinguishable from that of human Djs (independent t-test, p>=0.05). At most one prize will be awarded to any competitor. Two second prizes of $1000 will be given to the runner-ups, determined by closeness to human distribution.

Other Resources

Contestants might be interested in the lessons and resources gathered by the Music Information Retrieval Experiment Exchange (MIREX) User Experience Grand Challenge

The following musical data sources may also be useful:

Final Caveats: All competition details are subject to change, due to venue availability, etc.
While we anticipate smooth sailing, one never knows!

 

Judges: Will be announced in September - stay tuned!

Yes! I want to keep apprised of Turing Test developments!
Add me to the Turing Tests competition email list!

Enter your email address:


The Turing Tests in Creativity are a project of The Neukom Institute for Computational Science at Dartmouth College.
For more information contact turingtests@math.dartmouth.edu.