flowbotopresentation
TRANSCRIPT
D O M O A R I G AT O , M R . F L O W B O T O
Produced by Wes Berry
Imagine a rap battle between Jay-Z and a robot…A robot whose mic skills are indistinguishable from those of a human (i.e. one that passes the Turing Test)?Requires creativityWe can look at machine creativity by looking at necessary processes for a creative, rapping robot—Flowboto
W H AT I S C R E AT I V I T Y ?
Creating novel concepts and pieces?Is it something that springs randomly and suddenly or pours from séance-like concentration?
What about animals?
Some evidence of creative selection and processes in chimps, Vogelkop Gardener bowerbirds, and dogs
Robots?
Can robots be creative?Take potential rapping robot, FlowbotoLet’s examine a recipe for Flowboto…
Replicating Speech Patterns, Grammar, and Stressed Syllables, while realizing connections between words
• Siri, Watson, GPS navigational devices
Step 1
• Strategy a la NEIL• Mix in some Google NGrams
Robust programs exist that do many of these thingsThe Never Ending Image Learner (NEIL) + Google NgramsLovechild of Siri and WatsonStill boring…
Key Breakthroughs!• Google Translate can handle poetry and produce
rhymes in appropriate places using brute force search
• 2001: study concludes that speaking robots can mimic emotion well enough to be perceived by humans
• 2002: study suggests humans can recognize general communicative intent in robot-directed speech
What does it mean?
This means Flowboto can ‘emote’ in ways understandable to humans, critical to being a believable musician
Step 3:
Tempo, pitch, key all digitally accessible (pitch-shifters, beat-matchers, digital tuners)But, there are also complex, underlying elements to songs that human artists understand—part of creativityUnderstanding of these is critical to Flowboto as an artist
Human-Driven Approaches• Pandora’s Musical Genome Project• Crowd-Sourced Social Tagging Data
Two solutions:Both human-basedSimilar drawbacks to both, but existences demonstrate potential and precedence for converting musical intricacies into data for a computer
Step 4:Subject Matter
Flowboto needs subject matter—parse the web!Social trends, pop culture, and eventsCan create a web of the world
Step 5:Unique Sound Combinations and
Appealing New Music• Might use understandings of poetic forms,
musicality, and databases to recognize popular musical styles (what gets the people going?)
• Wide, data-parsing algorithms to produce various probable renditions of novel songs or sections, graded on likability scale
Finally…It can do the tasks, but it still needs to create!By…
FlowBot produces probable renditions of novel music
Unique Sound Combinations and Appealing New Music
Human producers select, sample, and mix the renditions that sound best
Release the hit!(And credit Flowbot to avoid a lawsuit)
Human producers as filters, differentiating good and bad song/section renditions
If Flowboto also incorporates learning algorithms, it can make better musical decisions going forward, creating and self-selecting more appropriate renditions in the future based on past human-choices
Interesting Implication…
Then…
Learning algorithms would allow Flowboto to self-filter more effectively in the future…
Flowboto is almost a real boy!(er…artist)
• Improving as an artist
• Similar to human artist, whose work is critiqued and improved through collaborative efforts with other artists
And we begin to see similarities to a human artist learning through collaboration
Speech Patterns
RhymingMusicality
Databases
Novel Sounds
FLOWBOTO
Existent, but dispersed technologies, themselves precedented or conceivable from past projects, allow for innovationLaying the building blocks for an artificially-creative musical agent like Flowboto
What does this mean for machine creativity?
Then…
Details different depending on the tasks, but the process translates well across creative formsMixing, reimagining, and improving the available tools, while using human and animal creative processes as models, may unlock machine creativity sooner rather than later
References
Then…
• Breazeal, C. (2001). Emotive qualities in robot speech. Proceedings from the International Conference on Intelligent Robots and Systems. Maui, HI: IEEE.
• Breazeal, C., & Aryananda, L. (2002). Recognition of Affective Communicative Intent in Robot-Directed Speech. Autonomous Robots, 12(1), 83-104.
• NPR Staff. (2011). Google’s artificial intelligence translates poetry. NPR. Retrieved from http://www.npr.org/2011/01/16/132959095/googles-artificial-intelligence-translates-poetry.
• Pandora. (n.d.). About the Music Genome Project. Retrieved from http://www.pandora.com/about/mgp.
• Saunders, R. (2002, February). Curious design agents and artificial creativity (Doctoral dissertation). Department of Architectural and Design Science, Faculty of Architecture, University of Sydney.
• Schmidhuber, J. (2006). Developmental robotics, optimal artificial curiosity, creativity, music, and the fine arts. Connection Science, 18(2), 173-187.
• West, K. (2008, September). Novel techniques for audio music classification and search (Doctoral dissertation). School of Computing Sciences, University of East Anglia.
• Goldman, J. G. (2014). Creativity: The weird and wonderful art of animals. BBC. Retrieved from http://www.bbc.com/future/story/20140723-are-we-the-only-creative-species.