Li e Code in he Loop: Pe o ming wi h an
Au onomous Agen
Elizabe h Wilson
(pe o ming/c ea ing as "digi al sel es")
C ea i e Compu ing Ins i u e
Uni e si y o he A s London
London, UK
Abs ac
Li e Code in he Loop: Pe o ming wi h an Au onomous Agen in TidalCycles
is a p oposed pe o mance whe ein collabo a i e musical in e ac ion is explo ed
wi h an agen sys em in he li e coding language TidalCycles. The pe o mance
will demons a e a hyb id app oach whe e he human can accep and ejec sugges-
ions om an au onomous agen , and he li e code has explici con ol o e he
sugges ions om he agen . The agen can gene a e pa e ns ha a e modi iable
based on changing a ec i e s a es o model human emo ion. Speci ically, his
pe o mance in e oga es he p ac ical and aes he ic dimensions o co-c ea ing wi h
a code-gene a ing AI agen , embodying he pape ’s call o sys ems ha p io i ise
e lec ion, sha ed agency, and li eness.
1 In oduc ion
This p oposed pe o mance p esen s Tidal-Me zA—a no el sys em designed o collabo a i e pe o -
mances be ween a human li e code and a machine agen in he con ex o li e coding, speci ically
ocusing on he gene a ion o musical pa e ns. By in eg a ing a ec i e modelling wi h compu a ional
gene a ion, his sys em le e ages ein o cemen lea ning echniques o dynamically adap music
composi ion pa ame e s wi hin he TidalCycles amewo k, ensu ing bo h a ec i e quali ies o he
pa e ns and syn ac ical co ec ness.
Li e coding is a e m used o e e o pe o me s c ea ing a by w i ing compu e code, usually in
on o an audience (Collins e al., 2003). In li e coding, compu e language is he p ima y medium
o no a ion and desc ibing he ules wi h which o syn hesise a wo ks, in his case we conside he
case whe e he ou pu is musical pa e n. The p ac ice o li e coding places a s ong ocus on he
elemen s o li eness, emb acing e o , he use o andom p ocesses and clea mappings be ween
syn ax and ou pu .
The TidalCycles li e coding language is used o c ea e au onomous pa e ns by Tidal-Me zA. TidalCy-
cles is an exp essi e language, known o i s lexibili y and e sa ili y in c ea ing complex s uc u al
ideas, h ough i s unc ional p og amming s yle and "mini-no a ion" syn ax (McLean and Wiggins,
2010). In Tidal-Me zA, hese "mini-no a ion" s ings—symbolic g oupings o deno e wide unc ions
in TidalCycles a e a c ucial aspec o ep esen ing and gene a ing new pa e ns .
Fo a p e ious i e a ion o his pe o mance, pe o med a S Ma y’s Chu ch, B igh on:
h ps:
//www.you ube.com/li e/d0RMUqcbhmQ?si=n8IaWeZ8QcKpOFbm& =8036
An audio composi ion ex ac om he pe o mance can be ound he e
h ps://on.soundcloud.
com/Qa 93RYJoUYJg7s57.
P oceedings o he 6 h Con e ence on AI Music C ea i i y (AIMC 2025),
B ussels, Belgium, Sep embe 10 h-12 h
Figu e 1: Li e pe o mance wi h he Tidal-Me zA au onomous agen sys em desc ibed. Image c edi :
Jona han Reus
2 Me hods Taken in De eloping The Wo k
The de elopmen o Tidal-Me zA in ol ed a mul idisciplina y app oach, blending compu a ional
c ea i i y, a ec i e compu ing, and li e coding p ac ices. The sys em was buil upon he TidalCycles
pla o m, a domain-speci ic language o li e coding o music, which p o ided a lexible and
exp essi e en i onmen o musical pa e n gene a ion.
The machine agen componen was ained using ein o cemen lea ning, wi h he goal o op imising
musical pa ame e s based on a ec i e goals (such as alence and a ousal le els). To achie e his,
a ewa d unc ion was de ined ha e alua es gene a ed musical pa e ns acco ding o a ec i e
es ima ions de i ed om p e- ained emo ion ecogni ion models, while simul aneously main aining
musicali y and syn ax alidi y. The aining p ocess included i e a i e in e ac ions wi h human
code s in simula ed li e sessions, whe e eedback was collec ed bo h explici ly ( ia anno a ions) and
implici ly ( h ough musical choices and iming).
Fo his hyb id sys em, wo agen s a e consecu i ely buil o a emp o cap u e all he dimensions
ou lined in he a ec i e model-ALCAA. An o e iew o how hese a e implemen ed a e ou lined in
igu e 2.
Fi s , ein o cemen lea ning (RL) echniques a e used o dynamically adap he pa ame e s gene a ed
by he a ec i e model wi hin he lexible amewo k o TidalCycles. This amewo k enables
he model o ha ness TidalCycles’ ex ensi e lib a y o unc ions and pa e ns o gene a e music
composi ions ha no only encapsula e desi ed emo ional a ibu es bu also adhe e o he syn ac ical
co ec ness o TidalCycles code. Secondly, speci ic mini-no a ion s ings a e p oduced ha ha ness
TidalCycles in e nal pa sing o sho -hand e en s.
In Me zA, he RL agen ’s ac ions co espond o he selec ion o musical elemen s wi hin he Tidal-
Cycles amewo k. These ac ions a e guided by a ewa d mechanism ha e alua es he alignmen
be ween he gene a ed music’s a ec i e a ibu es and he a ge a ec i e s a es de ined by he
ALCAA model. Th ough ial and e o , he agen e ines i s decision-making s a egies, g adually
lea ning which musical elemen s and TidalCycles unc ions o employ in o de o e oke speci ic
emo ional esponses. RL is pa icula ly well-sui ed o his ask because i allows he agen o
2
lea n and adap in a dynamic en i onmen and op imise i s ac ions. In pa icula , modelling equa-
ions o di e en musical s uc u al pa ame e s we e de ined, namely: hy hmic s uc u e, sound
le el/pe cep ual loudness, and empo, modali y, pi ch egis e and pi ch con ou . Th ough he c ea ion
o wo agen s, hese pa ame e s a e inco po a ed, p ese ing he o iginal equa ions in his new mode
o gene a ion.
In his p oposed pe o mance, he li e code will be engaging in a cons uc i e dialogue wi h he
Tidal-Me zA sys em, shaping musical s uc u e in eal ime alongside he machine agen . The li e
code s ole will in ol e cu a ing and esponding o he agen ’s ou pu s, s ee ing he di ec ion o he
composi ion h ough code, and in luencing he a ec i e ajec o y o he pe o mance. Ra he han
unc ioning as a ixed con olle , he li e code will be ope a ing in a co-c ea i e capaci y—adap ing
my coding s a egies in esponse o he machine’s decisions and emo ional cues.
Figu e 2: Li e pe o mance wi h he Tidal-Me zA au onomous agen sys em desc ibed
3 Technical Requi emen s
3.1 Technical Requi emen s om he Venue
3.1.1 Sound an Technical
1. PA Sys em:
• 2 o mo e coaxial speake s, wi h equency ange minimum 50- 17000Hz
• 1 o mo e subwoo e s
• Capable o s e eo sound ep oduc ion
3.1.2 Ligh ing and Visuals
1. P ojec o
2. P ojec ion Sc een
3.1.3 Cables
1.
HDMI cable (>10m, o al e na i ely his mus be long enough o connec hdmi ou pu om
he a is ’s lap op o he p ojec o )
2. 2x XLR cables o connec om PA o A is ’s Mixe o In e ace
3
3.2 A is Will b ing
1. Lap op
2. Audio in e ace (> 4 inpu , >4 ou pu )
3. 8 Channel Digi al Audio Mixe
4. Any addi ional ha dwa e syn hesise s o be used (1x Ko g Volca FM, 1x Moog We ks a )
5. USB-c o HDMI and USB adap e
6. Addi ional 3.5mm o phono cable
7. Addi ional Jack cables
3.3 Addi ional Reques s
• Timing should be alloca ed o load-in/soundcheck
• O ganise s should aim o ensu e di e si y in pe o me s.
Acknowledgmen s and Disclosu e o Funding
This wo k was suppo ed by EPSRC and AHRC unde he EP/L01632X/1 (Cen e o Doc o al
T aining in Media and A s Technology) g an a
Re e ences
Collins, N., McLean, A., Roh hube , J., and Wa d, A. (2003). Li e coding in lap op pe o mance.
O ganised sound, 8(3):321–330.
McLean, A. and Wiggins, G. (2010). Tidal–pa e n language o he li e coding o music. In
P oceedings o he 7 h sound and music compu ing con e ence, pages 331–334.
4
