frag­ments — street scale

Frag­ments — urb­an scale

People often find the pro­pos­al of teach­ing an algorithm to accur­ately anti­cip­ate the optim­al ingredi­ents for their din­ner con­des­cend­ing. ‘I am offen­ded by the assump­tion that I don’t have the cook­ing skills to cook with any res­cued ingredi­ent that fits my repertoire!’*

Tak­ing pride in flex­ibly adjust­ing recipes or enjoy­ing the chal­lenge of cook­ing with ad-hoc ingredi­ents is coun­ter­ing the max­im of optim­isa­tion in the kit­chen. The con­sequences of find­ing value in the serendip­ity intro­duced by ad-hoc deliv­er­ies reach bey­ond just giv­ing slack to the engin­eers of the algorithm. They are ena­blers for Machine Teach­ing Commons.

When it comes to dis­trib­ut­ing food, the pos­sib­il­ity of Machine Teach­ing Com­mons is linked to the pos­sib­il­ity of inhab­it­ing the space between big-data-driv­en, spe­cial­ized, highly optim­ised pre­dict­ive mod­els and manu­ally execut­ing ‘if this than that’ protocols.

We need to pop­u­late this space of the Machine Teach­ing Com­mons with data-gen­er­at­ing prac­tices legit­im­ised by com­mon­ers and machine teach­ing setups that com­mon­ers can autonom­ously oper­ate. A space in which the com­plex­ity of the input con­di­tions that drive decision-mak­ing and the cor­rel­a­tion between input con­di­tions and decision out­comes does not evade the grasp of the com­mon­ers. A space where com­mon­ers stay in the cyber­net­ic loop of algorithmic decision-making.

(image: screen­shot of algorithmic food dis­tri­bu­tion mod­el prototype)

A mod­el run­ning on a smal­ler, cur­ated set of (unwieldy) data, that lacks the pre­ci­sion of large data-driv­en mod­els. But can it still ‘good enough’ to serve com­mon­ing of food? Using food cat­egor­ies, such as ‘stir fry veg’ or ‘green veg’ instead of food items, such as ‘broc­coli’, reduces the com­plex­ity of the mod­el. Train­ing the mod­el for ‘green veg’ instead of train­ing the mod­el for broc­coli, kale, pak choi, beans, etc. takes less effort and less data and is sim­pler to negotiate.

Whatever food item hides behind ‘green veg’, many people would hap­pily serve it up as a side or use it as an ingredi­ent of their soup.

*Para­phrase of recur­ring remarks in work­shops sim­u­lat­ing col­lect­ive machine teaching.

Thing­sCon 2023 provided a bril­liant oppor­tun­ity to test the set­tings for col­lect­ively teach­ing a machine to dis­trib­ute res­cued food in par­ti­cipants’ kit­chens. No, tech skills are not neces­sary. The work­shop was dom­in­ated by shar­ing stor­ies about the sizes of fridges and freez­ers, being flex­ible with recipes, and tam­ing the chaos in fam­ily kit­chens. Story­ing and the abil­ity to reteach the machine emerged as key for coun­ter­ing the optim­ising tend­en­cies of the algorithm.

See Thing­sCon short video for a glimpse of the event and the workshop.

In machine teach­ing, labelling is the act of apply­ing a set of labels on train­ing data, which is then used to teach the mod­el that in turn will cat­egor­ise future data accord­ing to those labels. Col­lect­ive labelling is a key activ­ity in teach­ing an algorithm to decide where an ingredi­ent should be delivered at a giv­en moment in time. The work­shop Labelling Fric­tion rehearsed and reflec­ted strategies of nego­ti­at­ing labels and apply­ing them to a data­set to teach a model

The work­shop worked with 500 pantry images as train­ing data and the labels were recipes for food to which the ingredi­ents in the pantry could con­trib­ute. For the labels we tapped into and nego­ti­ated par­ti­cipants’ per­son­al lan­guages of meals and cook­ing habits. After agree­ing on a lim­ited num­ber of meals, the par­ti­cipants used these meals as labels and attached them to fur­ther images of pan­tries (train­ing data) on the plat­form Roboflow.

The train­ing data pantry images were gen­er­ated by a tool com­bin­ing our very own Data Gen­er­at­or soft­ware and Stable Dif­fu­sion AI image gen­er­at­or tool. 

The work­shop setup was inspired by the sub­vers­ive labelling prac­tice in van der Burg’s and de Boer’s pro­ject ‘Object­ive Por­trait’.  

Work­shop co-organ­iser: Iohanna Nicen­boim

Work­shop con­sult­ant: Vera van der Burg

Design f®iction exer­cise at the Design Fric­tion For Pre­dict­ive Food Com­mon­ing work­shop March 2022 Ams­ter­dam. 

 “AI-Drop, the cooper­at­ive for pre­dict­ive dis­tri­bu­tion of res­cued food is about to reach its first year of oper­a­tion. You are fol­low­ing their oper­a­tion closely. You are scrolling through your per­son­al bubble’s chat­ter on your Twit­ter feed. appears in your feed again. There is out­rage from com­munit­ies, design­ers and aca­dem­ics like­wise. What are the people you fol­low say­ing?” 

Pro­to­typ­ing cards and facil­it­a­tion cards are wide­spread utensils of co-design and facil­it­a­tion meth­od­o­lo­gies. The cards are abstrac­ted and mobile mani­fest­a­tions of accu­mu­lated prac­tice or, com­ing from the oth­er side, dia­gram­mat­ic and action­able sim­pli­fic­a­tions of the­ory. Either way, card sets intro­duce the usu­al sus­pects that emerge or ought to be addressed in the pro­cess. They com­mit to onto­lo­gies, the appro­pri­ate level of detail, and rel­ev­ant cat­egor­iz­a­tion. This way, card sets cre­ate the dis­pos­i­tions and pos­sible paths of the sys­tems, pro­cesses, organ­isa­tions or infra­struc­tures they are used to ima­gine, nego­ti­ate or prototype. Card Deck is a sub­vers­ive pro­to­typ­ing card deck for nego­ti­at­ing data streams, logist­ic­al solu­tions, and inter­faces of an auto­mated dis­tri­bu­tion plat­form for res­cued food with factors such as diet­ary require­ments, cook­ing cul­ture, avail­able pro­cessing, stor­age capa­cit­ies, hun­ger, mood, motiv­a­tion etc. It is sub­vers­ive as it comes pack­aged as an exten­sion card set for the IoT Ser­vice Kit pro­to­typ­ing card set; but instead of facil­it­at­ing the imple­ment­a­tion of a design, it aims at extend­ing or rearran­ging the onto­lo­gies of ingredi­ents, meals, bod­ies and waste in from the per­spect­ive of nego­ti­ation human and more-than-human use value.

The Card Deck aims at:

  • Nego­ti­at­ing the onto­lo­gies of meals, ingredi­ents, and waste trans­formed by logist­ics and pro­cessing (kit­chens, bod­ies, bacteria).
  • For­mu­lat­ing stor­ies of com­mon­ing-based infra­struc­tures and auto­ma­tion that emerge from spec­u­la­tion about sens­ing; think­ing about the kinds of data and sens­ing needed to tell the stor­ies worth telling.
  • Reveal­ing the extent of the over­lap between the ele­ments of com­mon­ing-based and more-than-human auto­mated infra­struc­tures and main­stream IoT ser­vice architectures.
  • Scru­tin­ise the dir­ect con­nec­tion between col­lec­ted data and the func­tion­al­it­ies it is fuel­ing to strive towards min­im­al feas­ible datafic­a­tion (see Undo­ing Optim­iz­a­tion by Alis­on Powell).
  • Nego­ti­ate the fric­tion between het­ero­gen­eous cook­ing habits and stand­ard­isa­tion of the system.

Basel Shape Walk with bach­el­or stu­dents of the “Grasp­ing The Future City” class at FHWN Academy of Arts and Design, Basel, Septem­ber 2021.

In On Non­scalab­il­ity, Anna Tsing com­pares colo­ni­al sug­ar­cane pro­duc­tion with defi­ant mat­su­take mush­room for­aging in today’s North Amer­ica to out­line the lim­its of scalab­il­ity. She cre­ates aware­ness of how scal­ing-up oper­a­tions change the very nature of the scaled pro­ject. Scal­ing Mater­i­al Urb­an Com­mons (SMUC) is inves­ted in under­stand­ing scale domains and the moment when the nature of com­mon­ing changes dur­ing city-wide scal­ing up. 

The Basel Shape Walk invest­ig­ates cog­nit­ive scale domain bound­ar­ies by explor­ing our capa­city for recall­ing the length and shape of urb­an walks. It loc­ates bound­ar­ies of recallab­il­ity by walk­ing dif­fer­ent paths and explor­ing when our impli­cit cap­ab­il­ity of recall­ing the walk starts to fail us. In the walk, a blind­folded per­son and a nav­ig­at­or both walk a pre­set path of around 20 minutes in the streets. After the walk, the blind­folded per­son attempts to recall and draw the route. The choice of the path’s dur­a­tion and shape make recall­ing dif­fi­cult but not impossible, even if it takes sev­er­al attempts and cor­rec­tions. In the Basel Shape Walk ses­sion, most par­ti­cipants could recall and pro­duce the path on paper (image 1–8). How­ever, the exer­cise proved to be more chal­len­ging in the ori­gin­al 2009 Zagreb ver­sion of the exper­i­ment, where only one per­son from five could repro­duce the Zagreb paths on paper (images 9–10). Alva Noë’s concept of trans­mod­al­ity between touch and vis­ion in Action in Per­cep­tion provides the experiment’s philo­soph­ic­al back­ground for recall­ing the per­cep­tion of the walked path (touch) through draw­ing (vis­ion). Trans­mod­al­ity sug­gests that spa­tial con­cepts that we acquire through touch and vis­ion are fun­da­ment­ally sim­il­ar (they pro­duce sim­il­ar sen­sor­imo­tor pat­terns). In effect, we can recall shapes that we have per­ceived through touch by draw­ing them; this is less a trans­la­tion between visu­al and tact­ile con­cepts as it is the recall of the same concept but in a dif­fer­ent mod­al­ity. The walk­ing exper­i­ment intro­duces the issue of scale to trans­mod­al­ity by play­ing with the ques­tion of what size of the path (meter or kilo­metre scale) is too large to be recalled in draw­ing (cen­ti­metre scale). With routes that are just large and com­plex enough that their recall prompts a chal­lenge, the exper­i­ment demarc­ates the bound­ar­ies of a scale domain in which grasp­ing and recall­ing an impli­cit, embod­ied spa­tial concept is still pos­sible. The struggle of expli­citly repro­du­cing an impli­cit spa­tial concept through draw­ing is the mani­fest­a­tion of that boundary.


Noë’s concept of trans­mod­al­ity between touch and vis­ion is thus exten­ded in the walk through the play­ful intro­duc­tion of scale. Trans­mod­al­ity sug­gests that we acquire the same spa­tial con­cepts through touch and vis­ion: on an abstract level there is an impli­cit sim­il­ar­ity between the sen­sor­imo­tor pat­terns of shapes pro­duced by touch and vis­ion, and recall is an indic­at­or of know­ledge, or in oth­er words a grasp, in the sense of a sen­sor­imo­tor concept. 

The Basel Shape Walk exper­i­ments with trans­mod­al­ity across scales (ca. 0.4 km of the walk and cm of the draw­ing) and explores the size (and com­plex­ity of the shape) that is on the brink of recallability.


FHNW Uni­ver­sity of Applied Sci­ences and Arts North­west­ern Switzerland
Academy of Art and Design
Insti­tute Exper­i­ment­al Design and Media Cul­tures (IXDM)
Freil­ager-Platz 1
CH-4002 Basel

About SMUC

Scal­ing Mater­i­al Urb­an Com­mons (SMUC) crafts ima­gin­ar­ies of urb­an futures that recon­cile auto­ma­tion and pre­dict­ive tech­no­lo­gies with com­mon­ing- and care-based use of resources. SMUC uses spec­u­lat­ive city-mak­ing to scale up the com­mon­ing of res­cued food. It does so by pro­to­typ­ing a pre­dict­ive algorithm-based sys­tem, dubbed, that orches­trates the col­lec­tion and drop-off of res­cued food in Basel and London.

Draft ana­tomy of an infra­struc­ture for food res­cue, sens­ing body and kit­chen con­di­tions, col­lect­ive machine teach­ing and algorithmic dis­tri­bu­tion of food items.

More detail to follow…

Work­shop about the clash between auto­mated food deliv­ery and het­ero­gen­eous, idio­syn­crat­ic eat­ing cul­tures, Ams­ter­dam and Online, March 2022.

Auto­ma­tion is closely inter­twined with paradigms of effi­ciency and thrives in repet­it­ive set­tings. Auto­ma­tion is con­du­cive to repe­ti­tion through its uni­fy­ing and stand­ard­ising tech­niques. Auto­ma­tion is argu­ably adverse to nego­ti­at­ing diverse interests, rela­tions or het­ero­gen­eous and idio­syn­crat­ic prac­tices. SMUC invest­ig­ates the fric­tion between plat­form-scale tech­no­lo­gies of auto­ma­tion and het­ero­gen­eous, situ­ated and messy loc­al eat­ing cultures. 

Design Fic­tion is a meth­od­o­logy for cre­at­ing ima­gin­ar­ies of the future that chal­lenge the status quo and enable the scru­tiny of pos­sible futures. SMUC pro­poses the Design F®iction meth­od to expose the fric­tion between the auto­ma­tion of res­cued food dis­tri­bu­tion and com­mon­ing-minded social, polit­ic­al and eco­nom­ic per­spect­ives. The inter­dis­cip­lin­ary work­shop aims to activ­ate the ima­gin­ary of the auto­mated com­mon­ing infra­struc­ture through storytelling and play­ful sim­u­la­tion, and map out where scru­tiny is required. Instead of dis­miss­ing auto­ma­tion in com­mon­ing prac­tices, the workshop’s ambi­tion is to crit­ic­ally but con­struct­ively probe the design space for auto­ma­tion in urb­an food com­mon­ing. The auto­mated infra­struc­ture is equal parts pro­pos­al and provocation.

Meth­ods used in the work­shop include: map­ping prob­lem­at­ic aspects of the infra­struc­ture, play­ful sim­u­la­tion of the infra­struc­ture by enact­ing urb­an data streams, and enact­ing and teach­ing a machine-learn­ing algorithm writ­ing fict­ive future Tweets about the back­lash against infrastructure. 

Work­shop par­ti­cipants were (either in one or both work­shop ses­sions): Jam­ie Allen, Cristina Ampatzidou, Roy Bendor, Jaz Choi, Joshua Ents­minger, Gab­ri­ele Ferri, Györgyi Gálik, Dan Lock­ton, Iohanna Nicen­boim, Iskander Smit, Mar­tijn de Waal, Gab­ri­ela Aquije Zegarra.

Concept pro­to­type for the infra­struc­ture. Ams­ter­dam, Basel, Decem­ber 2021.

One of the crit­ic­al aspects of auto­mat­ing the deliv­ery of res­cued food is explor­ing where res­cued food can real­ist­ic­ally pro­duce a bal­anced meal. SMUC partnered with the Mas­ter of Digit­al Design (MDD) at the Ams­ter­dam Uni­ver­sity of Applied Sci­ences to explore an exper­i­ment­al recipe gen­er­at­or mod­ule for the infra­struc­ture for dis­trib­ut­ing res­cued food. MDD stu­dents Anouk van Asbeck, Cater­ina Maluenda Guer­rero, Utkarsh Srivast­ava and Michaela Kořistová have con­duc­ted design research and developed a concept pro­to­type dubbed “Wasty”.

Wasty is a recipe-box-style ser­vice that gen­er­ates per­son­al­ised recipes in real time based on the avail­ab­il­ity of res­cued ingredi­ents, indi­vidu­al cook­ing habits, diet­ary needs, cook­ing cul­ture, avail­able cook­ing equip­ment and ingredi­ents already avail­able in the kit­chen. As a design research arte­fact, the recipe gen­er­at­or con­siders these factors and deliv­ers point­ers for teach­ing an algorithm about where to deliv­er reduced food.

Meth­od­o­lo­gic­al explor­a­tion with bach­el­or stu­dents of the “Grasp­ing The Future City” class at FHWN Academy of Arts and Design, Basel, Septem­ber 2021.

Ser­vice design prac­tice is pro­foundly integ­rat­ing user research and user feed­back in design decisions. How­ever, if the struggle for eco­lo­gic­al futures requires the inclu­sion of oth­er-than-human beings in the address­ees of design, pre­val­ent design research meth­ods for identi­fy­ing user needs and under­stand­ing user exper­i­ence will fail the design­er. This meth­od­o­lo­gic­al chal­lenge motiv­ated the cre­ation of a work­shop format­ted to extend the IoT Ser­vice Kit (Inter­net of Things Ser­vice Kit). 

The IoT Ser­vice Kit is an open access card set developed by the innov­a­tion agency Futurice to facil­it­ate the con­ver­sa­tion between tech­no­lo­gists and diverse stake­hold­ers about IoT-enabled ser­vices in homes, indus­tri­al sites, and the city. The kit is a card set con­tain­ing sensor types, inter­ac­tion types, devices, users, and data sets, all of which are build­ing blocks of IoT services. 

The work­shop incor­por­ated parts of the Moth City probe kit and exer­cises enact­ing non-humans to enable par­ti­cipants to shift from an exclus­ively human per­spect­ive, thus cre­at­ing an ‘oth­er-than-human exten­sion card set’ for the IoT ser­vice kit. Vikt­or hos­ted the first instance of the work­shop for design stu­dents in his exper­i­ment­al design prac­tice course ‘Grasp­ing the Future City’ at the Crit­ic­al Media Lab, Basel.

These exer­cises and fur­ther work­shops will res­ult in the pub­lic­a­tion of an exten­sion card set for the IoT Ser­vice Kit.

Work­shop par­ti­cip­a­tion and pub­lic­a­tion, Sum­mer 2021, London.

SMUC has a two­fold interest in more-than-human design principles: 

1) It is inter­ested in infra­struc­tures that turn food into com­mons by giv­ing it a new pur­pose once it has been taken off super­mar­ket shelves. The con­cern with the time win­dow between super­mar­ket shelves and food waste fur­ther extends to the cir­cu­la­tion of food after it is con­sum­able for humans. 

2) SMUC’s con­cern with care-based and eco­lo­gic­al approaches to auto­mated and pre­dict­ive tech­no­lo­gies in com­mon­ing motiv­ates an invest­ig­a­tion into ways of includ­ing non-human agents (such as algorithmic agents) with­in com­mon­ing practice. 

Vikt­or par­ti­cip­ated in the ‘Moth City’ work­shops held by the pro­ject More-than-Human: data inter­ac­tions in the smart city. The work­shop explored role­play and enact­ing urb­an crit­ters as a meth­od to decen­ter an exclus­ively human per­spect­ive and probe non-human per­spect­ives on urb­an life. The work­shop aimed to gen­er­ate embod­ied design know­ledge that enables spec­u­la­tion about data ser­vices with oth­er-than-human users in mind.

Viktor’s reflec­tions on the work­shops and the struggle of shift­ing to oth­er-than-human per­spect­ives appear in the Moth Cit­ies doc­u­ment­a­tion book­let.


FHNW Uni­ver­sity of Applied Sci­ences and Arts North­west­ern Switzerland
Academy of Art and Design
Insti­tute Exper­i­ment­al Design and Media Cul­tures (IXDM)
Freil­ager-Platz 1
CH-4002 Basel

About SMUC

Scal­ing Mater­i­al Urb­an Com­mons (SMUC) crafts ima­gin­ar­ies of urb­an futures that recon­cile auto­ma­tion and pre­dict­ive tech­no­lo­gies with com­mon­ing- and care-based use of resources. SMUC uses spec­u­lat­ive city-mak­ing to scale up the com­mon­ing of res­cued food. It does so by pro­to­typ­ing a pre­dict­ive algorithm-based sys­tem, dubbed, that orches­trates the col­lec­tion and drop-off of res­cued food in Basel and London.