Clara Girke & Valena Ammon – Bouncing on the Surface of Latent Space

Latent space is a space crea­ted within machi­ne lear­ning pro­ces­ses. Objects in this space will be posi­tio­ned depen­ding on dif­fe­rent resem­blan­ces defi­ned by varia­bles which results in the emer­gence of a space with many dimen­si­ons, con­nec­tions and networks.

We deci­ded to use the time in our resi­den­cy to explo­re the topic of arti­fi­ci­al intel­li­gence (AI) aka machi­ne lear­ning – hence to make our first steps into latent space. But let us start from the beginning.

Last year we heard about AI’s being able to crea­te images based on natu­ral lan­guage and it was an intri­guing visi­on. Being able to just type a sen­tence that comes to mind and short­ly after being pre­sen­ted an image (or even a varie­ty of images to pick from). We heard about it from the distance some­ti­mes saw some AI gene­ra­ted images—but never tried it by our­sel­ves. So coming to Fau­co­gney-et-la-Mer we felt that we had final­ly the time to explore.

As a duo we stri­ve to con­nect illu­stra­ti­on with cera­mic, color with mate­ri­al, lines with volu­me and visu­al with con­tex­tu­al. Usual­ly working in the field of ves­sels we deci­ded to stay in this area of objects. As a warm-up we deci­ded to gene­ra­ted dra­wings by our­sel­ves and then feed the text-to-image AI with the same prompt we gave our­sel­ves. Being able to compa­re our crea­ti­ons with the ones from the AI.
We obser­ved and slow­ly got a fee­ling for the algo­rithm and how it may work.

We stay­ed with some images gene­ra­ted (by us or the algo­rithm) and took one step fur­ther. Coming up with more and more detail­ed prompts we fed into dif­fe­rent AIs to gene­ra­te designs of cups that we wan­ted to have as output—working hard on our skill to per­form the craft of promp­ting. On the way we found many tools we could use, artists and desi­gners working with AI and com­mu­ni­ties occu­p­ied by the wish to make AI more equal and just.

One step fur­ther. Giving the algo­rithm an image to work with. To alter, to chan­ge and modi­fy. Picking one of the gene­ra­ted objects and alte­ring it again crea­tes a loop of crea­ting and reac­ting bet­ween the algo­rithm and us as human desi­gners.
Obser­ving the cur­rent abili­ty of the AI to gene­ra­te images we were inspi­red by glit­ches and the out­put we dee­med imperfect—making the AI a sym­pa­the­tic and may­be even humo­rous part­ner to work with.

And then final­ly brin­ging some­thing from a dra­wing with words to be gene­ra­ted by an AI and even­tual­ly mold­ed with clay into the ana­log world and trans­forming it into some­thing we can touch. Fol­lo­wed by fee­ding it as an image back into the digi­tal world to be pro­ces­sed again. Crea­ting a never ending loop of crea­ti­on and decis­i­on making—choosing with which out­put we want to keep on going and crea­ting. Lost in trans­la­ti­on bet­ween the ana­log and the digi­tal. Always wondering:

Is it pos­si­ble to tru­ly coope­ra­te with an AI on eye level?

Reflec­ting on our role as decis­i­on-makers in this pro­cess. Always fil­te­ring, clas­si­fy­ing, reor­de­ring and choo­sing with which out­put to work further—making it our next input.

The steps we took and the exer­cis­es we made within this explo­ra­ti­on are just the first to come fol­lo­wed by many more. During the pro­cess we stumb­led upon many que­sti­ons and not all could be ans­we­red.
To finish this blog­post we like to share some of them with you—an attempt to drag you deep into the latent space in our human heads.

How can Machi­ne Lear­ning be ethi­cal and just if alre­a­dy the data­set it is built upon reflects a white, male euro­cen­tric world view?
What is the role of peo­p­le trai­ning AI systems and what is our impact and respon­si­bi­li­ty as users?

What are the impli­ca­ti­ons of copy­right and owner­ship for a tool being able to copy styl­es and brush­strokes from one spe­ci­fic artist/designer/style?

Can text-to-image machi­ne lear­ning be part of our design pro­cess (Gestal­tungs­pro­zess)?

What rest­ric­tions are requi­red to pre­vent peo­p­le to use this tech­no­lo­gy in a harmful and/or vio­lent way?

How can we dream bey­ond how AI is used?

Dan­ke an Vil­la Vil­le­kul­la, die uns ein­ge­la­den haben an die­sen wun­der­schö­nen Ort und es geschafft haben uns mit die­sen inspi­rie­ren­den Men­schen zusam­men­zu­brin­gen. Mer­ci Ali, Ani­ta und Che­yenne fürs gemein­sa­me Ver­wei­len und Gedan­ken spin­nen.
Nikos Probst für die Ein­la­dung zum Mid­jour­ney Dis­cord Ser­ver.
Ame­lie Gold­fuß für den Zugang zu DALLE 2.
Die Platt­form Dre­a­ming­Bey­on­dAI für den schö­nen und inspi­rie­ren­den Launch-Event und die wun­der­vol­le Plattform.Danke an all die Men­schen, die ihre Zeit in das Ent­wickeln von Tools stecken und Infor­ma­tio­nen auf­ar­bei­ten und tei­len, um die­se ande­ren zugäng­lich zu machen.
Dan­ke an Men­schen, die sich mit ethi­schen, sozia­len, gesell­schaft­li­chen und öko­lo­gi­schen Aus­wir­kun­gen von Tech­no­lo­gien beschäf­ti­gen, die­se kri­tisch hin­ter­fra­gen und nach neu­en Wegen suchen.