PhD update – September 2022

Sev­en months since the last update. Much bet­ter than the gap of three years between the pre­vi­ous two. These past months I feel like I have begun to reap the rewards of the grunt work of the last cou­ple of years. Two papers final­ly saw the light of day, as well as a course syl­labus. Read on for some more details.

Things that hap­pened:

First, a pair of talks. In Feb­ru­ary I pre­sent­ed on “Con­testable AI & Civic Co-Design” as part of a pan­el chaired by Roy Ben­dor at Rein­vent­ing the City. A PDF of my slides is avail­able on the contestable.ai web­site, here. In March, I pre­sent­ed at the AiTech Ago­ra. The title of the talk is “Mean­ing­ful Human Con­trol Through Con­testa­bil­i­ty by Design” and the slides are avail­able here.

In Feb­ru­ary a short inter­view was pub­lished by Bold Cities, a smart city research cen­ter I am loose­ly affil­i­at­ed with.

Then, in March, came a big moment for me, with the pub­li­ca­tion of my first jour­nal arti­cle in AI & Soci­ety. Here’s the abstract, and ref­er­ence. It’s avail­able open access.

The increas­ing use of arti­fi­cial intel­li­gence (AI) by pub­lic actors has led to a push for more trans­paren­cy. Pre­vi­ous research has con­cep­tu­al­ized AI trans­paren­cy as knowl­edge that empow­ers cit­i­zens and experts to make informed choic­es about the use and gov­er­nance of AI. Con­verse­ly, in this paper, we crit­i­cal­ly exam­ine if trans­paren­cy-as-knowl­edge is an appro­pri­ate con­cept for a pub­lic realm where pri­vate inter­ests inter­sect with demo­c­ra­t­ic con­cerns. We con­duct a prac­tice-based design research study in which we pro­to­type and eval­u­ate a trans­par­ent smart elec­tric vehi­cle charge point, and inves­ti­gate experts’ and cit­i­zens’ under­stand­ing of AI trans­paren­cy. We find that cit­i­zens expe­ri­ence trans­paren­cy as bur­den­some; experts hope trans­paren­cy ensures accep­tance, while cit­i­zens are most­ly indif­fer­ent to AI; and with absent means of con­trol, cit­i­zens ques­tion transparency’s rel­e­vance. The ten­sions we iden­ti­fy sug­gest trans­paren­cy can­not be reduced to a prod­uct fea­ture, but should be seen as a medi­a­tor of debate between experts and citizens.

Alfrink, Kars, Ianus Keller, Neelke Doorn, and Gerd Kortuem. “Ten­sions in Trans­par­ent Urban AI: Design­ing a Smart Elec­tric Vehi­cle Charge Point.” AI & SOCIETY, March 31, 2022. https://doi.org/10/gpszwh.

In April, the Respon­si­ble Sens­ing Lab pub­lished a report on “Respon­si­ble Drones”, to which I con­tributed a lit­tle as par­tic­i­pant on work­shops that lead up to it.

A sec­ond big mile­stone for me was mak­ing pub­lic the syl­labus for indus­tri­al design engi­neer­ing mas­ter elec­tive course “AI & Soci­ety” (no rela­tion to the jour­nal) which I have been devel­op­ing under the guid­ance of my super­vi­sor Gerd Kortuem over the past cou­ple of years. The syl­labus con­tains a read­ing list, as well as many self-guid­ed design exer­cis­es. Here’s a short description:

Arti­fi­cial Intel­li­gence (AI) is increas­ing­ly used by a vari­ety of orga­ni­za­tions in ways that impact soci­ety at scale. This 6 EC mas­ter elec­tive course aims to equip stu­dents with tools and meth­ods for the respon­si­ble design of pub­lic AI. Dur­ing sev­en weeks stu­dents attend a full-day ses­sion of lec­tures and work­shops. Stu­dents col­lab­o­rate on a group design project through­out. At the end, stu­dents indi­vid­u­al­ly deliv­er a short paper.

ID5417 Arti­fi­cial Intel­li­gence and Society

The third big mile­stone was the pub­li­ca­tion of my sec­ond jour­nal arti­cle in Minds & Machines. It is the the­o­ret­i­cal cor­ner­stone of my the­sis, a pro­vi­sion­al frame­work for design­ing con­testa­bil­i­ty into AI sys­tems. Abstract and ref­er­ence fol­low. This one is also open access.

As the use of AI sys­tems con­tin­ues to increase, so do con­cerns over their lack of fair­ness, legit­i­ma­cy and account­abil­i­ty. Such harm­ful auto­mat­ed deci­sion-mak­ing can be guard­ed against by ensur­ing AI sys­tems are con­testable by design: respon­sive to human inter­ven­tion through­out the sys­tem life­cy­cle. Con­testable AI by design is a small but grow­ing field of research. How­ev­er, most avail­able knowl­edge requires a sig­nif­i­cant amount of trans­la­tion to be applic­a­ble in prac­tice. A proven way of con­vey­ing inter­me­di­ate-lev­el, gen­er­a­tive design knowl­edge is in the form of frame­works. In this arti­cle we use qual­i­ta­tive-inter­pre­ta­tive meth­ods and visu­al map­ping tech­niques to extract from the lit­er­a­ture sociotech­ni­cal fea­tures and prac­tices that con­tribute to con­testable AI, and syn­the­size these into a design framework. 

Alfrink, Kars, Ianus Keller, Gerd Kortuem, and Neelke Doorn. “Con­testable AI by Design: Towards a Frame­work.” Minds and Machines, August 13, 2022. https://doi.org/10/gqnjcs.

Around the same time in August, Fabi­an Geis­er, whom I had been men­tor­ing for some time, grad­u­at­ed with a fas­ci­nat­ing mas­ter the­sis and project with the title “Reimag­in­ing the smart allo­ca­tion of road space in Ams­ter­dam for fair­ness”.

And final­ly, as these things were going on, I have been qui­et­ly chip­ping away at a third paper that applies the con­testable AI by design frame­work to the phe­nom­e­non of cam­era cars used by munic­i­pal­i­ties. My aim was to cre­ate an exam­ple of what I mean by con­testable AI, and use the exam­ple to inter­view civ­il ser­vants about their views on the chal­lenges fac­ing imple­men­ta­tion of con­testa­bil­i­ty in the pub­lic AI sys­tems they are involved with. I’ve sub­mit­ted the man­u­script, titled “Con­testable Cam­era Cars: A spec­u­la­tive design explo­ration of pub­lic AI that is open and respon­sive to dis­pute”, to CHI, and will hear back ear­ly Novem­ber. Fin­gers crossed for that one.

Look­ing ahead:

So what’s next? Well, I have lit­tle under a year left on my PhD con­tract, so I should real­ly begin wrap­ping up. I am con­sid­er­ing a final pub­li­ca­tion, but have not set­tled on any top­ic in par­tic­u­lar yet. Cur­rent inter­ests include AI sys­tem mon­i­tor­ing, visu­al meth­ods, and more besides. Once that final paper is in the can I will turn my atten­tion to putting togeth­er the the­sis itself, which is paper-based, so most­ly requires writ­ing an over­all intro­duc­tion and con­clu­sion to book­end the includ­ed pub­li­ca­tions. Should be a piece of cake, right?

And after the PhD? I am not sure yet, but I hope to remain involved in research and teach­ing, while at the same time per­haps get­ting a bit more back into design prac­tice besides. If at all pos­si­ble, hope­ful­ly in the domain of pub­lic sec­tor appli­ca­tions of AI.

That’s it for this update. I will be back at some point when there is more news to share.