Nathalie Smuha

Nathalie Smuha's central preoccupation is what happens when governments hand over the execution of law to algorithms, and she is unambiguous that this is never a neutral technical transfer. "Die vertaalslag van een wet naar code, dat is geen puur technisch iets, dat is eigenlijk iets dat politieke en normatieve implicaties heeft" ¹². Her go-to illustration is Indiana, where "refusal to cooperate" with welfare authorities got coded as simply not answering the phone — an interpretation no legislator intended, but one baked permanently into a system now applied at scale ¹². For Smuha this is the crux of the problem: coders end up making normative choices that were supposed to belong to democratically accountable lawmakers, and once made, those choices get executed uniformly and invisibly across an entire population.

Scale is her second recurring theme. Where a human bureaucrat's error affects one case, an algorithmic error affects everyone processed by that system, and she points to the UK Post Office accounting scandal as the sobering proof — a bug in the system that led to wrongful theft prosecutions, imprisonment, and suicides before anyone caught it ¹. She's equally alert to feedback loops that manufacture their own confirmation: predictive policing sends officers to neighborhoods the algorithm flags, crime gets found there because that's where police are looking, and the algorithm appears validated while crime elsewhere goes unrecorded, locking the same communities into permanent suspicion ¹². Function creep is the third recurring pattern — infrastructure justified for one purpose quietly migrates to another. Cameras installed to protect Antwerp's Jewish quarter after terrorist attacks were later used during COVID to check whether that same community was following lockdown rules ¹², and Belgium's COVID pass was legalized in a way that lets the system return in a future crisis. Her warning here is pointed: "Er is eigenlijk geen limiet aan wat je kan doen in de naam van veiligheid en we moeten ons daar een beetje tegen verzetten" ¹².

Smuha's framing for why this matters isn't abstract civil-liberties anxiety but a structural argument about the rule of law itself. Laws are supposed to be contestable, revisable, and applied through processes that allow for judgment and correction; she worries that algorithmic administration converts law into "algorithmic rule by law" — a technically enforced version of law that becomes rigid, opaque, and resistant to the political correction that ordinary law allows for. Crucially, she insists this isn't a bet you make on the current government's good intentions: "Denk niet aan de overheid die we vandaag hebben, denk aan volgende verkiezingen en uw nachtmerrie politieke partij wordt verkozen" ¹². The infrastructure and the powers you build now are inherited by whoever comes next, and "wat vandaag legaal is kan morgen illegaal worden gemaakt door een andere overheid" ¹² — meaning safeguards need to be built for the worst plausible future government, not the current one.

Underneath these institutional arguments sits a philosophical line she returns to explicitly: some judgments are not the kind of thing that should be delegated to a machine at all. "Die waardeoordelen is iets dat we niet mogen of kunnen delegeren aan AI-systemen... zelf uitmaken wat voor ons een waardevol leven is" ¹². This is her check on techno-solutionism — not every decision that can be automated should be, because some decisions constitute the very autonomy that law and democracy are meant to protect.

On the EU AI Act, her position has a clear structure: necessary but insufficient, and mis-designed in a specific way. She names the flaw precisely — the Act relies on self-assessment, so even high-risk systems like welfare-decision algorithms or predictive policing get evaluated by the deploying organization itself, with independent scrutiny only arriving after complaints, "potentially three years of damage later" ¹. Her deeper critique is conceptual: the Act treats AI like a product to be certified safe through technical standards, "zoals een wasmachine," but risks to human rights, democracy, and the rule of law don't reduce to technical specifications ¹². Against Macron's push to loosen the regulation, she argues for the opposite move — not looser, but smarter: simplify the compliance burden while closing the gaps that currently let the most consequential government uses slip through the Act's lists ¹². She's equally skeptical of transparency as a cure-all — publishing an algorithm's code doesn't help most citizens, doesn't stop errors slipping in, and doesn't guarantee anyone is actually accountable for the final decision ¹².

She closes her critique of legal instruments with a warning that cuts against a common liberal assumption — that having law on your side is automatically protective. "Het recht kan dus eigenlijk ook worden gebruikt om onze rechten en vrijheden uit te hollen en dat is wat rule by law is" ². Law, in other words, can be the mechanism of erosion rather than its remedy, which is precisely why she wants sunset clauses, independent verification, and structural limits built in now, before infrastructure and precedent make reversal costly.

Smuha extends the same skepticism about hype and necessity to the humanoid robotics wave, treating it as a useful analogy for how technological narratives outpace evidence. She calls the current rush of humanoid robot announcements "communicative acceleration" — once a major player announces progress, competitors must respond in kind mainly to keep investors interested, without corresponding proof of real technical leaps ¹². Her practical judgment is that general-purpose humanoid robots are probably the wrong design goal altogether: specialized machines that do one task well, like a robot vacuum, make more sense, since even a task as simple as picking up an egg without crushing it remains extremely hard for robots. The pull toward humanoid form, she argues, comes from fascination with building something human-like rather than actual need ¹² — the same diagnosis, in miniature, that she applies to governments adopting algorithmic systems: a mix of fascination with the technology and pressure to cut costs, with little evidence the systems actually deliver on efficiency or effectiveness ¹².