Sartre’s Bad Faith and The Algorithm Decided

What is Sartre’s concept of bad faith?

Bad faith is the act of deceiving yourself about your own freedom, pretending that external circumstances or roles determine your choices when you are, in fact, radically free to choose otherwise.

Sartre’s most famous illustration is the waiter who plays at being a waiter. His movements are too precise, too eager, too perfectly “waiter-like.” He has collapsed his identity into his role so completely that he no longer experiences himself as a free being who chose this job and could choose otherwise. The role has become an excuse to stop choosing.

The concept is uncomfortable because it applies universally. Every time you say “I had no choice,” Sartre would argue you are in bad faith. You had a choice. You may not have liked the alternatives. The consequences may have been severe. But the choice existed, and pretending otherwise is a flight from the anxiety of freedom.

How does “the algorithm decided” function as bad faith?

Saying “the algorithm decided” transfers agency from the humans who designed, trained, and deployed the system to the system itself, creating a fiction of mechanical objectivity that shields decision-makers from moral scrutiny.

I sat in a meeting where a product manager explained that loan applicants were being denied because “the model flagged them.” When I asked what features the model used, the room went quiet. When I asked who chose those features, the room went quieter. When I asked who decided the threshold for denial, someone changed the subject.

The algorithm did not decide. A team of humans decided what data to collect, what features to engineer, what objective function to optimize, what threshold to set, and what populations to test against. At every step, a human made a choice. The algorithm is the crystallization of those choices into a form that can be executed without further human deliberation. That is its utility and its danger: it automates the choices while obscuring the choosers.

Sartre would recognize this pattern instantly. It is the waiter, again. The organization plays at being determined by its algorithms. The movements are too precise: “data-driven decision-making,” “automated workflows,” “model-based recommendations.” Each phrase is a small performance of objectivity that conceals a prior act of subjective choice.

“Man is condemned to be free; because once thrown into the world, he is responsible for everything he does.” — Jean-Paul Sartre, Existentialism Is a Humanism

Where does technological determinism become self-deception?

Technological determinism becomes self-deception at the exact moment when humans who could change the system claim they cannot, transforming a design choice into an apparent fact of nature.

I have identified 4 distinct patterns of algorithmic bad faith in organizations I have worked with:

• The Delegation Dodge: “The recommendation engine suggested it.” But someone built the recommendation engine, chose its training data, and decided to follow its suggestions without human review.
• The Complexity Shield: “The model is too complex to explain.” But someone chose to deploy a model too complex to explain in a context where explanation matters.
• The Data Defense: “The data shows this is the right decision.” But someone chose what data to collect, how to clean it, and what questions to ask of it.
• The Scale Excuse: “We cannot review every decision at this scale.” But someone chose to operate at a scale that requires automation, and that choice carries moral weight.

Each pattern has the same structure: a human choice is repackaged as a mechanical inevitability. The packaging is the bad faith. The choice was real. The inevitability is a story.

What does authentic engagement with algorithmic systems look like?

Authentic engagement means owning every decision embedded in the system, from data selection to deployment context, and accepting that “the algorithm decided” is never a complete sentence.

Sartre’s alternative to bad faith is what he calls authenticity: the willingness to confront your freedom directly, to acknowledge that you are choosing and that your choices have consequences you cannot delegate. In the context of algorithmic systems, authenticity has concrete expressions.

An authentic model card does not say “the model predicts X.” It says “we trained this model on Y data, optimizing for Z objective, knowing that this would produce disparate outcomes for population W, and we decided this trade-off was acceptable because of reasons A, B, and C.” Every passive construction becomes active. Every “the system does” becomes “we chose to build a system that does.”

I rewrote the documentation for a content moderation system using this framework. The original document was 12 pages of passive voice: “content is classified,” “decisions are rendered,” “appeals are processed.” The rewrite was 8 pages of active voice: “we classify content using these criteria,” “we render decisions at these thresholds,” “we process appeals within this timeline.” The engineering did not change. The moral clarity did.

Why does this matter beyond philosophy?

Algorithmic bad faith has regulatory, legal, and social consequences: the EU AI Act, GDPR’s right to explanation, and emerging case law all reject the premise that algorithms make decisions independent of human responsibility.

The legal landscape is catching up to Sartre. The EU AI Act, effective since 2024, requires that high-risk AI systems provide meaningful human oversight. GDPR’s Article 22 gives individuals the right not to be subject to decisions based solely on automated processing. Courts in 9 countries have ruled that “the algorithm decided” is not an acceptable defense against discrimination claims.

These legal developments are philosophical developments in disguise. They encode the principle that automated systems are extensions of human agency, not replacements for it. Every algorithm is a delegation of choice, and delegation does not eliminate responsibility. It extends it.

Sartre wrote that we are “condemned to be free.” The engineers and product managers who build algorithmic systems are condemned to choose. The training data is a choice. The objective function is a choice. The deployment context is a choice. The decision not to audit for bias is a choice. Saying “the algorithm decided” is the most expensive form of bad faith in the modern economy, not because the philosophy is wrong, but because the consequences land on people who never consented to be subjects of someone else’s flight from freedom.