When AI Interviewer Meets AI Job Candidate: Lessons from an Absurd Interview
An absurd experiment of an AI interviewer meeting an AI candidate reveals the circular dilemma and decision-making boundaries in automated recruiting.
Using a short Hacker News post about "AI interviewing AI" as a starting point, this article dissects the deeper issues behind the AI recruiting wave. It points out that AI interview tools have spread rapidly due to efficiency demands, but the historical bias embedded in their training data makes "objective assessment" a false premise. When both sides of the interview are AI, evaluation criteria fall into a self-referential loop resembling a GAN game, while the meta-signals humans rely on to judge authenticity become completely ineffective. This leads to three serious issues: the lack of explainability in AI decisions, the institutional vacuum of accountability, and the irreplaceability of human judgment in dimensions such as teamwork and growth potential. The author ultimately argues that AI should serve as an auxiliary tool rather than the final decision-maker, and that technology must serve human goals rather than letting human goals disappear into the maze of technology.
Introduction: An Interview Without Humans
What happens when an AI interviewer meets an AI job candidate? The answer: chaos. This short post on Hacker News, with 5 points and no comments yet, captures in a single sentence a deeply ironic scenario—both ends of the hiring pipeline are AI, and humans have been completely excluded.
But this seemingly comical "automated interview" experiment exposes some issues worth serious consideration in the current wave of AI adoption: Are we handing decision-making power to systems that aren't fully mature yet? When both ends are AI, how much of the original purpose of hiring—finding the most suitable person—actually remains?
The Rise of AI Interviews: Technology Deployment Driven by Efficiency
AI interview tools have rapidly proliferated in corporate recruiting in recent years. From resume screening to video interview analysis, AI is being used to evaluate candidates' verbal expression, emotional states, and even "cultural fit." The logic from supporters is straightforward: hiring processes involve massive repetitive labor and human bias; AI can improve efficiency, reduce costs, and claims to offer "more objective" assessments.
The problem is that the training data for AI interview systems often comes from historical hiring decisions, which themselves contain bias. When AI is used to judge "who is a good candidate," it may simply be replicating past inequalities rather than eliminating them.
The bias problem in AI interview systems has been confirmed in multiple empirical studies. In 2018, Amazon was revealed to have an internal AI recruiting tool that automatically downgraded resumes containing the word "women" because its training data was predominantly male resumes; the project was ultimately abandoned. In 2021, Dutch researchers found that a commercial video interview AI systematically scored non-white candidates lower. The common root cause of these cases is that the model learns statistical patterns from historical hiring decisions, and the historical data itself carries human bias. A more insidious risk lies in the "proxy variable" problem—even if sensitive fields like gender and race are removed, information such as zip code, alma mater, and hobbies can still indirectly encode these attributes, causing bias to persist in hard-to-detect ways. Therefore, "data debiasing" alone is not enough to solve the problem; systematic constraints must also be established at the levels of model design, validation processes, and continuous auditing.
When AI Interviews AI: A Technical Circular Dilemma
What makes this "AI vs. AI" experiment noteworthy is that it reveals a technical dead loop.
Self-Referential Evaluation Criteria
The AI interviewer evaluates based on a preset scoring model, and if the AI job candidate is also a system optimized to pass AI interviews, then the interview essentially becomes a game between one model and another—both sides optimizing for the other's rules rather than genuinely demonstrating or assessing ability. This is strikingly similar to the principle behind Generative Adversarial Networks (GANs): two AI systems "deceiving" each other.
Generative Adversarial Networks (GANs) were proposed by Ian Goodfellow in 2014. The core architecture contains two neural networks: the Generator, which produces samples as realistic as possible from random noise, and the Discriminator, which distinguishes whether input samples are real or generated. The two compete during training: the Generator tries to "fool" the Discriminator, while the Discriminator continuously improves its detection ability, ultimately reaching a Nash equilibrium in the ideal state where the Generator's output is indistinguishable from real data. This GAN mechanism is highly isomorphic to the "AI interviewer vs. AI candidate" dilemma—both sides optimize for the other's rules rather than serving some external real objective. The difference is that the GAN game has a clear mathematical objective function (minimizing the divergence between the generated distribution and the real distribution), whereas AI recruiting has no such objective "true ability distribution" as an anchor, making this game more prone to degenerating into pure rule arbitrage.
Unverifiable "Authenticity"
In human interviews, interviewers can assess a candidate's true level by following up with questions, observing body language, and judging how natural the answers are. But when the interviewer itself is AI, these "meta-signals" cannot be captured. A carefully tuned AI candidate can pass all AI interview metrics while completely lacking actual work ability.
The Serious Questions Behind the Absurdity: The Boundaries of Automated Decision-Making
The value of this thought experiment lies not in its entertainment value, but in forcing us to think about several key questions:
The lack of explainability in AI decisions. If the AI interviewer rejects the AI candidate, who explains why? If neither can explain its own judgment logic, the entire process becomes a black box.
A vacuum of accountability. When AI recruiting leads to wrong decisions—whether missing out on excellent talent or hiring unqualified people—who should bear responsibility? The developers, the company, or the HR staff using the AI?
The irreplaceability of human judgment. Hiring is not just about skill matching; it also involves judging teamwork, growth potential, and cultural fit—precisely the areas where current AI is weakest.
A Warning for Practitioners: Don't Let the Tool Become the Goal
Although this Hacker News post contains limited information, the trend it points to is real. More and more companies are exploring AI-driven recruiting automation, and while the scenario of "using AI to interview AI" is currently a joke, it is not far from partial reality.
For tech practitioners, this absurd drama offers an opportunity for reflection: when designing AI systems, their capability boundaries and applicable scenarios must be clearly defined. AI can serve as an auxiliary tool to help humans filter information and improve efficiency, but completely handing over decision-making power to AI—especially when the evaluated party is also AI—will only produce meaningless loops.
Conclusion
When both sides of an interview are AI, what humans lose is not just the sense of participation, but also control over the direction of technological development. This short post speaks volumes in reminding us: technology should serve human goals, not let human goals disappear into the maze of technology.
Background Supplement
The accountability of automated decision-making legally involves the concept of a "multi-party responsibility chain." Article 22 of the EU's General Data Protection Regulation (GDPR) explicitly grants individuals the right "not to be subject to a decision based solely on automated processing, including profiling, which produces legal effects concerning him or her or similarly significantly affects him or her," and requires companies to provide channels for human intervention and objection. The EU AI Act passed in 2024 further classifies recruiting AI as a "high-risk" application, requiring risk management systems, technical documentation, human oversight, and transparency obligations. However, in practice, when companies procure third-party AI recruiting tools, responsibility often becomes blurred among the vendor, the purchaser, and the HR staff actually using the AI. If the model is a black box lacking auditable logs, aggrieved candidates have almost no way to prove their case. This institutional vacuum means that even if the technology itself can be improved, the lack of a clear responsibility framework will still leave victims of automated recruiting with nowhere to seek redress.
Related articles
Bilibili's Hit Dify Tutorial: A Practical Guide to Building 20+ AI Apps from Scratch
A distilled guide to Bilibili's top-rated Dify tutorial: Windows + Docker deployment, MySQL connection, five app types, model integration strategy, and publishing.
Agent Skills Beginner's Guide: From Concepts to Multi-Agent Collaboration in Practice
A comprehensive guide to Agent Skills: core concepts, components, the importance of Description, standardization, beginner learning paths, and multi-agent collaboration.
One Neuron Detects Both Cats and Cars? We Reproduced It and Found a New Explanation
The cat-car neuron reinterpreted: researcher Cammarata argues it distinguishes pointy from round, not cats from cars. We reproduced the experiment and found the new explanation holds on most inputs—but cars are the exception. The interpretability debate continues on language models.