The so-called Eliza effect – the tendency to attribute human characteristics to machines – is nothing new. Experiments with the chatbot ELIZA in the 1960s demonstrated how easily we are fooled by even simple algorithms. The program identified key words used by the participant and responded with phrases that deepened the conversation, such as, “What does this mean to you?” and “Tell me more about it.” This allowed ELIZA to conduct meaningful conversations in which participants felt heard, as if they were talking to a human.
Today’s chatbots are far more advanced. They write, speak, and even send videos featuring their avatars. Chatbots in the form of phone apps are currently the most popular, but AI holograms and robots also exist. The ways we can interact with artificial intelligence are multiplying. You can talk to a chatbot psychotherapist, astrologer, teacher, or friend. There is also a growing market for AI romance apps, where you can choose – or even design – a partner: from body type and personality to interests and conversational style.
Do you want your girlfriend to be a housewife or a clone of a famous actress? Do you dream of a relationship with a vampire? Nothing could be simpler – you can have any partner you can imagine! Yet the question remains: are these relationships just harmless fun?
Self-deception in relationships with AI – a harmless game?
The popularity of AI-girlfriend apps is on the rise. In February 2024, the Mozilla Foundation analyzed 11 AI apps simulating interpersonal relationships. Within a year, they were downloaded 100 million times on the Google Play Store alone. Unlike virtual assistants such as ChatGPT, social chatbots are designed to build relations, not perform specific tasks. “Relational AI” aims to meet the social, emotional, or erotic needs of its users.
Discussions about evaluating “relational AI” raise various concerns: protecting the privacy of romance app users, the risk of addiction or manipulation, and the protection of minors. The limits of replacing humans with artificial intelligence are also debated. Australian philosopher Robert Sparrow argues that, however much joy we may experience interacting with AI, it is ultimately morally deplorable because it is based on illusion. We convince ourselves that we’ve established a relationship, even though in reality we are alone.
But is self-deception in an emotional relation with AI actually morally deplorable? Dr. Emilia Kaczmarek, from the Department of Ethics at the Faculty of Philosophy, University of Warsaw, explores this question.
Relationships with AI – not always morally deplorable?
We need more empirical research to determine the extent to which people who enter deep emotional relationships with AI are self-deceived, and to what extent they are consciously engaging with fiction, similar to role-playing. Let us assume, however, that this is a form of self-deception in which our desires distort our perception of reality. For example, we do not want to be lonely, so we convince ourselves that we are not.
Even if we accept that feeling liked by a chatbot constitutes self-deception, this does not necessarily make it morally wrong. Dr. Emilia Kaczmarek argues that self-deception in human-AI relations can be morally problematic, but it is not always morally deplorable:
“I analyzed the reasons why it is worth striving for a true, rather than a false, view of the world and ourselves. Some of these reasons help us achieve morally important goals, while others are valuable in themselves – such as being honest with oneself or upholding the ideal of authenticity,” the researcher explains.
First, the obligation to be honest with oneself is crucial for fulfilling other moral obligations, such as avoiding harm to others. For example, wishful thinking can prevent us from considering the negative consequences of our actions.
The ultimate moral evaluation of self-deception in a relationship with AI does not depend solely on the fact of being deceived. The purpose, consequences, and degree of awareness with which we approach the digital simulation also matter. From this perspective, forming such relationships does not have to be morally reprehensible – but it can be problematic. Therefore, we should be conscious of the needs we aim to satisfy when engaging in emotional interactions with artificial intelligence.
Second, the potential moral concern of self-deception should be proportional to the person’s degree of autonomy, as well as their cognitive, social, and emotional capacities. Self-deception is not morally blameworthy when it is beyond one’s control. It becomes particularly concerning when it results from conscious negligence.
Third, motivation and consequences matter. The goals and outcomes of entering into a relationship with AI can vary, and so can their moral evaluation. Do we choose a safe relationship with a chatbot because someone has wronged us? Or because we expect our partner to admire us unconditionally? Morally significant is also which personal traits and predispositions are reinforced or suppressed in a relationship with AI. How might succumbing to the illusion of being adored by AI affect our relationships with real people, whose recognition may be harder to earn?
Fourth, we should not morally condemn others for indulging in comforting illusions if it causes no harm. Someone who consciously overlooks facts to avoid emotional pain does not necessarily deserve moral criticism. We can demand honesty from ourselves, but do we have the right to demand it from others?
“Harmless self-deception in emotional relationships with AI can be considered morally questionable, although it will very rarely warrant moral condemnation. Such self-deception may be considered a violation of the prima facie obligation (editor’s note: ‘at first glance;’ this means that other obligation may take precedence) to strive for an accurate understanding of oneself and the world around us,” argues Dr. Kaczmarek.
Ultimately, the moral assessment of self-deception in a relationship with AI does not depend solely on the fact of being deceived. The purpose, consequences, and degree of awareness with which we engage in the digital simulation are also crucial. From this perspective, forming such relationships does not have to be morally deplorable, but it can be ethically complex. Therefore, we should remain mindful of the needs we aim to satisfy when engaging in emotional interactions with artificial intelligence.
The article was originally published in Polish on the Serwis Naukowy UW website on April 9, 2025.
