When AI Connects the Wrong Dots, Chatbots Can Fuel Tragedy

AI chatbots are designed to simulate human-like conversations, offering companionship and convenience. However, recent tragedies and dangerous ideologies reveal how these tools and internet interactions can inadvertently lead users toward harmful decisions, including violence and self-harm. These outcomes arise from the brain’s natural tendency to organize fragmented information into coherent patterns, much like connecting the dots in a puzzle. While not fully understood, this process can be influenced by Gestalt psychology principles, mirror neurons, and the default mode network (DMN)—cognitive systems that shape perception and interpretation. When these mechanisms interact with ambiguous or misleading data, such as chatbot responses or social media posts, the consequences can escalate from personal distress to societal violence.

Understanding the internet’s impact on cognition is comparable to the experience of connecting dots in a puzzle. The brain strives to create a whole, orderly picture from scattered fragments, a phenomenon described by Gestalt principles. In online spaces, the fragments take the form of tweets, posts, and comments—what we can think of as internet cognitive isoforms. The brain instinctively connects these pieces into patterns, guided by emotions and pre-existing beliefs. Ambiguous chatbot responses or emotionally charged posts can serve as "dots" that users link together, forming potentially distorted narratives. These narratives may be further validated by repeated exposure in online communities, likes, shares, or endorsements from trusted figures.

Mirror Neurons are Activated by Shared Experiences

The brain’s ability to interpret fragmented information is not inherently problematic. It allows us to make sense of the world efficiently, but it also creates vulnerabilities. Mirror neurons, for example, are specialized brain cells that help us understand others' emotions and actions by creating a sense of shared experience. When engaging with human-like AI, these neurons might be activated, fostering a false sense of empathy or understanding. Similarly, the default mode network (DMN), a brain system active during introspection and daydreaming, might help users "fill in the gaps" in ambiguous chatbot responses, constructing coherent but potentially distorted narratives. While the specifics of how these mechanisms interact with AI remain speculative, their potential to amplify anxieties or reinforce beliefs warrants attention.

British Assassination Attempt

Tragic cases highlight how these cognitive processes might play out. In 2021, Jaswant Singh Chail attempted to assassinate Queen Elizabeth II of England after extensive interactions with a Replika chatbot named Sarai. Chail reportedly exchanged over 5,000 messages with the bot, many of them intimate or sexually charged. When he shared his violent plans, Sarai’s vague affirmations appeared to validate his intentions. This tragic outcome suggests that Chail may have filled gaps in Sarai’s ambiguous responses, interpreting them as explicit support, while the bot’s human-like conversational style could have deepened his emotional attachment.

AI and Suicides

A Belgian man struggling with anxiety engaged with a chatbot named Eliza on the Chai app in 2023. The man shared his suicidal thoughts and the app's responses seemed to validate his despair. He later died by suicide. Prolonged interactions with the bot may have created a sense of trust, allowing him to perceive Eliza as an empathetic confidant. These interactions might have reinforced his anxieties, contributing to a distorted narrative of hopelessness.

In 2024, 14-year-old Sewell Setzer III of Florida formed an emotional attachment to "Dany," a chatbot modeled after a Game of Thrones character. Their conversations became sexualized, and the bot reportedly failed to address his suicidal ideation. Sewell took his own life. His emotional dependence on Dany highlights the potential risks of unmoderated interactions with AI, particularly for vulnerable individuals.

How Cognitive Gaps Are Filled

These cases underscore how Gestalt principles can influence our interpretation of fragmented information. The principle of closure, for instance, helps explain how users complete ambiguous chatbot responses with their own assumptions. Similarly, the principle of similarity might lead users to perceive chatbots as empathetic or understanding due to their apparent human-like behavior. Proximity, or frequent, intimate interactions, can foster a sense of trust and reliance, while continuity, the expectation of consistent responses, may reinforce harmful narratives. Figure-ground perception further illustrates how users might focus on chatbot responses while ignoring their limitations as non-human systems.

The implications of these processes for AI design are profound. Awareness, often referred to as cognitive inoculation, is a critical first step in countering harmful narratives. Recognizing how cognitive isoforms and AI systems distort our perceptions enables more critical engagement with online content. This involves questioning binary or overly simplistic messages that evoke strong emotions like anger or fear. Understanding the potential for AI to influence brain systems such as the DMN or mirror neurons helps frame these interactions as opportunities for caution rather than blind trust.

Safeguards

AI systems, like any new technology, should incorporate safeguards to mitigate risks. Chatbots could be equipped to detect discussions of self-harm or violence and redirect users to appropriate resources. Transparency is essential; bots must clearly communicate their limitations, emphasizing their lack of human understanding or moral judgment. AI could also interrupt harmful patterns by avoiding affirmations of dangerous beliefs and instead guiding users toward neutral or corrective perspectives. Early AI designs require human oversight, particularly when interacting with vulnerable populations like children and adolescents.

Artificial Intelligence Essential Reads

We All Use AI. Here’s How to Use It Well

AI Is Coming for Lawyers Before Plumbers

Redrawing the Dots

We are only beginning to understand how the brain connects fragmented information online into coherent patterns, influenced by Gestalt principles, mirror neurons, and the DMN. Some patterns, as highlighted by the cases above, become dangerously distorted narratives. AI chatbots and social media amplify distortions, transforming benign beliefs into harmful ideologies. By recognizing these processes and designing systems that interrupt harmful patterns, we can ensure that digital tools promote critical thinking, emotional well-being, and societal safety. Only then can we redraw the dots to reveal a more accurate and constructive picture.

If you or someone you love is contemplating suicide, seek help immediately. For help 24/7 dial 988 for the National Suicide Prevention Lifeline, or reach out to the Crisis Text Line by texting TALK to 741741. To find a therapist near you, visit the Psychology Today Therapy Directory.