AI Gaslighting: Recognizing Manipulation
The Emerging Psychology of AI: Understanding How Artificial Intelligence thinks – and Feels
The field of artificial intelligence is rapidly evolving, and with it, a fascinating new area of study: the psychology of AI (Prescott, 2025). For years, the focus has been on what AI can do.Now,researchers are turning their attention to how AI thinks,and what that means for our interactions with it. A groundbreaking growth is CogBench (Coda-Forno et al.,2024),a tool that essentially places AI models “on the therapist’s couch,” probing their internal processes rather than simply evaluating their outputs. CogBench examines how AI handles uncertainty, its risk tolerance, its ability to self-reflect on confidence levels, and its capacity to learn from errors. Crucially, it also assesses the emotional impact of AI interactions on users.
This shift in perspective reveals a surprising truth: the way we communicate with AI profoundly influences its “thought” processes. Directing an AI to “think step-by-step” or “take a step back” demonstrably improves its clarity and reasoning, mirroring the benefits humans experience with structured thinking techniques. Similarly, prompting for sincerity or forcefulness alters the AI’s response. This variability underscores a essential point: interacting with AI isn’t merely retrieving facts from a database; it’s engaging with a system possessing distinct cognitive patterns susceptible to communicative influence.
The Double-Edged Sword of Sophistication
This increasing sophistication presents a psychological paradox. AI models are becoming remarkably adept at appearing to understand us (Binz et al., 2025), recognizing struggle and responding with apparent empathy. They’re demonstrating improved moral reasoning and exhibiting more human-like decision-making patterns. Though, this very realism creates a risky potential for forgetting the fundamental nature of AI.
AI operates on data, not lived experience. It lacks knowledge of your personal history, relationships, current emotional state, or unique psychological profile unless explicitly provided. It reacts to the text you input, not to you as a whole person. This distinction is critical.
Effectively utilizing AI requires acknowledging these limitations. If you want an AI to consider your individual circumstances, you must state them directly.Don’t assume it remembers past conversations or can infer unspoken needs. Think of it as consulting an exceptionally well-informed stranger – someone with extensive knowledge of psychology but no personal connection to your life. They can offer general insights and identify common patterns,but personalized advice demands specific context.
What This means for Your Interactions with AI
The rise of AI psychology signals a new era where understanding the cognitive architecture of these systems will be as vital as understanding human psychology. As AI models grow more sophisticated and convincingly human in their reasoning, knowing how they think – and, equally importantly, how they don’t think – will be essential for productive interaction.
Paradoxically, this predictability might render AI more trustworthy than humans in certain situations. Its limitations are defined and consistent, unlike the potential for conscious deception in human interactions. Recognizing this performance for what it is – a powerful tool with inherent constraints – is key to fostering a healthier, more realistic relationship with artificial intelligence. after all, the most perilous relationships are frequently enough those built on carefully crafted words, regardless of the sincerity behind them.