Do AI Models Reason?

February 22, 2025


"Reasoning" in the context of AI has to be one of the most misused terms. That's because LLMs don't reason exactly like humans. They can infer incredibly fast, but they do not reason. Nor do they think. They loop until a condition is met. Nonetheless, because we don't have a universally agreed-upon set of definitions, you'll often hear descriptions of "LLMs' ability to reason."

Turing Post recently covered the subject with an article based on Charles Fadel's and Dr. Alexis Black's research paper, "Does Present-Day-GenAI Actually Reason?" Whether you prefer to say that AI models reason or infer, I've condensed the above article into a table format so you can quickly determine how well AI models perform when attempting to replicate human modes of thinking.

Based on extensive research in cognitive science, psychology, and artificial intelligence, human modes of thinking can be categorized into distinct cognitive strategies that form the foundation for understanding and evaluating AI reasoning capabilities.

Researchers can systematically assess how well artificial intelligence systems replicate or fall short of human-like reasoning by breaking down human cognition into these specific modes. This categorization provides a framework for understanding the current limitations of AI reasoning and potential paths forward.

Key Performance Categories:

Mode of Thinking Description & AI Performance
Abductive Thinking Description: Making the most plausible explanations from incomplete information.
AI Performance: Partial - Can generate plausible explanations but lacks intuition and creativity in forming genuine novel hypotheses.
Abstract Thinking Description: Working with concepts not directly observable.
AI Performance: Limited - While it can process abstract concepts, it lacks true understanding of their deeper meaning.
Analogical Thinking Description: Using comparisons between situations to gain insights.
AI Performance: Partial - Can recognize similarities but struggles to generalize effectively across different domains like humans do.
Analytical Thinking Description: Breaking down complex problems into components.
AI Performance: Strong - Good at systematic evaluation and breaking down problems into smaller parts.
Associative Thinking Description: Connecting ideas based on patterns and relationships.
AI Performance: Strong in pattern matching but lacks deeper understanding of meaningful connections.
Computational Thinking Description: Using algorithmic, structured approaches.
AI Performance: Excels - Highly effective at structured, algorithmic problem-solving.
Concrete Thinking Description: Processing information literally and specifically.
AI Performance: Strong - Good at processing explicit, literal information.
Convergent Thinking Description: Narrowing multiple possibilities to one correct solution.
AI Performance: Excels - Very effective at determining optimal solutions from a set of possibilities.
Creative Thinking Description: Generating original ideas and novel solutions.
AI Performance: Poor - Lacks true originality, mainly recombines existing ideas rather than creating genuinely new ones.
Critical Thinking Description: Analyzing information objectively for reasoned judgments.
AI Performance: Limited - Can analyze but lacks true objectivity and deeper understanding.
Deductive Thinking Description: Moving from general principles to specific conclusions.
AI Performance: Partial - Can apply formal logic but only when explicitly programmed.
Design Thinking Description: Iterative problem-solving considering user experience.
AI Performance: Limited - Can follow design principles but lacks true understanding of user needs.
Divergent Thinking Description: Exploring multiple possible solutions.
AI Performance: Partial - Can generate multiple solutions but often relies on trained patterns.
Emotional Thinking Description: Making judgments based on emotions rather than logic.
AI Performance: Fails - Cannot truly experience emotions or think emotionally.
Holistic Thinking Description: Understanding systems as interconnected wholes.
AI Performance: Poor - Does not integrate perspectives flexibly or understand true interconnections.
Inductive Thinking Description: Deriving general principles from specific observations.
AI Performance: Partial - Can generalize from data but struggles with nuance and bias.
Integrative Thinking Description: Synthesizing conflicting perspectives into innovative solutions.
AI Performance: Poor - Does not effectively synthesize conflicting perspectives.
Intuitive Thinking Description: Making rapid, unconscious judgments from experience.
AI Performance: Limited - Can make quick assessments but lacks true intuition based on experience.
Lateral Thinking Description: Finding unconventional problem-solving approaches.
AI Performance: Limited - Primarily relies on trained patterns rather than truly novel approaches.
Logical Thinking Description: Using structured, principle-based reasoning.
AI Performance: Strong - Good at following logical structures and rules.
Metacognitive Thinking Description: Reflecting on one's own thought processes.
AI Performance: Poor - Lacks self-awareness and true introspection.
Narrative Thinking Description: Creating meaning through storytelling.
AI Performance: Partial - Can generate narratives but lacks deeper understanding of meaning and context.
Pattern Recognition Description: Identifying recurring structures in information.
AI Performance: Excels - Very effective at identifying trends and relationships across large datasets.
Reflective Thinking Description: Analyzing past experiences for deeper understanding.
AI Performance: Poor - Lacks ability for true self-reflection and learning from experience.
Sequential Thinking Description: Processing information in a logical, step-by-step order.
AI Performance: Excels - Very good at processing information in structured, sequential ways.
Strategic Thinking Description: Planning with consideration of long-term consequences.
AI Performance: Partial - Can assist in planning but lacks true adaptability and long-term understanding.
Systemic Thinking Description: Understanding component interactions within a system.
AI Performance: Limited - Can model interactions but lacks deep understanding of complex system dynamics.
Temporal Thinking Description: Reasoning about sequences and relationships over time.
AI Performance: Partial - Can process temporal sequences but may struggle with complex temporal relationships.