Chapter 3: Further Reading — Mental Models, Cognitive Science, and AI Collaboration

The resources below extend the frameworks from Chapter 3 across three areas: the cognitive science of mental models and how they shape behavior, the specific mental models that have been proposed for AI collaboration, and the practical research on what distinguishes effective from ineffective AI users.

The Cognitive Science of Mental Models

"Mental Models" — Philip Johnson-Laird (1983) The foundational academic work that established the concept of mental models in cognitive science. Johnson-Laird argued that human reasoning operates through the construction and manipulation of internal models of the world rather than through formal logical rules. Dense but influential — the introduction and opening chapters are accessible and foundational for understanding why mental models matter for all cognitive performance, including AI collaboration. Level: Academic | Johnson-Laird, P. N. (1983). Mental models. Harvard University Press.

"The Design of Everyday Things" — Don Norman (1988, revised 2013) Norman's chapter on mental models is one of the most accessible treatments of the concept in print. He distinguishes between the designer's model (how the system actually works), the user's model (the user's beliefs about how it works), and the system image (what the system communicates to the user). The gap between these is the source of most usability problems — and the same framework applies directly to AI tool design and use. Level: Accessible

"Thinking, Fast and Slow" — Daniel Kahneman (2011) Kahneman's account of System 1 and System 2 thinking is directly relevant to how mental models operate. System 1 thinking (fast, automatic, pattern-based) is where mental models live — they are the cognitive shortcuts that produce automatic judgments and expectations. System 2 thinking (slow, deliberate, analytical) is what the Model Diagnostic exercises draw on. Understanding this distinction helps explain why deliberately updating mental models is hard: you are asking System 2 to correct the automatic outputs of System 1. Level: Accessible

"Mindware: Tools for Smart Thinking" — Richard Nisbett (2015) A practical book on the scientific thinking tools — statistical reasoning, cost-benefit analysis, counterfactual thinking — that improve decision quality. Nisbett's core argument is that the right mental tools produce better thinking regardless of intelligence, and that these tools can be learned. The book's framework maps onto AI collaboration: having the right mental tools for interacting with AI systems is the prerequisite for effective use, not just technical sophistication. Level: Accessible

Expert Performance and Mental Models

"The Cambridge Handbook of Expertise and Expert Performance" — Ericsson et al. (2006) A comprehensive academic resource on what distinguishes expert from novice performance across domains. The chapters on mental representations are particularly relevant: a consistent finding is that experts differ from novices primarily in the richness, accuracy, and organization of their mental models of the domain — not primarily in factual knowledge or deliberate strategy application. This research grounds the claim in Chapter 3 that mental model quality is the key differentiator in AI collaboration skill. Level: Academic

"Sources of Power: How People Make Decisions" — Gary Klein (1998) Klein's research on decision-making under pressure (firefighters, military commanders, emergency responders) found that experts rarely use formal decision procedures — they recognize patterns and run mental simulations based on their experience-derived mental models. The "recognition-primed decision" model he develops has direct implications for how AI collaboration expertise develops: through experience that builds accurate models, not through the accumulation of formal rules. Level: Accessible–Academic

Mental Models Specific to AI and Technology

"Human-Computer Interaction" — Dix, Finlay, Abowd, and Beale (various editions) The standard HCI textbook covers mental models extensively in the context of interface design. Chapter treatments of how users build mental models of interactive systems and how those models affect usage patterns provide useful grounding for understanding AI-specific mental model development. Level: Technical–Accessible

"Robo-Advisors, Artificial Intelligence, and the Future of Human Collaboration" — MIT Sloan Management Review An ongoing stream of research articles examining how organizational contexts shape human-AI collaboration. Several articles directly address the mental models held by professionals working with AI systems in business contexts, including how initial framing affects adoption and use quality. Available via MIT SMR subscription or university library access. Level: Accessible

"People and AI Research (PAIR) Guidebook" — Google A practical guide to human-centered AI product development that includes substantial discussion of how users form mental models of AI systems and how designers can support more accurate model formation. Though aimed at product designers, the framework is useful for users thinking about their own model-building process. Level: Accessible | Available at: pair.withgoogle.com

AI-Specific Collaboration Research

"How Does Human-AI Collaboration Affect Creative Output?" — various authors A growing body of research examining creativity in human-AI collaboration across domains — writing, design, music. The consistent finding is that the quality of the collaboration is mediated by how humans conceptualize their role relative to the AI: users who see themselves as directors and evaluators (consistent with the first draft and amplifier models) produce better work than users who see the AI as an autonomous creator. Accessible via Google Scholar using search terms "human-AI collaboration creativity"

"Measuring Human-AI Partnership Quality" — Harvard Business Review HBR has published several accessible research summaries on effective human-AI collaboration in professional settings. Articles from 2022–2024 are most current for the current generation of generative AI tools. Themes that recur: the importance of user expertise in evaluating AI output, the cost of replacing rather than augmenting human judgment, and the organizational conditions that enable effective AI use. Level: Accessible | Available at: hbr.org

"Co-Intelligence: Living and Working with AI" — Ethan Mollick (2024) Mollick, a Wharton professor who has studied AI's impact on knowledge work extensively, offers a framework for AI collaboration that aligns closely with the productive models in Chapter 3. His distinction between using AI as a "brilliant but junior employee" versus using it as an autonomous decision-maker is directly parallel to the brilliant intern versus replacement model distinction. Particularly good on the practical implications for knowledge workers across different domains. Level: Accessible

Calibration and Trust in Human-AI Systems

"Trust in Automation: Designing for Appropriate Reliance" — Parasuraman and Riley (1997) A foundational paper in the human factors literature on trust calibration in automated systems. The concept of "appropriate reliance" — trusting automation in situations where it is reliable and maintaining human oversight in situations where it is not — is central to effective AI use. The paper's framework predates current AI tools but applies directly to the pattern matcher model and calibrated trust practices. Level: Technical

"Appropriate Reliance on AI Assistants: The Role of Mental Models" — various authors Recent research examining how users' mental models of AI capabilities affect their reliance calibration — whether they over-trust, under-trust, or appropriately trust AI suggestions in different contexts. Consistently finds that users with more accurate mental models of AI limitations have better-calibrated trust and better outcomes. Search Google Scholar for recent publications on this topic. Level: Technical–Accessible

On the "Brilliant Intern" Concept

"Turn the Ship Around!" — L. David Marquet (2013) Marquet's account of transforming a submarine crew from a leader-follower model to a leader-leader model is, unexpectedly, one of the best frameworks for thinking about high-leverage human direction of capable agents. His core practice — giving intent rather than instructions, making context explicit, and building the judgment of others rather than executing decisions for them — maps directly onto the brilliant intern model of AI collaboration. If you are going to internalize one book about directing capable agents effectively, this is an excellent choice. Level: Accessible

"The First 90 Days" — Michael Watkins (2003) A management book about executive transitions that contains excellent frameworks for explicit context-setting with new team members — exactly the kind of context-provision the brilliant intern model requires. The framework for understanding organizational context, identifying constraints, and communicating priorities is directly transferable to the practice of briefing an AI tool on a new project. Level: Accessible