Thinking Machines by Luke Dormehl: Study & Analysis Guide

Understanding the history of artificial intelligence is not merely an academic exercise; it is essential for navigating today’s polarized debates about existential risks, job displacement, and superintelligent futures. In Thinking Machines, Luke Dormehl provides an accessible chronicle of AI’s turbulent journey, revealing a field defined by recurring cycles of extravagant promise, profound disappointment, and unexpected rebirth. This guide analyzes Dormehl’s historical narrative to extract crucial lessons about how technological progress actually happens, empowering you to critically assess contemporary claims against the sobering—and often surprising—backdrop of the past.

From Optimistic Birth to the First AI Winters

Dormehl’s story begins with the foundational intellectual work that made AI conceivable. He places central importance on Alan Turing, whose 1950 paper introduced the Turing Test as a conceptual benchmark for machine intelligence and fundamentally questioned the boundaries between human and machine cognition. This theoretical groundwork culminated in the pivotal 1956 Dartmouth Conference, where the term "artificial intelligence" was officially coined. Attendees, including pioneers like John McCarthy and Marvin Minsky, were buoyed by immense optimism, believing that core human intellectual capabilities could be replicated in machines within a generation.

This initial wave of optimism, however, collided with the harsh reality of technological limitations. Early research focused on symbolic AI or good old-fashioned AI (GOFAI), which attempted to encode intelligence through explicit rules and logic. Programs could solve narrowly defined problems like logic puzzles, but they failed catastrophically when faced with the messy, unstructured real world. For instance, early machine translation efforts, which attempted direct word-for-word substitution, produced famously nonsensical results. This gap between promise and delivery led directly to the first AI winter of the 1970s, a period of severe funding cuts and disillusionment where the field’s grand ambitions were seen as a failure. Dormehl highlights how this pattern—a surge of hype fueled by theoretical breakthroughs, followed by a crash when practical applications fall short—becomes a central rhythm of AI history.

The Institutional Rollercoaster and the Rise of Connectionism

A key strength of Dormehl’s analysis is his attention to the institutional and cultural forces shaping AI’s path. The cycles of boom and bust were not just about technology but about money, politics, and public perception. The revival in the 1980s, for instance, was heavily driven by the commercial promise of expert systems. These were rule-based programs designed to emulate the decision-making of human experts in fields like medicine or geology. Corporations and governments invested heavily, believing they had found a commercially viable form of AI. However, expert systems proved brittle, expensive to maintain, and incapable of learning. Their limitations, coupled with the fading of another overhyped paradigm (Japan’s "Fifth Generation" computer project), triggered the second AI winter in the late 1980s and early 1990s.

Parallel to the rise and fall of symbolic approaches, Dormehl traces the enduring, if marginalized, alternative: connectionism. This approach, inspired by the neural networks of the brain, argued that intelligence emerges from simpler, interconnected units. While overshadowed by GOFAI for decades due to computational constraints and theoretical critiques (like Minsky and Papert’s analysis of perceptrons), connectionism persisted. Its proponents continued refining the concepts of neural networks with multiple layers. The lesson here is that progress is not linear; sidelined ideas can re-emerge triumphantly when the enabling conditions—in this case, vastly more powerful computers and massive datasets—finally align.

Breakthroughs by Accident and the Unpredictability of Progress

Perhaps the most compelling theme in Thinking Machines is that AI’s most transformative breakthroughs often arrived in unexpected ways, solving problems they were not originally designed to address. The eventual triumph of deep learning—a modern incarnation of multi-layered neural networks—exemplifies this. It was not a direct, funded pursuit of general intelligence. Instead, key advances came from researchers applying these models to specific, data-rich domains like image recognition. The 2012 victory of a deep neural network (AlexNet) in the ImageNet competition was a watershed moment, dramatically exceeding expectations and catalyzing the current AI boom.

This pattern repeats throughout the history Dormehl recounts. The pursuit of general human-like intelligence (AGI) repeatedly stalled, while focused applications leveraging statistical power and pattern recognition succeeded spectacularly. Modern AI applications in recommendation systems, language models, and autonomous vehicles are offspring of this pragmatic, engineering-driven thread. Dormehl’s narrative suggests that the field advances not through top-down, conscious design of a mind, but through bottom-up, often accidental, discoveries of what immense computational scale applied to vast data can achieve. This should inform your view of today’s claims: the path to advanced capabilities is likely to be non-linear and unpredictable.

Critical Perspectives: History’s Lessons for Today’s AI Landscape

Evaluating Dormehl’s historical narrative against the current landscape reveals critical lenses for assessment. His chronicle of recurring AI hype cycles is a direct warning for evaluating contemporary claims about artificial general intelligence (AGI). Each previous cycle was characterized by experts making confident, timeline-specific predictions about human-level machine intelligence—predictions that consistently proved wildly optimistic. When you encounter similar predictions today, Dormehl’s history urges skepticism toward the timelines while taking seriously the underlying technological momentum.

Furthermore, the book illuminates the persistent mismatch between technological capability and societal adaptation. Each breakthrough phase—from expert systems to deep learning—has triggered debates about societal impact, ethics, and job displacement that the field itself was unprepared to address. The current debates about bias in algorithms, mass surveillance, and economic disruption are not new in kind; they are historical echoes at a larger scale. Dormehl’s work implies that understanding AI’s trajectory requires studying not just algorithms, but the economic, political, and cultural systems that fund, regulate, and are reshaped by them.

Finally, the history of AI winters teaches that progress is fragile and contingent on sustained investment. Winters were caused by a loss of faith when overpromising met underdelivering. The current boom is fueled by unprecedented commercial investment and tangible products. A critical question is whether this creates a more resilient ecosystem or sets the stage for a more catastrophic winter if commercial returns plateau. The lesson is to look for sustainable drivers of progress beyond mere hype.

Summary

• AI progresses in non-linear cycles: The field’s history is a repeated pattern of intense optimism (boom), followed by disillusionment when overpromises meet technical limits (bust/winter), leading eventually to unexpected breakthroughs that reignite progress.
• Breakthroughs are often accidental and applied: Major advances, like deep learning, frequently come from pursuing narrow, practical applications with new tools (e.g., greater compute and data), not from a direct, top-down pursuit of general human-like intelligence.
• Institutional and cultural forces are decisive: Funding, corporate interests, and public perception have repeatedly shaped AI’s direction as much as pure scientific discovery, from the Dartmouth Conference to expert systems to today’s corporate AI labs.
• Hype is a historical constant, and a warning: Grandiose predictions about achieving AGI within specific decades have a perfect track record of being wrong. Historical awareness cultivates healthy skepticism toward absolutist claims about timelines.
• Societal questions lag behind technical ones: Every upswing in AI capability has forced a rushed confrontation with ethical, economic, and philosophical questions. Today’s debates about bias, control, and impact are part of this long-standing pattern.