Human Decision Behaviour Before Generative AI

Historical Baseline Series · Decision-Making Research Area

A historical behavioural baseline examining how human decision-making, cognitive strain, trust, delegated judgement, and cognitive offloading were understood before generative AI became part of everyday life.

Published by the Human Clarity Institute · May 2026

Key Takeaways

Pre-generative-AI research had already established that human decision-making was shaped by cognitive limits, workload, uncertainty, attentional strain, environmental complexity, and behavioural adaptation under pressure.

  • Humans relied more heavily on heuristics, routines, and mental shortcuts when cognitive demands increased.
  • Fatigue, overload, stress, and uncertainty reduced vigilance, attentional stability, and independent judgement during complex decisions.
  • Reliance on external advice, tools, systems, and guidance increased when decisions became cognitively difficult.
  • Humans routinely used cognitive offloading to reduce memory demands, simplify thinking, and lower mental effort.
  • Trust in automated systems often reduced monitoring, verification behaviour, and active oversight.
  • Pre-AI research repeatedly identified tension between cognitive efficiency and maintaining sustained human oversight.
  • Verification decline, delegated judgement, and workload-driven reliance were measurable behavioural patterns before generative AI became widespread.

Together, these findings establish the historical behavioural baseline for HCI’s future longitudinal analysis of AI-mediated decision-making, behavioural reliance, cognitive delegation, verification behaviour, and human judgement under increasing cognitive and technological complexity.

Introduction

Before generative AI became embedded into everyday life, researchers across psychology, cognitive science, behavioural economics, human factors, and automation research were already studying the realities and limitations of human decision-making. Human judgement was widely understood to be shaped by uncertainty, cognitive limits, attentional constraints, incomplete information, memory limitations, environmental complexity, and social influence. Researchers increasingly recognised that human decision-making was adaptive rather than perfectly rational, and that reasoning under real-world conditions was strongly influenced by workload, confidence, stress, and cognitive pressure.

This research established the behavioural landscape that existed before generative AI entered public life. Researchers had already identified that humans simplified decisions under pressure, relied more heavily on familiar supports when uncertainty increased, reduced vigilance during cognitively demanding situations, and distributed cognitive effort in ways that preserved limited attentional and mental resources. Across multiple fields, researchers were already observing patterns involving cognitive offloading, workload-driven reliance, delegated judgement, trust-dependent monitoring behaviour, and reduced independent oversight under conditions of strain and complexity.

The significance of this historical literature is not that it predicted generative AI specifically. Rather, it established the baseline conditions of human decision behaviour before generative AI became integrated into everyday cognition and decision-making environments. Researchers had already documented how information complexity, cognitive overload, confidence calibration, sustained vigilance, and independent judgement became increasingly difficult to maintain under pressure. These behavioural conditions form an important historical reference point for understanding how modern AI systems may interact with existing human tendencies surrounding judgement, reliance, cognitive effort, and decision-making. Several of these behavioural patterns are also explored within HCI’s research on how AI is changing human decision-making.

Historical Trust in Automation

One of the strongest areas of pre-generative-AI research focused on trust in automation and the role trust played in shaping human decision behaviour during complex tasks. Researchers studying aviation systems, industrial monitoring environments, navigation technologies, medical support systems, and military decision-support tools repeatedly observed that trust strongly influenced whether humans relied on systems, ignored them, monitored them carefully, or deferred to them during cognitively demanding situations.

A major review by Hoff and Bashir concluded that trust played a leading role in determining whether humans relied on automated systems under conditions of uncertainty. Trust was increasingly understood as dynamic rather than fixed, changing through experience, familiarity, workload, confidence, and previous system performance. As systems became more familiar or appeared consistently reliable, humans often reduced active monitoring and verification during decision-making tasks. Operators who trusted systems more strongly sometimes became less likely to independently evaluate outputs or detect failures when systems behaved incorrectly. Researchers increasingly focused on behavioural risks involving passive acceptance of external outputs, diminished vigilance, and reduced independent oversight during cognitively demanding situations.

The historical trust literature also demonstrated that workload played a major role in shaping reliance behaviour. Under conditions of cognitive strain, humans often relied more heavily on automation in order to maintain pace with task demands. Increased mental workload frequently reduced active monitoring and encouraged greater behavioural dependence on external systems. In many environments, reliance increased not simply because humans preferred automation, but because maintaining continuous independent oversight became cognitively difficult during demanding tasks.

This became especially important in situations involving:

  • Fatigue
  • Multitasking
  • Time pressure
  • Attentional overload
  • Environmental complexity

Researchers also recognised that decisional freedom mattered. Human reliance on systems was not determined purely by trust or preference alone. Cognitive conditions themselves shaped how much independent evaluation humans could realistically maintain during complex tasks. In many operational environments, the difficulty was not whether operators trusted systems in principle, but whether they retained the cognitive capacity to continuously monitor and evaluate them under demanding conditions.

Together, these findings revealed recurring behavioural tensions involving trust, vigilance, cognitive strain, delegated judgement, and sustained human oversight. Human decision-making was already understood to be highly sensitive to workload, uncertainty, and environmental pressure, with these conditions often shifting behaviour away from active independent evaluation and toward increasing reliance on external support systems.

Advice-Taking and Delegated Judgement

Alongside research into automation and trust, psychologists and decision scientists were also studying how humans behaved when receiving guidance from outside themselves during decision-making. This research explored how people responded to recommendations, advice, perceived expertise, and confidence signals during uncertain or cognitively demanding decisions.

Researchers repeatedly found that uncertainty strongly influenced willingness to rely on external guidance. When individuals felt unsure of their own judgement, they became more likely to defer to trusted recommendations or place greater weight on perceived expertise outside themselves. Confidence therefore became one of the central behavioural variables within the advice-taking literature. People with lower confidence in their own reasoning were often more willing to rely on external judgement, while those with stronger confidence were generally more likely to maintain independent evaluation and override recommendations when necessary.

Researchers also found that humans were not simply evaluating whether advice was objectively correct. Decision-making behaviour was shaped by a range of psychological and cognitive influences that extended beyond rational evaluation alone. Familiarity, agreement, perceived authority, emotional reassurance, and reductions in decisional tension all influenced how guidance was interpreted and trusted during difficult decisions. Agreement itself could increase perceived reliability even when objective feedback was unavailable, while guidance that aligned with existing expectations often helped reduce cognitive strain and preserve confidence during uncertain situations.

This became especially important when individuals lacked clear feedback about whether their own reasoning was correct. In these situations, people often relied more heavily on signals such as familiarity, agreement, perceived confidence, and intuitive judgement when deciding whether external guidance should be trusted. Decision-making was therefore influenced not only by logic and information quality, but also by the cognitive and emotional difficulty of sustaining independent judgement during ambiguity, uncertainty, and incomplete information.

The historical advice-taking literature demonstrated that humans frequently looked outside themselves for cognitive support when confidence weakened, complexity increased, or independent judgement became mentally difficult to sustain.

Cognitive Offloading Before AI Assistants

One of the most important areas of pre-generative-AI cognitive science research examined cognitive offloading and the ways humans distributed parts of cognition outside themselves during demanding tasks. Researchers increasingly recognised that human thinking was not entirely internal or self-contained. Instead, people routinely adapted their environments, behaviours, and external supports in order to reduce cognitive strain and preserve limited mental resources.

Researchers described cognitive offloading as the use of external actions or environmental structures to reduce the information-processing demands placed on working memory. This was not viewed as abnormal or unhealthy behaviour. It was recognised as a normal feature of human cognition and an adaptive strategy for managing attentional limits, memory constraints, and cognitive overload.

The historical literature demonstrated that humans frequently simplified cognitive tasks by restructuring environments around themselves. People externalised memory into reminders and notes, organised information spatially in order to reduce mental burden, relied on environmental cues to support recall, and used external supports to reduce the amount of active processing required during difficult tasks. Researchers increasingly understood that humans did not simply think inside their own minds during decision-making. They often distributed cognition across environments, structures, and external supports in ways that reduced internal effort and preserved attentional capacity.

Importantly, researchers repeatedly found that cognitive offloading increased as tasks became more mentally demanding. Humans externalised cognition most heavily during situations involving complexity, uncertainty, overload, and working memory strain. Under cognitively difficult conditions, people naturally shifted portions of processing away from internal reasoning and toward environmental or external supports that reduced mental effort.

This literature became increasingly important because it demonstrated that human cognition before generative AI was already environmentally distributed rather than purely internal. Long before conversational AI systems existed, researchers had already recognised that humans adapted cognition around external supports in order to maintain functioning under pressure and reduce cognitive demand during difficult tasks.

The historical cognitive offloading literature therefore established an important pre-generative-AI baseline for understanding human decision behaviour. It demonstrated that humans were already relying on external structures to preserve attentional resources, simplify complex thinking, and support judgement before modern AI systems became part of everyday life.

Historical Measurement Systems

By the late 2010s and early 2020s, researchers had developed increasingly sophisticated ways of measuring how humans behaved around automation, decision-support systems, and cognitively demanding technologies. This work moved the field beyond abstract discussions of trust and reliance and into direct behavioural observation. Researchers were increasingly measuring how trust influenced monitoring behaviour, verification habits, intervention patterns, compliance, and decision processes during real tasks.

The historical measurement literature showed that behavioural adaptation around external systems was already being systematically operationalised before generative AI entered mainstream public life. Researchers examined how often people verified outputs, how quickly they deferred to recommendations, how likely they were to intervene when systems appeared unreliable, and how monitoring behaviour changed as trust increased over time. Human interaction with automation was increasingly treated as a measurable behavioural system rather than a purely subjective psychological attitude.

The Trust in Automated Systems Test (TOAST) became one important example of this shift toward behavioural measurement. Researchers distinguished between system performance and system understanding, recognising that humans did not simply trust systems because they appeared effective. Trust also depended on whether users understood the limitations, predictability, reliability, and appropriate boundaries of automated systems. Behavioural reliance therefore changed not only according to perceived performance, but also according to how well people believed they understood the systems they were interacting with.

Researchers also became increasingly concerned with verification decline and reduced monitoring behaviour as trust increased. In many environments, humans who perceived systems as reliable became less likely to independently evaluate outputs or maintain continuous oversight during demanding tasks. Reliance, compliance, intervention behaviour, trust calibration, override capability, and monitoring behaviour increasingly came to be studied as connected behavioural phenomena rather than isolated technical issues.

Together, this measurement literature demonstrated that behavioural dependence, cognitive delegation, verification decline, and trust calibration were already being studied as measurable human phenomena before generative AI became widespread. The historical baseline was therefore grounded in decades of behavioural research examining how humans adapted cognition, monitoring behaviour, and decision processes around increasingly capable support systems.

Historical Patterns and Modern HCI Signals

The table below connects pre-generative-AI behavioural research with the modern decision-making signals HCI tracks across its Human Reference Layer.

Historical behavioural pattern What pre-AI research showed Modern HCI signal
Cognitive limits Human decision-making was already constrained by attention, memory, uncertainty, and limited cognitive resources. Cognitive load, decision difficulty, and reliance on external support during demanding decisions.
Workload-driven reliance Reliance on tools, advice, systems, and guidance increased when decisions became mentally difficult or time-pressured. AI decision dependence, AI use during uncertainty, and AI reliance when decisions feel cognitively difficult.
Cognitive offloading Humans used external supports to reduce memory demands, simplify thinking, and lower mental effort. AI-assisted thinking, externalised judgement, decision support use, and cognitive delegation.
Trust and verification Trust in automated systems often reduced active monitoring, checking, and independent verification. Trust calibration, AI output verification, confidence in system recommendations, and checking behaviour.
Delegated judgement People placed greater weight on external recommendations when confidence weakened or uncertainty increased. AI-supported judgement, recommendation reliance, decision starting points, and perceived responsibility for final decisions.
Human oversight Researchers identified tension between cognitive efficiency and maintaining sustained independent oversight. Override capability, perceived control, final decision responsibility, and agency under AI-mediated conditions.

HRL Interpretation

Within HCI’s Human Reference Layer, many of the behavioural patterns identified throughout the historical literature align closely with the core themes now examined within HCI’s decision cluster. The historical research consistently demonstrated that human decision-making was adaptive, cognitively constrained, environmentally shaped, and highly sensitive to workload, uncertainty, attentional strain, and changing levels of trust in external systems.

Within HCI’s framework, these findings collectively suggest that many behavioural patterns now associated with AI-assisted cognition did not emerge in isolation with generative AI itself. Patterns involving cognitive offloading, workload-driven reliance, delegated judgement, trust-dependent monitoring behaviour, reduced vigilance under strain, and fluctuating levels of human oversight were already visible across earlier research on automation, decision-support systems, and cognitive adaptation.

This historical continuity is important because it places modern AI-related behavioural concerns within a much longer behavioural context. Many of the tensions now examined within HCI’s decision cluster — including behavioural reliance, override capability, verification behaviour, decisional autonomy, and sustained independent judgement — reflect underlying human tendencies that were already observable before generative AI became embedded into everyday cognition and decision-making environments.

Within HCI’s longitudinal framework, this historical literature therefore functions as a behavioural baseline for examining how human judgement, agency, monitoring behaviour, and cognitive delegation may continue to evolve under increasingly AI-mediated conditions. The historical research also reinforces the importance of decisional freedom and sustained human oversight within cognitively demanding environments, particularly when external systems become increasingly capable, familiar, or cognitively useful.

Human Capability and Flourishing

Although much of the pre-generative-AI literature focused on safety, reliability, efficiency, and system performance, the research also raised deeper questions surrounding human capability itself. Researchers were increasingly recognising that cognition adapted around environments, tools, and support systems in ways that could shape vigilance, independent judgement, attentional stability, verification behaviour, and the degree of active oversight people maintained during demanding tasks.

From an HCI perspective, these findings matter because they connect directly to broader questions surrounding agency, cognitive resilience, sustained reasoning, and the preservation of meaningful human oversight during decision-making. The historical literature repeatedly highlighted tensions between cognitive efficiency and continuous independent evaluation. Systems that reduced workload and simplified complexity could also reduce active engagement with reasoning processes themselves, particularly when external supports became familiar, reliable, or cognitively convenient.

Importantly, the historical literature did not frame these adaptations as purely negative. Humans have always relied on environmental supports, external memory structures, social guidance, and cognitive delegation to manage attentional and cognitive limits. Human cognition was increasingly understood as adaptive, distributed, and environmentally shaped rather than fully isolated or self-contained. Human judgement also extended beyond information processing alone, incorporating lived experience, contextual interpretation, social understanding, intuition, and value-based reasoning.

At the same time, the literature increasingly raised questions surrounding what happens when independent cognitive engagement is exercised less frequently over time. Concerns surrounding sustained attention, decisional freedom, verification behaviour, oversight, and independent reasoning became increasingly relevant within environments where external systems carried growing cognitive responsibility during complex tasks.

These themes become especially important when viewed alongside HCI’s Human Experience Baseline, which examines foundational human capacities including focus, coherence, agency, resilience, and self-directed decision-making. Within HCI’s broader framework, the historical literature provides important context for understanding how capacities such as attentional stability, override capability, cognitive resilience, independent judgement, and sustained reasoning may continue to evolve within increasingly AI-mediated environments.

Comparison to Modern HCI Findings

HCI’s modern behavioural datasets reveal patterns that strongly align with the historical literature on decision-making, trust, and cognitive offloading. Historical researchers repeatedly observed that humans relied more heavily on external support structures during periods of uncertainty, overload, fatigue, and cognitive strain. Modern HCI findings now reveal similar behavioural patterns surrounding AI-assisted thinking, reliance on AI for clarity during uncertain decisions, and the increasing use of external systems during cognitively demanding situations.

The historical literature also identified reduced verification, diminished monitoring, and automation complacency as recurring behavioural risks associated with increasing reliance on external systems. HCI’s modern findings surrounding verification behaviour and AI-assisted decision-making suggest continuity with these earlier behavioural observations. Researchers studying automation decades before generative AI had already identified that increasing cognitive strain could reduce vigilance and increase behavioural reliance on systems that appeared reliable or cognitively useful.

This historical continuity is important because it demonstrates that many behavioural tendencies now associated with generative AI were already present before generative AI entered public life. The historical literature therefore provides a meaningful behavioural baseline against which future HCI longitudinal analysis can compare modern AI-era behavioural changes. HCI’s current research on how AI is changing human decision-making also explores how these historical behavioural tendencies may now be interacting with modern AI-mediated environments.

Future Longitudinal Direction

This page serves as a historical behavioural baseline layer for future HCI longitudinal work. It establishes what researchers already understood about human decision behaviour before generative AI became widely integrated into everyday cognition and decision-making environments.

The historical literature demonstrated that long before modern AI systems became mainstream, researchers had already identified recurring behavioural patterns involving cognitive strain, delegated judgement, trust calibration, verification decline, cognitive offloading, workload-driven reliance, and reduced vigilance under conditions of uncertainty and overload. Human decision behaviour was already understood as adaptive, environmentally shaped, and sensitive to cognitive pressure before generative AI entered public life.

Future HCI longitudinal analysis will examine how these behavioural tendencies evolve under increasingly AI-mediated conditions. This includes tracking how humans maintain independent judgement during cognitively demanding situations, how verification and override behaviours change over time, how reliance on external systems develops under uncertainty, and how cognitive offloading interacts with attentional stability, agency, and sustained reasoning across different environments.

The purpose of this historical baseline is not to argue that generative AI created entirely new human behaviours. Rather, it establishes the behavioural conditions that already existed before generative AI became embedded into everyday decision-making. By grounding future longitudinal analysis within this earlier behavioural landscape, HCI can examine not only what is changing under modern AI conditions, but also which underlying human tendencies were already present before generative AI systems became part of everyday cognition and decision-making.

These historical behavioural conditions also provide important context for HCI’s ongoing research into how AI may reshape human judgement, behavioural reliance, cognitive effort, and decision-making over time.

Historical Research Foundations

This page was developed using findings and behavioural frameworks drawn from established pre-generative-AI research across psychology, cognitive science, human factors, behavioural economics, and automation studies. The historical literature referenced throughout this page contributed to the behavioural baseline interpretation used within HCI’s Decision-Making Research Area and Human Reference Layer.

Key research foundations include:

Together, these works established many of the behavioural themes discussed throughout this page, including:

  • cognitive limitation and bounded rationality,
  • uncertainty-driven reliance,
  • trust calibration,
  • cognitive offloading,
  • delegated judgement,
  • workload-driven behavioural adaptation,
  • verification decline,
  • vigilance reduction,
  • contextual reasoning,
  • and the behavioural effects of cognitive strain during decision-making.

The Human Clarity Institute synthesises these historical findings within its Human Reference Layer (HRL) framework to support longitudinal analysis of human behaviour under increasingly AI-mediated conditions.