BEYOND ACCURACY
Ethical challenges of artificial intelligence in cost estimation
Artificial intelligence (AI) is increasingly used to support cost and carbon estimation across infrastructure, defence and major capital programmes, with evaluation typically focused on predictive reliability and efficiency. This paper argues that such a focus is insufficient, because cost and carbon estimates function as decision-shaping socio-technical artefacts, rather than neutral technical outputs. Drawing on AI ethics scholarship, professional guidance from the Association for Project Management and governance expectations set out in the UK Government AI playbook, the paper examines how AI restructures accountability, authority and trust within cost and carbon estimation processes. It identifies key ethical risk areas, including automation bias, reinforcement of historical bias, opacity and auditability challenges, dilution of professional responsibility and the misalignment between numerical precision and ethical decision quality. The paper concludes, by outlining implications for cost and carbon engineering governance and proposing principles for the responsible and ethically informed use of AI in cost estimation that extend beyond performance metrics.
Cost and carbon estimation plays a central role in the planning, approval and delivery of projects and programmes. Estimates influence investment decisions, procurement strategies, risk allocation and long-term performance assessment. In public and regulated sectors, they also shape political accountability and public trust. Despite this, cost and carbon estimation is often framed as a technical activity whose primary objective is numerical accuracy.
Robustness alone is an insufficient lens through which to evaluate the impact of AI on cost estimation.
Recent advances in AI have accelerated the adoption of data-driven approaches to cost and carbon estimation. Machine learning models are increasingly used to analyse historical project data, generate forecasts and identify patterns that promise greater consistency and speed, than traditional expert-led methods. These systems are frequently positioned as objective improvements, that reduce human bias and subjectivity.
However, robustness alone is an insufficient lens through which to evaluate the impact of AI on cost estimation. Cost and carbon estimates do not merely describe expected outcomes. They actively shape decisions, behaviours and incentives within organisations. When AI systems are introduced into this context, they do not simply augment calculation. They redistribute authority, alter perceptions of objectivity and change how responsibility is understood and exercised. These shifts raise ethical questions that are not captured by conventional performance metrics.
Professional guidance, such as the APM ACostE (now CaSA) Estimating Guide, emphasises transparency of assumptions, treatment of uncertainty and the exercise of professional judgement. These principles implicitly assume a human estimator, who can explain, challenge and revise an estimate. At the same time, the UK Government guidance on AI adoption frames AI as a high-impact decision support capability, that still requires clear accountability, auditability and contestability, where it informs consequential financial decisions. Cost and carbon estimation for major programmes sits at the intersection of these professional and governance expectations.
This paper addresses this gap, by examining the ethical challenges associated with the use of AI in cost and carbon estimation. It adopts a conceptual and practice-informed approach, that treats cost estimation as a socio-technical system. The paper asks the following question: What ethical risks arise when AI is used to generate, or influence cost and carbon estimates and how do these risks differ from those associated with non-AI enhanced estimation methods?
AI in cost and carbon estimation typically refers to narrow systems designed to support specific tasks, rather than general intelligence. Common applications include parametric models, using machine learning, regression and ensemble methods, trained on historical project data, natural language processing applied to cost and carbon breakdown structures and probabilistic forecasting tools that simulate cost and carbon uncertainty.
These approaches differ from traditional estimating techniques in several important ways. Traditional models are often assumption-driven, transparent in structure and reliant on professional judgement. AI-based approaches prioritise pattern recognition and optimisation based on statistical relationships within data. While such systems can outperform human estimators on narrowly defined accuracy metrics, they frequently encode assumptions, implicitly within data selection, feature engineering and model architecture.
This distinction matters ethically. Robustness does not guarantee appropriateness. An estimate may be numerically robust, while still being misleading, over-trusted, or misused. The authority granted to AI outputs can alter how uncertainty is communicated and how challenge is exercised within decision-making processes. As a result, AI should be understood not simply as a computational enhancement, but as an intervention that reshapes estimation practice.
AI ethics scholarship highlights recurring concerns related to delegation, responsibility and opacity when decision authority is shared between humans and machines. These concerns are particularly salient in cost and carbon estimation, where professional standards already embed ethical expectations around transparency, judgement and accountability.
The authority granted to AI outputs can alter how uncertainty is communicated and how challenge is exercised within decisionmaking processes.
Delegation of estimation tasks to AI systems complicates notions of professional responsibility. While AI tools are typically described as decision support, their outputs may carry significant authority in governance forums. This creates a risk that responsibility for estimates becomes distributed across model developers, estimators and decision makers in ways that are poorly defined.
The UK Government AI playbook emphasises that accountability for decisions informed by AI cannot be delegated to the system itself. Responsibility remains with human decision makers and organisations. In cost and carbon estimation, this principle aligns with professional expectations that estimators remain accountable for the estimates they present. AI, therefore, challenges existing ethical assumptions not by introducing new values, but by destabilising how those values are enacted in practice.
Cost and carbon estimation should, therefore, be understood as a socio-technical system in which models, data, organisational incentives and professional judgement interact over time. Ethical risk emerges from these interactions, rather than from individual technical components.
Automation bias and deference
AI-generated estimates may be perceived as objective and authoritative, leading decision makers to defer to model outputs, even when uncertainty is high. This risk is well recognised in AI governance guidance and directly conflicts with professional expectations that estimates should be challenged and interrogated.
Where AI outputs are presented without clear articulation of assumptions and limitations, challenge may be discouraged, rather than enabled. Automation bias, therefore, represents not a failure of robustness, but a failure of ethical decision support.
Reinforcement of historical bias
AI systems trained on historical cost and carbon data risk embedding optimism bias and structural underestimation into future estimates. Past practices become normative baselines. Rather than correcting bias, AI may normalise it at scale. This creates an ethical risk that is difficult to detect, because models may appear internally consistent, while reproducing systemic error.
Opacity and auditability
Professional cost and carbon estimating guidance emphasises transparency of assumptions and traceability of reasoning. Many AI models lack a clear line of sight between inputs and outputs, making meaningful audit challenging. Post-hoc explanations may provide comfort, without enabling genuine scrutiny. This undermines both professional accountability and governance assurance.
Dilution of professional responsibility
Traditional cost and carbon estimation places responsibility on the estimator. AI-assisted estimation blurs ownership of the estimate across multiple roles. When outcomes diverge from forecasts, it may be unclear who is accountable. This dilution of responsibility conflicts with both professional standards and public sector governance expectations.
Precision without ethical improvement
AI can improve numerical precision, without improving honesty, integrity or decision quality. Increased precision may obscure uncertainty, rather than clarify it. Ethical cost and carbon estimation requires not only accurate numbers, but truthful communication of confidence and risk.
Implications for cost engineering practice
The ethical risks identified challenge existing cost and carbon estimation governance frameworks. Professional standards assume explainability, judgement and accountability. AI adoption requires additional safeguards, to ensure these principles remain meaningful, rather than symbolic.
Transparency of assumptions, treatment of uncertainty and the exercise of professional judgement are not new ethical demands imposed by AI.
Human oversight alone is insufficient if reviewers lack the authority or capability to challenge AI outputs. Governance arrangements must, therefore, address skills, incentives and role clarity. AI-assisted cost estimation should be treated as a high-impact decision support use case requiring proportionate assurance.
Ethically responsible use of AI in cost and carbon estimation should emphasise contestability, traceability and clear ownership. AI systems should support professional judgement, rather than replace it. Organisations should define who owns AI-informed estimates and how uncertainty is communicated to decision makers.
The UK Government AI playbook highlights the importance of capability as well as control. Cost engineers working with AI require not only technical skills, but ethical literacy to understand how models shape decisions. Ethical governance must, therefore, extend beyond model validation, to include professional development and organisational culture.
This paper has argued that the ethical challenges associated with AI in cost estimation extend well beyond questions of predictive accuracy. By examining AI-assisted cost and carbon estimation through the lenses of professional practice and public sector governance, the analysis shows that AI acts less as a neutral technical enhancement and more as a force that reshapes authority, accountability and trust within decision-making systems.
A central insight of the paper is that many ethical principles relevant to AI-assisted cost and carbon estimation already exist within established professional guidance. Transparency of assumptions, treatment of uncertainty and the exercise of professional judgement are not new ethical demands imposed by AI. Rather, they are long-standing expectations embedded within cost and carbon estimating practice. What AI does, is expose the fragility of these principles when estimation is partially delegated to systems, whose internal reasoning may be opaque and whose outputs can carry disproportionate authority.
The ethical risks identified in this paper should, therefore, not be interpreted as failures of individual technologies. They arise from misalignment between technical capability and governance maturity. Automation bias, historical bias reinforcement and responsibility dilution are not inherent properties of machine learning models. They emerge when organisational incentives, assurance processes and professional roles fail to adapt to the changed epistemic status of AI-generated estimates.
In this sense, AI functions as a stress test of existing estimation ethics, rather than a novel ethical problem in its own right. This perspective has important implications for how AI adoption in cost and carbon estimation is framed. Much current discourse implicitly assumes that ethical risk can be mitigated through improved explainability, validation or model performance.
While these measures are necessary, the analysis suggests they are insufficient on their own. Ethical risk persists even where models are technically sound, if governance arrangements allow AI outputs to substitute for professional judgement, rather than support it. Oversight that lacks authority or understanding risks becoming performative rather than protective.
The discussion also highlights a critical tension between numerical precision and ethical decision quality. AI can increase the apparent precision of estimates, while simultaneously reducing the visibility of uncertainty and contestability. In high-stakes environments, this can lead to ethically problematic outcomes, where confidence in numbers replaces meaningful scrutiny of assumptions. Ethical cost and carbon estimation, therefore, requires deliberate design choices, that prioritise truthful communication of uncertainty over apparent exactness.
For major programmes and long-lived infrastructure investments, these issues are particularly acute. Cost and carbon estimates establish baselines that shape future decisions and institutional memory. When AI-driven estimates become embedded within organisational processes, ethical risks compound over time. Biases can become normalised, accountability can diffuse across roles and trust can shift from professional judgement, to model outputs, in ways that are difficult to reverse. This temporal dimension reinforces the need to treat AI-assisted cost and carbon estimation as a governance challenge, rather than a one-off technical deployment.
Taken together, the discussion underscores that responsible use of AI in cost and carbon estimation cannot be achieved through technical fixes alone. It requires explicit alignment between professional ethics, organisational governance and AI system design. Without this alignment, AI risks amplifying existing weaknesses in cost and carbon estimation practice, while giving them the appearance of objectivity and rigour.
By framing AI-assisted cost and carbon estimation as a socio-technical ethical system, this paper contributes a perspective that moves beyond performance metrics and tool evaluation. It invites cost engineers, decision makers and policymakers to reconsider not only how estimates are produced, but how they are trusted, challenged and owned in an era of increasingly algorithmic decision support.
The UK government AI playbook highlights the importance of capability as well as control.
This paper is conceptual and does not present empirical case studies. Future research should examine real-world deployment of AI in cost and carbon estimation and explore how ethical risks manifest across different organisational contexts and programme scales.
AI ethics in cost and carbon estimation is not primarily about robustness. It is about how authority, accountability and trust are redistributed when estimation is partially delegated to machines. Existing professional guidance already embeds ethical principles. AI makes these principles visible, fragile and in need of reinforcement. Ethical governance must, therefore, be designed into AI-assisted cost and carbon estimation, rather than applied after failure.
In recognition of the growing role of AI across cost and carbon estimation practice, readers are directed to the AI policy published by CICES, which sets out principles for responsible adoption, professional competence and accountability in the use of AI-enabled tools.
While this paper has focused on ethical risks and governance considerations, the Institute’s policy provides practical guidance for practitioners seeking to implement AI in a manner that remains consistent with professional standards and public trust. Together, these perspectives highlight the importance of aligning technical innovation with professional responsibility, as AI becomes embedded within estimation practice.