Insight最新文献

{"title":"Trustworthy AI for Quantitative Systems Engineering via Deterministic Delegation to Verified Tools","authors":"Raymond Madachy","doi":"10.1002/inst.70056","DOIUrl":"https://doi.org/10.1002/inst.70056","url":null,"abstract":"<div>\u0000 \u0000 <p>Generative AI offers substantial capability in systems engineering processes for multimodal interaction supporting elicitation, interpretation, and synthesis. However, large language models (LLMs) remain unreliable for exact quantitative computation as they may produce incorrect or hallucinated results when executing mathematical and domain-specific quantitative calculations. Systems engineering is pervasively grounded in quantitative models that demand determinism, correctness, and traceable results, requirements that probabilistic language models cannot reliably satisfy on their own. A deterministic delegation pattern is presented that addresses this limitation by constraining the LLM to input interpretation, assumption documentation, and results presentation while assigning all authoritative computation to verified external tools through formally defined API endpoints. Identical inputs produce identical results, every output carries full provenance, and users can independently inspect or rerun any analysis via a returned source URL. The pattern is demonstrated through application domains for systems engineering cost modeling using the Constructive Systems Engineering Cost Model (COSYSMO) and system reliability analysis using fault tree and reliability block diagram methods. Examples at increasing levels of complexity show how the assistants translate diverse user inputs into complete, reviewable input configurations before invoking the deterministic backend, combining the accessibility of conversational AI with the correctness and auditability that engineering decisions require.</p>\u0000 </div>","PeriodicalId":13956,"journal":{"name":"Insight","volume":"29 3","pages":"58-66"},"PeriodicalIF":0.8,"publicationDate":"2026-07-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148386404","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Challenges and Opportunities for Systems Engineering in the Age of AI: SERC Perspective","authors":"Zoe Szajnfarber,&nbsp;Tom McDermott,&nbsp;Valerie Sitterle","doi":"10.1002/inst.70063","DOIUrl":"https://doi.org/10.1002/inst.70063","url":null,"abstract":"<div>\u0000 \u0000 <p>This perspective summarizes recent work across the SERC network in terms of five major buckets: 1) AI4SE – AI tools developed to support systems engineering work; 2) SE4AI — leveraging systems principles to assure AI-enabled systems; 3) feedback to AI4SE – where system architecting principles are leveraged to re-map what systems engineers do as AISE progresses; 4) feedback to SE4AI — where novel AI tools are developed to support challenges arise when assuring AIES; and 5) supporting the broader workforce transformation that will be needed for systems engineering to evolve with AI. In each section we highlight ongoing work and point out emerging trends. We conclude with a roadmap, suggesting key needs and trajectories for the next decade.</p>\u0000 </div>","PeriodicalId":13956,"journal":{"name":"Insight","volume":"29 3","pages":"10-14"},"PeriodicalIF":0.8,"publicationDate":"2026-07-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148386694","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"On (Human) Learning in Design Space Exploration and the Impact of AI Assistants","authors":"Daniel Selva","doi":"10.1002/inst.70065","DOIUrl":"https://doi.org/10.1002/inst.70065","url":null,"abstract":"<div>\u0000 \u0000 <p>Design space exploration (DSE) is a foundational activity in early-phase systems engineering aimed at identifying promising designs through modeling, optimization, and analysis. With the rapid emergence of AI-powered assistants, engineers can now delegate increasingly complex DSE tasks, from generating candidate designs to analyzing entire tradespaces. While this delegation may improve efficiency and solution quality, it raises an important question: what is lost when engineers offload cognitive effort to AI in DSE? This article argues that DSE is not only an optimization process but also an information discovery and learning activity through which engineers develop intuition, uncover trade-offs, and build defensible rationales for design decisions. Drawing on some of our recent empirical studies, we examine how AI assistance affects both DSE performance and learning, highlighting a nuanced relationship between the two. The findings suggest that while AI can enhance DSE outcomes, excessive automation may hinder learning, particularly for less experienced engineers. We conclude with implications for designing AI assistants that balance performance gains with the need to preserve and enhance human understanding in engineering design.</p>\u0000 </div>","PeriodicalId":13956,"journal":{"name":"Insight","volume":"29 3","pages":"43-47"},"PeriodicalIF":0.8,"publicationDate":"2026-07-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148386407","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Twenty Tips for Using Generative AI in Systems Engineering","authors":"Ray Madachy,&nbsp;Dan O'Leary,&nbsp;Tom McDermott,&nbsp;Sinan Bank","doi":"10.1002/inst.70053","DOIUrl":"https://doi.org/10.1002/inst.70053","url":null,"abstract":"<div>\u0000 \u0000 <p>Generative AI (GenAI) is rapidly moving from experimentation to routine use in systems engineering practice. We are already using these tools across virtually all process activities. The potential benefits are substantial, but the risks are equally real including inconsistent outputs, hidden assumptions, unverifiable calculations, weak traceability, context loss across sessions, and avoidable rework caused by poorly framed prompts and informal review habits. Current use is often ad hoc, whereas we need to move to disciplined individual and team processes that are measurable and thus manageable.</p>\u0000 <p>This article is dedicated to Dr. Barclay Brown, previous chair of the INCOSE AI Systems Working Group. Barclay recently passed away and was a strong advocate in our community for applying AI in systems engineering with practical engineering discipline. He could have written this article as well, and this contribution aims to carry his spirit forward by offering concrete guidance that engineers can apply immediately while preserving rigor, transparency, and accountability.</p>\u0000 <p><b>A SHORT TRIBUTE TO DR. BARCLAY BROWN</b></p>\u0000 <p>Barclay was very fond of Star Trek references. He would say about GenAI “It's life, Jim, but not as we know it.” He liked to describe GenAI like it was a new kind of being: “Neither human nor sentient; no soul, no goals; but with vast knowledge and information; creativity of a sort; little (or no) judgment; great speed; ‘eager’ to please; and happy to re-do things many times.” Why would we not want to converse with an entity like this?</p>\u0000 <p>He would note “Large Language Models (LLMs) have no interests, desires, ambitions, understandings, willingness, emotions, or intentions; and no capacity for these, though they can TALK like they do. But treating AIs like beings is the most productive way to relate to them. They know all about human behavior, so they respond better when addressed in that same way. So have them act in any way that serves you.” This makes them extremely easy to use but also makes it easy to accept what they tell you as correct, complete, verifiable, and immediately usable. Easy to use works for short information summaries, but not for complex systems engineering tasks.</p>\u0000 <p>Barclay enjoyed presenting tips on using GenAI in INCOSE and other conferences. In his spirit we have distilled some essential tips for systems engineers using GenAI. Use it but use it right.</p>\u0000 </div>","PeriodicalId":13956,"journal":{"name":"Insight","volume":"29 3","pages":"29-31"},"PeriodicalIF":0.8,"publicationDate":"2026-07-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148386751","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A Decade of AI and Systems Engineering: Looking Back and Forward","authors":"Tom McDermott","doi":"10.1002/inst.70062","DOIUrl":"https://doi.org/10.1002/inst.70062","url":null,"abstract":"","PeriodicalId":13956,"journal":{"name":"Insight","volume":"29 3","pages":"8-9"},"PeriodicalIF":0.8,"publicationDate":"2026-07-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148386747","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"STPD – Operational Design as a Model-Based Extension of STPA","authors":"Avi Harel","doi":"10.1002/inst.70067","DOIUrl":"https://doi.org/10.1002/inst.70067","url":null,"abstract":"&lt;div&gt;\u0000 \u0000 &lt;p&gt;The paper argues that the growing complexity of distributed socio-technical systems exposes a fundamental limitation of the STPA methodology. Although STPA transformed safety engineering (Leveson 2011; Leveson 2004) by treating safety as a control and constraint problem rather than purely a reliability problem, it still assumes that hazardous scenarios and the constraints needed to prevent them can largely be identified during design time. The paper claims that this assumption no longer holds in modern adaptive systems (Rasmussen 1997; Woods 2015), where operational contexts evolve continuously, AI-based components change behavior dynamically, distributed actors develop inconsistent interpretations of the operational situation, and unexpected interactions emerge during runtime.&lt;/p&gt;\u0000 &lt;p&gt;According to the paper, safety in such environments cannot rely only on predefined constraints. Instead, systems must continuously validate during runtime whether operational assumptions, coordination structures, and safety constraints still remain valid under changing conditions. This shifts the engineering challenge from specifying static safe behavior toward maintaining operational coherence under uncertainty. To address this challenge, the paper proposes extending STPA into a broader framework called system-theoretic process design, (STPD). STPD treats runtime surprise (Woods 2015; Dekker 2011) as an inherent property of complex adaptive systems (Rasmussen1997; Woods 2015) rather than as an exceptional case outside the model. Under this approach, systems must be capable of detecting deviations from expected operational assumptions, adapting coordination structures, managing degraded modes, recovering from anomalies, and learning from unforeseen situations while still constraining behavior within acceptable operational limits.&lt;/p&gt;\u0000 &lt;p&gt;The paper emphasizes that this extension substantially increases analytical complexity. The analysis must address not only fixed control loops and predefined unsafe control actions, but also adaptive behaviors, runtime transitions, resilience mechanisms (Hollnagel, Woods, and Leveson 2006; Woods 2015), evolving operational contexts, and partially unknown future states. Resilience itself introduces additional coupling, variability, coordination overhead, and verification difficulty. Consequently, a central tradeoff emerges: systems must remain adaptive enough to survive operational surprise while simultaneously remaining constrained enough to preserve safety, controllability, productivity, and usability.&lt;/p&gt;\u0000 &lt;p&gt;A central criticism presented in the paper concerns the scalability limitations of classical STPA. The number of possible socio-technical operational states grows exponentially with the number of system variables, making exhaustive analysis infeasible. The paper also argues that STPA's constraint-based logic requires engineers to identify all hazardous control actions expli","PeriodicalId":13956,"journal":{"name":"Insight","volume":"29 3","pages":"48-57"},"PeriodicalIF":0.8,"publicationDate":"2026-07-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148386442","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Trustworthy LLM-Augmented Data Reconciliation for Deployment-Constrained Environments","authors":"Danny J. O'Leary,&nbsp;Allison B. Ledford","doi":"10.1002/inst.70055","DOIUrl":"https://doi.org/10.1002/inst.70055","url":null,"abstract":"<div>\u0000 \u0000 <p>Digital engineering initiatives depend on trustworthy data lineage and a shared understanding of what enterprise data means. In practice, those conditions are often absent. The same concept may be represented in databases, forms, spreadsheets, email, and conversation-based exchanges, with different labels, different levels of structure, and different degrees of documentation. Organizations attempting to build a digital thread or expand model-based systems engineering (MBSE) often discover that the limiting factor is not tool availability. It is uncertainty about how data actually moves through work and how apparently related elements should be interpreted across sources (Ledford 2023; Ledford, Harris, and Purdy 2023; Ledford et al. 2025).</p>\u0000 <p>That problem is the foundation for data element mapping and analysis (DEMA). DEMA was developed to make hidden data work visible by tracing information at the levels of function, vessel, and individual data element. Prior DEMA applications have shown that a large share of enterprise data handling remains undocumented and that many exchanges still depend on manual transfer rather than integrated digital flow (Ledford 2023; Ledford, Harris, and Purdy 2023). Those findings are important because a digital thread is only as reliable as the organization's understanding of the data relationships it claims to connect.</p>\u0000 </div>","PeriodicalId":13956,"journal":{"name":"Insight","volume":"29 3","pages":"37-42"},"PeriodicalIF":0.8,"publicationDate":"2026-07-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148386693","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"","authors":"","doi":"","DOIUrl":"","url":null,"abstract":"","PeriodicalId":13956,"journal":{"name":"Insight","volume":"29 2","pages":"18-29"},"PeriodicalIF":1.0,"publicationDate":"2026-04-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148070838","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Exploring the Test and Evaluation Space using Model Based Conceptual Design (MBCD) Techniques","authors":"David Flanigan,&nbsp;Kevin Robinson","doi":"10.1002/inst.70041","DOIUrl":"10.1002/inst.70041","url":null,"abstract":"<div>\u0000 \u0000 <p>During the initial concept development phase, systems engineers focus on defining the problem space and system functions to explore candidate concepts that may address the systems engineers' problems. Model-based conceptual design (MBCD) techniques may be used to assist the customer and other stakeholders develop a greater understanding of the system concept, as well as identifying areas in the system that are affected by changes in requirements. This approach has generally been documented for describing the system concept in the early stages in the lifecycle, without significant focus on the test and evaluation (T&amp;E) space that would be needed to evaluate these concepts, or identifying where the T&amp;E space would be affected with a change in requirements. Our hypothesis is that decision makers would equally gain insight into the T&amp;E considerations as well as system space considerations using MBCD techniques. An approach is offered to extend the previously published MBCD methodology to better consider the T&amp;E space.</p>\u0000 </div>","PeriodicalId":13956,"journal":{"name":"Insight","volume":"29 2","pages":"13-17"},"PeriodicalIF":1.0,"publicationDate":"2026-04-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147683625","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Carbon Considerations for Systems Evolution","authors":"David Flanigan,&nbsp;Kevin Robinson","doi":"10.1002/inst.70043","DOIUrl":"10.1002/inst.70043","url":null,"abstract":"<div>\u0000 \u0000 <p>In the early stages of systems development, systems engineers will typically evaluate alternatives based on performance, cost, risk, and schedule to evaluate the solution space of alternatives. While these criteria have proven to be successful, there is growing interest in the analysis of carbon costs as well to contribute to the decision making. These decision criteria are very good to help the decision maker select the best alternative within the solution space in which to develop a system concept. We offer another criterion for consideration to account for carbon expenditure throughout the systems engineering lifecycle. We believe that including this dimension can influence decision makers to evaluate a richer portion of the solution space. This approach is developed and exercised with a notional example.</p>\u0000 </div>","PeriodicalId":13956,"journal":{"name":"Insight","volume":"29 2","pages":"30-36"},"PeriodicalIF":1.0,"publicationDate":"2026-04-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147683639","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}