How can structural engineers leverage Conversational AI benefits while managing its risks?
Structural engineers spend an estimated 20-40% of their work hours writing. Do you ever struggle with writer’s block or spend too much time perfecting work emails? It’s understandable to feel frustrated and wonder if this is what you really signed up for as a structural engineer.
Conversational AI tools like ChatGPT can solve this problem, acting as a writing assistant with the potential to drastically cut writing time so you can do more work you enjoy while more effectively communicating your value to teams, clients, and society.
In December 2022, based on an entrepreneurial colleague’s enthusiastic recommendation, I first tried ChatGPT. A feeling of “this is going to change everything” came over me as I explored; the last time I felt similarly was when I viewed my first Revit demonstration 20 years prior. I jumped in with both feet to learn all I could. The result has been working with AI experts, sharing my learnings in articles and virtual sessions, and collaborating on a report-writing workshop using this tool.
AI language models and OpenAI’s ChatGPT have since made headlines. ChatGPT is the fastest-adopted tool in the Internet age (See Figure 1). The rapid adoption rate means firm leaders must get up to date quickly; an estimated 43% of employees have already used it, yet 68% of those users have not told their bosses they are doing so.
AI tools like ChatGPT have the potential to be game-changing in helping engineers communicate effectively. At the same time, like any other software tool, it must be used with guardrails and a comprehensive understanding of the benefits and risks.
What is ChatGPT?
ChatGPT (and other tools like Claude, Google’s Bard, and Microsoft’s Bing) is an AI-powered language model that generates human-like text based on the input – called “prompts” – it receives. Conversational AI tools are built on large language models (LLMs); the current version is GPT-4. For example, GPT-3 was the latest model in December 2022. These models can currently pass standardized exams like the SAT, GRE, and lawyer BAR exams in the top 10% of test takers. Read more about the capabilities in OpenAI’s white paper, available for free here: https://openai.com/research/gpt-4.
With each model advancement, the responses to prompts have improved, sometimes dramatically. For example, the model refused to output source data in GPT-3. The current models will share sources.
The LLM data used to train ChatGPT and other AI tools is proprietary and not publicly available. The information for developing GPT-3 indicates that it was trained on publicly available Internet data from 2021 and earlier, including Wikipedia and Reddit. Less information is currently known about GPT-4, the latest large language model. Access to GPT-3.5 and earlier is available via multiple free tools, including ChatGPT, Google Bard, and the new Bing search engine. Access to GPT-4 currently requires a ChatGPT Plus paid subscription. It’s also important to understand that access to any of the free (and some of the paid) tools means that any data inputted is being used to train the next model unless you take deliberate steps available in some (but not all) tools to prevent that access. Any data you share may become public information, as Samsung employees painfully learned when their data leak went viral.
The information shared above and herein is current as of 9.1.23; however, these technologies are rapidly advancing and subject to change as new and better tools become available.
Structural Engineering Applications For ChatGPT
LLMs have the capabilities for technical applications for structural engineers who understand how to code or have IT departments developing proprietary AI tools. However, this article focuses on applications that any structural engineer can use right now, even without coding knowledge. For most structural engineers, the best application currently is as a writing assistant, helping to automate mundane but required tasks like writing meeting notes or reports.
One of the most common complaints among structural engineers is that the industry is becoming commoditized, and clients don’t appreciate structural engineers. There’s also the constant balance of wearing multiple hats in a seller-doer role, especially for project managers and above. Engineers at these levels may be responsible for both internal functions (structural engineering design, coordination, and managing staff) and external processes (client meetings, QA/QC, troubleshooting eld issues, producing reports). Seller-doer engineers also participate in business development and building firm reputations through thought leadership. SE leaders, especially of small and mid-size firms, often do all three jobs in their role, resulting in high burnout and overwork, a long-standing issue in structural engineering.
It’s here – in this area of communicating value to clients and others without your technical understanding to appreciate your expertise – where tools like ChatGPT can be a game-changer. With good prompts, ChatGPT can output responses in context, allowing it to present and summarize the same information from differing perspectives.
Examples of ChatGPT Use
The following are four practical ChatGPT structural engineering application examples:
- Explaining technical concepts to non-technical decision-makers: You are working with an architectural client in the early stages of building design. The team is discussing occupancy loads in one public-gathering space. You recognize that the higher occupancy load may drive the building into a higher occupancy category (thus a higher load importance factor and structural costs). You could input your technical understanding into ChatGPT and ask how you would explain it to the bottom-line conscious owner so they can understand and contribute to that decision. In this scenario, you’re using ChatGPT to become the client’s trusted partner/educator to help them make cost-conscious business decisions.
- Report writing: Structural engineers love going on-site, but only some enjoy writing reports. Imagine returning to the office, inputting your chicken-scratch notes, hitting a button, and immediately having a first report draft to review. This is possible right now (I’ve run workshops on this topic!) and can save you hours in writing. To be clear, I’m not saying the ChatGPT export is the report you hand over to a client in its final form. AI is a technology tool. Like other tools, it must undergo review before leaving the office. However, the most procrastinated and time-consuming part of writing is going from blank page to first draft, and tools like ChatGPT can get you there in a fraction of the time.
- Writing emails (especially when sharing unwelcome information): You receive an email from a client with a last-minute change the day before a deadline. That change has significant structural implications (such as a column move), and you need to craft an email response that advises that the deadline needs to move if the client requires this change. This type of email used to take me an hour to write (especially in my early career) to strike a balance between setting that boundary and preserving the client relationship. In a post-ChatGPT world, writing a first draft of this email takes minutes. This video shows an example of this type of prompt and output: https://bit.ly/SEChatGPTEmailexample.
- Thought leadership content: Client-facing engineers must publicly share their expertise to build their reputations. Examples include content creation, such as blogs, presentations, and social media posts. ChatGPT can act as a writing assistant and is especially helpful when repurposing already-written public content. For example, I used ChatGPT to generate an outline and first draft for this article based on my (100% human-written) abstract. Follow along with this ChatGPT example: https://bit.ly/ SEChatGPTExample4
Even Perfectly-Crafted Input Could Still Equal Garbage Output
Structural engineers can harness the power of ChatGPT and similar conversational AI tools to streamline their workflows, write faster and more effectively, and communicate their value better to non-technical stakeholders. Like any other software tool, you must also have the expertise to vet the accuracy of the output. Extra concern is warranted because the output of AI tools like ChatGPT sounds exceptionally knowledgeable, even when factually inaccurate.
For example, I asked ChatGPT to write my biography; I have had a robust online presence since before 2021, so it seemed a good test. ChatGPT output was factually incorrect, sharing that I had attended MIT and Stanford instead of my correct alma mater, Penn State. I have also asked ChatGPT to define structural engineering terms with similar inaccuracies.
The takeaway: Do not rely on ChatGPT or AI tools to provide accurate outputs. You must have the expertise to vet results for accuracy. From an organizational standpoint, you must have a QA/QC process to use this tool. The alternative is risking your professional reputation because AI output could be wrong or misleading.
Risk Management Concerns for Firm Leaders
AI integration in firms raises concerns over data protection, ethics, and reputation. The FTC investigated OpenAI for alleged harmful statements by ChatGPT, emphasizing the importance of risk management strategies.
AI writing tools like ChatGPT process large amounts of information and data, including proprietary methodologies, designs, and plans. Free tools currently available often use data inputted to train the model, resulting in concerns around protecting intellectual property (IP). Tech companies are racing to produce tools that will alleviate this concern. One example is Microsoft, which has been Beta-testing the Microsoft Copilot program in large organizations and is expected to roll it out to Office Suite Microsoft 365 users when testing is complete.
Other risks include the “black box” nature of the data on which the LLMs are trained. We know that training data includes publicly available Internet data. Therefore, some of that data is guaranteed to be inaccurate. False information, including societal identity-related bias (i.e., age, gender, race, veteran status, etc.), may be included in AI responses. This is especially problematic for documents that must avoid using discriminatory and non-inclusive language, such as job descriptions, promotion qualifications, resumes, or requests for proposals. Organizations must have a robust QA/QC process around any output partially or fully generated by AI tools.
Ethical use of AI tools is another risk. It will be years before state licensure and ethics codes catch up with AI. It’s up to engineers and organizations to use engineering judgment on proper use. Broad adoption rates mean many engineers are using it without telling their managers. This scenario raises questions about accountability, responsibility, intellectual property protections, and the potential risks associated with the undisclosed use of AI.
Right now, leaders can promote an environment that encourages open discussion regarding AI use, use it with guardrails to explore benefits and risks, and develop policies guiding responsible use. Policy examples are included in this LinkedIn article: https://bit.ly/ SEChatGPTLIpolicy.
The other ethics-related issue that has reared its head in recent conversations is the impact on client trust. Because conversational AI is new, many individuals have differing thoughts on “proper use.” Only individual firms can answer the question: If a client learns you wrote something with AI, does that diminish that client’s trust in you? Will they wonder what other shortcuts you took? Or be happy with your faster communication response time?
What Structural Engineers Should Do Now
Avoiding or ignoring the advancement of AI tools like ChatGPT isn’t a pragmatic solution. Telling staff not to use it is akin to saying don’t use the internet or smartphones at work – an impractical and arguably short-sighted approach. Instead, focus on understanding the technology, openly discussing its use, exploring best practices, and implementing robust guidelines and safeguards.
Developments like Microsoft’s anticipated public rollout of Bing Chat Enterprise and AI tool “Copilot” in Office 365 emphasize that AI capabilities are rapidly becoming mainstream and will soon become integral to how we work. Microsoft co-pilot Beta users tell me that this AI/GPT-trained tool is designed to integrate seamlessly with the Microsoft Office/Outlook/Teams platform. This is one example of a tool that will soon be available and expected to have capabilities to keep private data safe from being used to train the public LLM models.
When your data can be kept separate from the public models (while still allowing the use of the public model in analysis), the most significant current risk management concerns will be minimized. Technology companies are in a race to provide this capability, and I predict AI tools will become more rapidly adopted for SEs soon after. Individuals and firms taking steps now to understand and harness this technology can give themselves a competitive advantage in the marketplace.
Where do you start with ChatGPT and other conversational AI tools? Begin with example 4 (thought leadership content) because that information is or will soon be public knowledge. Report writing is the most advanced of the four examples herein. Reserve this exploration for AFTER your structural engineering organization has decided if reports are an appropriate application for the firm; many firms may prefer to wait until better data protection is available.
Structural engineering firms should take immediate action to explore which AI applications are helpful for their organizations and establish usage guidelines around AI use. This includes safeguarding data and IP and developing a QA/QC policy to vet AI outputs.
AI Tools Are Here to Stay
Adopting AI writing tools isn’t a trend. It is a fundamental shift promising to bring transformative benefits when navigated with understanding and proper QA/QC practices. We have reached an inflection point in the technological landscape for this technology (Figure 3).
This technology is changing quickly and will likely have changed in the months between my writing of this article and your reading of it. Structural engineers and firms must start planning for this future now. The first step is fostering an open discussion and learning environment regarding using AI tools in the workplace. Leaders should promote conversations about AI in the workplace, develop organizational use cases, and encourage experimentation using QA/QC criteria that protect proprietary firm data. This will help ensure ethical use, prevent misuse, and mitigate potential risks.
Conversational AI tools like ChatGPT have the potential to solve long-standing communication challenges in helping clients and the public see structural engineers’ value. Those communication challenges lead directly to the fees our industry can command and our ability to attract and retain top talent. By embracing conversational AI as a communication enhancement tool with guardrails in place, we can shape the future of structural engineering using that technology to drive industry progress.