Can you trust your AI assistant?

An AI assistant can provide a wide range of benefits to a structural engineer. Here are some further ways that an AI assistant can aid a structural engineer:

Data analysis and optimisation: AI can analyse large amounts of data and identify patterns that may not be apparent to a human designer. By analysing data on building materials, construction techniques, and environmental factors, an AI assistant can help optimise the design of a building for factors such as cost, energy efficiency, and structural integrity.

Risk assessment: AI can help identify potential risks and hazards associated with a building design, such as earthquake or wind load risks. It can also provide simulations to help evaluate and mitigate those risks.

Design automation: AI can automate certain aspects of the design process, such as creating preliminary designs, generating 3D models, and generating construction plans. This can save time and reduce the risk of human error.

Collaboration and communication: An AI assistant can facilitate communication and collaboration between different members of a design team, allowing for more effective sharing of ideas, feedback, and revisions.

Continual improvement: AI can continually monitor and analyse data from a building's performance after it is constructed, providing feedback to the designer for future designs.

Assistance with code compliance: An AI assistant can help ensure that building designs comply with relevant codes and regulations, such as the Building Regulations in the UK.

Close

Despite the benefits of using AI and language models in structural engineering, there are also potential risks that must be considered. One of the biggest risks is the potential for errors in the AI or language model. If the AI or language model is not properly trained or programmed, it could make mistakes that could lead to serious safety issues in the building structure.

Another risk of using AI and language models is the potential for bias in the data used to train the model. If the data used to train the model is biased in some way, this could lead to biased or inaccurate results when the model is used to design a building structure.

Structural engineers can use several methods to discern whether or not generative pre-trained transformer (GPT) or similar language models are providing correct information on a subject related to design. Here are some ways to ensure the accuracy of information provided by AI in the design process:

Validation and verification: Structural engineers can validate and verify the information provided by the AI using established principles and data sources. They can cross-check the AI's output against existing building codes, standards, and regulations to ensure that the design meets the required standards.

Expert review: Engineers can also seek the assistance of domain experts in the field, such as other experienced engineers or architects, to validate the design or provide feedback on the AI's output. This can help confirm the accuracy of the design and provide suggestions for improvement.

Testing and simulation: Engineers can use physical testing and simulations to verify the accuracy of the AI's output. They can run simulations of the building design under different conditions to assess its performance and compare the results with the AI's predictions.

Data quality assessment: Engineers can assess the quality of the data used by the AI, ensuring that the data is relevant and reliable. They can identify potential sources of bias or inaccuracies in the data and adjust the AI's parameters accordingly.

Regular calibration: Finally, engineers can regularly calibrate the AI to ensure that it remains accurate and up-to-date. This involves periodically checking the AI's performance against the validation criteria, making any necessary adjustments or improvements, and revalidating the design as needed.

Overall, it's essential for engineers to critically evaluate the output provided by AI and language models and ensure that they use appropriate validation and verification techniques to confirm the accuracy of the information provided. By implementing these methods, structural engineers can confidently integrate AI into the design process, helping to improve design quality and efficiency.

For those of you who have navigated this far down in the blog, you may be interested (or alarmed) to hear that everything prior to this sentence was generated by ChatGPT, a language model developed by OpenAI. OpenAI's other offering DALL-E 2.0 generated images to complement the text. The irony of asking an AI how an engineer would go about checking an AI's output is quite obvious, but the detail provided by the software is quite impressive regardless.

The response provided, whilst fairly general and not particularly detailed, does provide domain specific advice rather than simply parroting general advice about AI. ChatGPT has ‘understood’ the context of the question asked, ‘understood’ that it is specifically related to structural engineering, buildings and a field of design. It has then taken it’s awareness of the existing industry and provided output that provides reasonable advice to someone in our profession, including the interface with other professions such as architects.

Readers of this blog may have noticed the repetitive nature of the phrasing and word choice. This may have raised suspicions for some of you about the author of the prose. However, it also betrays a potential pitfall with using AI – lack of creative thinking. We could assume this would carry over into the design space. AI design suggestions and solutions are based on the training data, and thus may not offer the creative solutions a human engineer would.

Another potential concern (and perhaps a more interesting  one) is that I find myself tending towards trusting the output implicity and - as when with given the answer to a difficult maths problem - losing the ability to think beyond the bounds set by the answer provided. The psychology of critical thinking and problem solving when given an AI solution is certainly a space to watch.

Figure 2 - Over reliance on digital technologies can proliferate errors much more quickly
 

Close

If we combine several of these issues, we can think of many scenarios that could be catastrophic for structural design. Consider an AI trained on data from existing structures that neglect a particular design criteria. The model would continue to suggest the deficient design/methodology and offer variations within those bounds that never considers the deficiency.

An engineer is given an authoritative design suggestion and the psychology of being given an answer constrains their thinking, supressing their instinct to question the output. The structure is built and added to the training data with the existing issue unresolved. The validity of the previous training data is strengthened by the addition and the problem remains unidentified and unaddressed. We have inadvertently implemented a form of digitised confirmation bias.

The implications of this technology are as yet unknown, but with the speed of development and its convincing output, a pandora's box has certainly been opened.