Since the release of ChatGPT in November 2022, Generative AI has dominated the headlines. Every industry – including Architecture, Engineering, and Construction (AEC) – is exploring ways to use the technology to gain a competitive edge.
Traditional AEC software is typically designed for specific tasks or disciplines, leading to multiple isolated systems, fragmented workflows, and limited interoperability. These solutions rarely adapt to changing project requirements. GenAI’s ability to ingest data from various sources brings agility that traditional industry software has lacked. GenAI can analyze vast amounts of data, generate accurate simulations, and provide real-time insights. These capabilities will significantly enhance efficiency, reduce costs, and improve overall project outcomes in the AEC industry.
GenAI can integrate essential AEC industry data from isolated software systems, including engineering design, project management, cost estimates, and valuable stakeholder feedback. This integration allows for a centralized and comprehensive view of project information, enabling better decision-making. GenAI can analyze the centralized project data and provide intelligent insights by leveraging machine learning algorithms. The output could include identifying potential risks, suggesting design improvements, or predicting project delays. These insights can help all the stakeholders make informed decisions and take proactive measures to optimize project outcomes.
Additionally, GenAI can use its ability to learn by ingesting historical project data to optimize project outcomes. By analyzing past projects, it can identify patterns, best practices, and areas for improvement. This knowledge can be applied to current projects, leading to better cost management, schedule adherence, and quality control. In essence, GenAI can be the silent partner in every project. Whether assisting architects in material selection, presenting designers with alternative designs, guiding contractors on safety protocols, or empowering project managers with data-driven insights, GenAI can act as a reliable assistant throughout the project lifecycle.
GenAI Adoption in AEC: Challenges to Overcome
A combination of organizational readiness, strategic planning, and targeted use cases can facilitate the adoption of GenAI in the engineering and construction industry. Initially, capturing the industry’s complicated technical engineering data across structural, mechanical, electrical, plumbing, and project management disciplines may be challenging. Construction also relies heavily on physical and situational awareness. Consequently, this hinders GenAI’s ability to perform and can lead to inaccurate results and hallucinations, which might undermine construction decisions through misguided outputs (incorrect design outputs and incorrect critical paths in managing construction schedules). Several strategies need to be employed to mitigate these potentially unsafe hallucinations. These strategies include using high-quality training data, verifying GenAI’s grasp of domain-specific engineering and construction content, instituting simulated testing to validate predictions, continuously monitoring uncertainty, and introducing human oversight throughout the AI decision-making process.
A Strategy to Efficiently Adopt GenAI in AEC
Considering the challenges above, AEC companies should define a systematic approach to adopting GenAI. The guidelines below can help AEC companies develop a successful roadmap.
- Awareness and understanding: Start by educating non-technical stakeholders about GenAI's capabilities and potential benefits. Utilize workshops, seminars, or training sessions that teach the principles of AI, explain how to use specific GenAI tools, and explore how to interpret and validate the outputs. These sessions should include clear discussions of GenAI's ethical implications and potential risks.
- Data quality: The success of AI depends on data quality. When data quality is high, AI models provide reliable and accurate outputs, which leads to smarter decision-making, improved business performance, and ultimately better employee and customer experiences. Companies may need to proactively invest in data collection tools and develop a more robust data management strategy, ensuring high-quality data is available to train AI models.
- Fine-tuning pre-trained models: Start by fine-tuning pre-trained language models using construction-specific data. This data includes design documents, building codes, contractual documents, technical documents, and BIM data. This approach helps the model understand the specialized vocabulary and context of the construction industry.
- Pilot projects: Start with small pilot projects to understand Generative AI’s practical challenges and benefits, explore the technology, and make necessary adjustments.
- Human oversight: GenAI systems require human oversight to ensure quality and accuracy. Humans have a richer understanding of context, culture, and real-world knowledge, which enhances the model’s understanding through feedback and interactive editing. The human-in-the-loop approach combines AI generation with human judgment for improved results.
- Evaluating business impact: Assess GenAI's business impacts through experiments that measure key performance indicators such as productivity, cost, time, and risks. Constant evaluation can help quantify the benefits of investing in GenAI.
- Developing custom models: Collaborating with AI experts and companies will produce more powerful custom language models that deliver additional value in construction-related tasks. This process involves compiling extensive datasets from the AEC domain and training models on this specialized content that consider challenges like data labeling, computational power, potential biases, overfitting risks, and evaluation difficulties. While large language models have their place, small language models (SLMs) offer a tailored approach, especially with context and specialized terminology like those in construction workflows. Context-aware models like SLMs can significantly enhance domain-specific applications.
- Model updates and interpretability: Regularly updating AI models must be a priority. Training data can quickly become outdated as materials, methods, and regulations frequently change. Without recent data, models can experience hallucinations and provide unreliable guidance. For example, an AI chatbot trained before the COVID-19 pandemic may overlook the subsequent supply chain disruptions and labor shortages. Regularly retraining models on new data is essential but costly and complex at scale and represents a higher level of GenAI maturity.
- Managing expectations: Companies should manage expectations about GenAI's capabilities and be prepared for adjustments as projects move from one phase to another. Unrealistic expectations can lead to project failure.
- Governance and oversight: Companies should have sufficient governance and oversight when implementing GenAI projects, including establishing and adhering to policies that ensure AI applications are responsible and ethical, and verify responsible data handling, quality, transparency, security, protection, etc.
