🔊 Play Transcript

Introduction:

This white paper summarizes key insights from a podcast featuring Senior Researcher Dr. Michael Kraus and Doctoral candidate Sophia Kuhn, both from ETH Zurich, discussing the application of Artificial Intelligence (AI) in the field of construction, particularly construction maintenance.

AI in Construction

Artificial Intelligence demonstrates exceptional versatility within the building sector. Its applications extend to computer vision, language processing, data mining, and the Internet of Things. The research endeavors of Michael Kraus and Sophia Kuhn are particularly focused on developing AI algorithms designed to improve the structural planning and the evaluation of projects in their early phases.

AI in Practice

I.      Programming Languages and Software

The programming languages that form the foundation of AI in this context are varied. Python is the primary language due to its versatility, followed by R, which is favored for statistical applications, and C++ for tasks that demand higher performance. The software landscape in this domain is rich and varied, with both startups and established construction companies contributing to a burgeoning market of specialized applications.

II.    AI Models in Use

Choosing the right AI model is contingent on the data available. The spectrum of models includes machine learning algorithms for tasks such as supervised machine learning (classification and regression) and unsupervised machine learning (e.g. clustering algorithm). There is also an increasing interest in generative AI technologies, such as Autoencoders and Transformers, due to their sophisticated capabilities

III.   Data Quality and Collection

• High-quality data is the backbone of effective AI applications. Inclusion of domain experts is crucial in AI pipeline development. The preprocessing of data, which includes “meta-data extraction”, is crucial for making the data usable for AI algorithms. Data must be relevant, accurate, representative, and free from unwanted biases (bias in data collection, especially in subjective areas like bridge inspection, can affect AI outcomes).
• Data collection in construction often involves capturing diverse data types, including images, videos, sensor data, and textual records.

The triad of AI effectiveness: the task definition, performance metrics, and quality of data.

Future Perspectives

I.      Potential

Artificial Intelligence is redefining the potential for rapid data processing within the construction industry, providing significant opportunities for sustainable resource planning and optimization. A notable potential of AI is in its ability to bridge the knowledge gap anticipated as seasoned experts from the baby boomer generation retire. By capturing their deep insights, AI can preserve invaluable expertise for future use. AI also presents the opportunity to harness the vast array of data sources in construction, whether they are spoken or written.

II.    Limitations

However, the journey of AI integration is not without its challenges. The complexity and the often opaque nature of AI algorithms pose significant hurdles, particularly in traditional fields such as construction. The industry is tasked with the challenge of quickly adopting AI technologies while also thoroughly understanding their technical and practical limitations. The integration of AI into construction necessitates a fundamental reevaluation of organizational processes. Expert knowledge needs to be harnessed to ensure practical and effective applications.

Technical Challenges in AI Integration

• Data Quality and Availability: Ensuring data is accurately preprocessed, cleaned, and formatted, including the acquisition of labeled, representative data.
• AI ‘Black Box’/Explainability: Improving transparency in AI decision-making to build trust and accountability.
• Algorithmic Bias: Developing methods to identify and mitigate biases within AI algorithms to ensure fair outcomes in construction projects.
• Interoperability: Ensuring that various databases, systems, and construction technologies can work together seamlessly.

Practical Challenges in Organizational Integration

• Integration with Existing Processes: Streamlining the adoption of AI into existing construction workflows and systems. Restructuring consulting practices to incorporate AI-driven insights and decision-making processes.
• Use of Information: Establishing regulations governing the application of AI outputs to ensure responsible use.
• Cybersecurity: Implementing robust security measures to protect against cyber threats.
• Skills Gap: Investing in training and development to equip current personnel with the necessary AI skills.
• Resistance to Change: Addressing cultural and organizational hesitance in adopting new AI technologies.

For industry professionals, ongoing training and lifelong learning have become indispensable in adapting to the ever-evolving landscape of AI technologies.

Conclusion

AI in infrastructure and construction offers significant opportunities for efficiency, safety, and sustainability. However, its successful implementation hinges on quality data, interdisciplinary collaboration, and continuous education in AI technologies. Education and training in AI must become integral parts of the curriculum for construction and engineering professionals. Collaboration between technologists, industry experts, and policymakers is essential to address the challenges and leverage the full potential of AI in construction. With concerted efforts, AI can lead to smarter, safer, and more sustainable construction practices, shaping the future of the built environment.