Now we move into Session 4: AI in Building Performance.
Here, we shift from asking:
“What does the building look like?”
to asking:
“Does the building perform?”
Building performance includes:
structure,
energy,
comfort,
function,
daylight,
ventilation,
thermal response,
environmental behaviour.
A beautiful building can still fail if it overheats, wastes energy, floods, creates glare, performs badly, or does not support users properly.
That is why AI in building performance is important.
There are many AI-supported or computational tools that help evaluate building performance.
Examples include:
Autodesk Insight,
Sefaira,
Ladybug,
Honeybee,
ClimateStudio,
Radiance,
Velux Daylight Visualizer,
Karamba3D,
Autodesk Robot Structural Analysis,
ETABS,
CFD tools,
BIM-integrated analysis systems.
These tools can help analyse:
energy use,
daylight,
thermal comfort,
ventilation,
structural behaviour,
circulation,
flood risk,
environmental impact.
But many of these are narrow-function AI tools. They are specific, computational, and task-based.
They are different from conversational generative AI.
This distinction is important.
A BIM-based performance tool may be more accurate because it is narrow, computational, and specific.
Generative AI, such as ChatGPT, Gemini, or Grok, is broader and more intuitive, but it can also make mistakes.
Generative AI works through language patterns, semantic reasoning, and trained data. It may produce convincing answers that are wrong.
So always verify.
Use generative AI to explore, explain, compare, and reflect.
Use narrow AI tools to simulate, calculate, and analyse specific performance conditions.
But in both cases, the architect must make the final judgment.
In Malaysia, performance is strongly connected to climate.
A glass building may look beautiful at night, but at 2:00 pm in Kuala Lumpur, it may suffer from high solar heat gain, glare, and cooling load.
Traditional Malay houses, on the other hand, were designed without AI, but they already show climate wisdom:
raised floors,
cross ventilation,
shading,
deep roof overhangs,
climate responsiveness.
This shows that wisdom existed before AI.
AI can simulate and visualise performance, but architects must still understand local climate, culture, regulation, and behaviour.
AI may also support regulation checking, including UBBL-related logic such as:
travel distance,
escape routes,
compartmentation,
fire safety logic.
Some firms may train AI using local regulations and office standards. In future, these tools may become more common.
But many AI tools are trained generally for global use. They may not automatically understand Malaysian regulations, local authority requirements, or project-specific constraints.
Therefore, the architect must provide context and verify the output.
AI can assist.
But the architect remains responsible.
AI can simulate, predict, and visualise possible outcomes.
But simulation is not reality.
Simulation depends on assumptions, data quality, modelling accuracy, and input conditions.
Once a building is constructed, many other factors appear:
workmanship,
material changes,
site conditions,
user behaviour,
weather,
operation,
maintenance.
AI can predict, but it cannot guarantee.
That is why the architect must evaluate reality.
Session 3 and Session 4 connect through one principle:
AI can expand your thinking, but it must not replace your thinking.
Generative AI helps you create, compare, and reflect.
Performance AI helps you simulate, calculate, and test.
But in both cases:
AI provides answers.
Architects provide judgment.
You are not only a user.
You are the orchestrator.
You are the evaluator of reality.