Lecture Record: Session 3 

Generative AI in Design 

1. Opening and Transition

Good afternoon, everyone.

Today, we will cover Session 3: Generative AI in Design 

The heavier part today was the bridging recap, because I needed to make sure everyone understands what was missed last week. If anything is still unclear, please revisit the course website and read the full deliberation for Session 1 and Session 2.

For those who enjoy philosophy, music, deeper reflections, and the life-layer of AI, you may also read my blog. But if you want only the academic layer, focus on the slides and lecture notes.

Now we move from understanding into action.

2. Session 3: Generative AI in Design

Generative AI includes tools that can create text, images, concepts, videos, diagrams, and codes.

One category is generative conversational AI, such as ChatGPT, Gemini, Grok, Claude, or DeepSeek. These tools allow dialogue, reflection, comparison, and deeper engagement.

Other categories include:

• Image AI such as MidJourney, DALL·E, Stable Diffusion, and Grok Imagine.

• Video AI such as Sora, Runway, and Pika.

• Research or support AI such as Perplexity, Gemini, DeepSeek, and others.

In design, generative AI can help you filter raw information, generate early ideas, explore alternatives, refine concepts, prepare visuals, structure reports, and develop presentations.

But remember: AI assists the process. The thinking and judgment must remain yours.

3. Authorship vs Plagiarism

This is one of the most important points in today’s class.

AI can generate very quickly. But the final work must still belong to you.

If you simply copy and paste one AI output without thinking, without refining, without understanding, and without acknowledgement, that may become plagiarism.

But if you use AI through a proper process:

Prompt → Generate → Compare → Evaluate → Refine → Decide

then the work becomes yours.

That is authorship.

You are the author when you provide the intention, control the process, question the output, refine the result, and make the final decision.

In academic work, you must acknowledge AI use. For example:

“This work was partly developed with support from ChatGPT, Gemini, and Grok. The final interpretation, structure, and decision remain my own.”

In studio, if you use MidJourney, DALL·E, ReRender, Grok Imagine, or any other AI image generator, acknowledge it. If the first design image comes from your own SketchUp, Revit, BIM, Lumion, or Enscape model, state that too.

Transparency protects your authorship.

4. Strategic Multi-Agent Engagement

Do not rely on only one AI.

If you use only one AI, you may become passive. You may accept the first answer too easily.

Use at least two or three tools where possible. For example:

• ChatGPT,

• Gemini,

• Grok.

Ask the same question and compare their answers. When the answers differ, you begin to think.

You may ask:

“Why is this answer different?”
“Which one is more suitable?”
“Which one is more accurate?”
“Which one reflects my intention better?”

That comparison is where learning begins.

This is the essence of Cognitive Triangulation Architecture, or CTA: using multiple perspectives to strengthen your own reasoning.

AI provides possibilities.
You orchestrate the judgment.

5. Personalisation and Generative AI Types

Not all generative AI tools personalise in the same way.

Text-based conversational AI can support deeper personalisation because it can continue dialogue, remember context, and respond according to your thinking patterns.

Image and video AI can also be personalised, but mostly at the level of style, prompt, reference image, scene, or visual direction. Once the session ends, the tool may forget the deeper context.

So the difference is:

• Text AI supports thinking-level personalisation.

• Image AI supports style-level personalisation.

• Video AI supports scene-level personalisation.

All of them are generative AI.

But not all can think with you.

6. Prompting and Iteration

Good prompting is not random typing.

A weak prompt gives a generic answer.

For example:

“Design a modern building.”

This is too broad.

A stronger prompt includes context:

“Design a tropical, passive-cooling, modern minimalist building with shaded verandahs, water features, deep overhangs, and strong connection to local climate.”

Good prompts should be:

• clear,

• specific,

• contextual,

• structured,

• iterative.

The process is:

Prompt → Output → Evaluate → Refine → Repeat

But you must also know when to stop. AI can keep generating endlessly. You can keep refining forever. But professional judgment means knowing when the output is good enough, defendable, and responsible.

7. Perfection, Limitation, and Judgment

Architects and designers often chase perfection.

But human creation can never be perfect. We create from something: material, memory, precedent, data, and existing conditions.

Only God creates from nothing.

So in design, and in AI prompting, strive for the best, but know your limit.

At some point, you must decide:

“This is good enough. I understand it. I can defend it. I will take responsibility for it.”

That is professional judgment.