POLYU SCHOOL OF DESIGN / SD2112 / WEEK 07
AI + X.
This week in AI. Recap. Quiz. Your team and product.
AI NEWS · 13 OCT 2026 · GOOGLE
Gemini 4 Argon
Complex coding and spatial reasoning; Google shows Blender / Three.js demos.
Released: paid API, Gemini Pro / Ultra; Antigravity access for Ultra.
Vendor demos and performance claims; not an independent class test.
Which design choices would you keep control of?
AI NEWS · 7 OCT 2026 · ANTHROPIC
Haiku 5.5: choose the job
Small model with adjustable effort for narrow, repeated agent tasks.
Released on Claude Platform, AWS, Google Cloud and Azure.
Larger Claude models remain stronger for complex coding in vendor tests.
Which small production task could you delegate and check?
AI NEWS · 7 OCT 2026 · OPENAI
An answer can be an interface
GPT-6 Intelligent UI creates interactive diagrams, forms and small tools.
Chat rollout: paid plans from 7 October; Free / Go from 8 October.
Enterprise access depends on admin settings. Check the generated behaviour.
Does the interaction explain something a static image cannot?
AI NEWS · 8 OCT 2026 · ANTHROPIC
Motion you can edit
Claude Motion animates text, charts, shapes and images as editable code.
MP4 export; Team / Enterprise beta. Enterprise admins must enable it.
Code-based animation, not a video-generation model producing footage.
When would editable animation serve your team video mark?
AI NEWS · STATUS CHECK · 14 OCT 2026 · MELTY
Games become remix material
Windows launcher for mods and mashups of games you already own.
Give Claude Code or Codex its instructions to build and publish a mashup.
Free download; compatibility and safety claims need checking per mod.
What makes a mashup a new creative work?
AI NEWS · 11 OCT 2026 · PHOTOCRAFT 0.7.0
Rebuilding creative tools
ArtCraft lists 13 Rust creative apps, from image editing to audio.
PhotoCraft 0.7.0: downloadable desktop builds; Apache-2.0 source.
The lineup is early alpha / in development; familiar UI is not feature parity.
What must a replacement preserve: habits, files or results?
AI NEWS · 11 OCT 2026 · REA 6.4.0
If agents can inspect software...
REA connects agents to app / binary analysis through MCP and a CLI.
MIT-licensed tool; 6.4 adds Android inspection and analysis improvements.
Evidence can support reimplementation; inferred behaviour still needs tests.
Where is your line between inspiration, compatibility and copying?
WEEK 7 / RECAP THEN QUIZ THEN TEAMS
Today
01 11:30-11:45 / Weeks 1-6 / twenty key concept slides
02 11:45-12:00 / ask me anything / send your questions
03 12:00-12:30 / mid-term quiz on Canvas
04 Gio demos Canvas team registration live
05 Step 1 / twenty minutes / one core-concept text per team
06 Step 2 / twenty minutes / one product image per team
05 · A WORKING DEFINITION · GIO, 2025
Intelligent-like output or behaviour, achieved through computation.
05 · TWO MACHINES
Write the rule, or show the examples.
Machine A: symbolic AI, 1956 onwards — definitions, logic, expert systems. Machine B: machine learning, 1958 / 1986 / 2012 — statistics over examples.
06 · THE DISTINCTION THAT ORGANISES THE COURSE
Using AI, or incorporating AI.
WEEKS 1 – 6 · THE PROCESS
Using AI
AI as a tool in how you design: a brief drafted with a chatbot, a moodboard from a diffusion model, generative fill, layouts from Figma Make.
You stay the author. You become the curator, the briefer, the editor.
WEEKS 8 – 12 · THE PRODUCT
Incorporating AI
AI as a material in what you design: a feed, a recommendation, an assistant, a filter — the product decides something for each person, on its own.
You design a behaviour, not a picture. Data, bias, trust and accountability become design problems.
02 · THE DEAL
Exact. Explainable. Brittle.
EXACT
Same input, same output.
A rule does the same thing every time. Even chance can be made repeatable: give the die a seed and the same picture comes back, on any machine, forever.
EXPLAINABLE
You can point at the line that decided.
Every output has a reason you can read. When a rule-based product does something strange, you can find out why. Machine B cannot say why. Next week.
BRITTLE
Nothing outside the rule can happen.
The definition does not know beanbags exist. Every new case needs a new rule, written by hand. That bottleneck ended the expert-systems boom.
06 · RANDOM() · RANDOMSEED()
Same rule, three seeds.
random(600) is a number between 0 and 600, different every run. randomSeed(1) loads the die: the same picture every run, on every machine. Where you call random() is where the rule lets go; everything else is the rule.
Three runs of the ten-point sketch with randomSeed(1), (2) and (3). Change the seed, keep the rule: a new picture that is still yours.
07 · THE SPEC IS THE DESIGN
One spec. Three executors.
LeWitt's drafters; you, in the editor; a language model on genai.polyu.edu.hk. Same words, delegated execution, and never quite what you meant. Machine B writes machine A, and machine A does exactly what it is told.
A language model is not a rule: the same prompt gives different code every time. The code it writes is a rule: the same seed gives the same picture every time.
03 · PROTOTYPE THEORY
A concept is its best examples.
THE THEORY
A structured representation of what members tend to have.
Not a list of conditions but a picture of the typical case, and a distance from it. Membership is a degree: a robin is a very good bird, a penguin a poor one, and neither needs a definition.
THE GAIN
Only similarity is needed.
No definition to write: you judge a new thing by how much it resembles what you have seen. That is learning from examples, and it is why a machine can hold "chair" with no rule for it.
THE COST
Outliers, and combinations.
Exceptions are hard: there are fewer examples at the edge, so the edge is unsure. And combining concepts is a puzzle: which properties of "pet" and "fish" does "pet fish" keep?
04 · ONE NEURON
A neuron is a weighted vote.
Two inputs, two weights, a bias, a threshold. Multiply, add, compare with zero: that is the whole unit. Three numbers hold everything it knows. Rosenblatt, 1958, called it a perceptron; the shape and the size are the week-1 cup.
Rosenblatt, "The perceptron: a probabilistic model for information storage and organization in the brain", Psychological Review 65, 1958. In the html deck the neuron learns live, one example at a time.
Open the live sketch › — click = pause · C = start again
05 · BACKPROPAGATION · 1986 · THE PICTURE, LIVE
The guess goes forward. The error comes back.
Nine pixels in, five and three hidden units, two out: nineteen units, 66 weights. Rumelhart, Hinton & Williams, "Learning representations by back-propagating errors", Nature 323, 1986.
Open the live sketch › — click = pause · C = start again
07 · FAUCONNIER & TURNER · 2002
Two inputs. One new space.
Two input spaces: a house and a boat. A generic space of what they share: a structure, a place, people who use it.
The blend: a houseboat. It took the rooms from the house and the hull from the boat, and it is one thing, not two.
After Fauconnier & Turner, The Way We Think: Conceptual Blending and the Mind's Hidden Complexities, 2002.
02 · TRANSFORMER · CAUSAL SELF-ATTENTION · VASWANI ET AL. 2017 ↗
One sequence. Many positions working together.
Open the live sketch › — choose a step · click a token · replay training or generation
03 · THREE ANSWERING SYSTEMS
A tool call alone does not make a system an agent.
RULED FAQ
Search a prepared set.
“What time?” returns the answer written into its rule.
RAG · ONE PASS
Retrieve, then generate.
Find the official event page, then answer once from its text.
AGENT
Choose, act, observe, continue.
If access details are missing, search an allowed source or ask.
04 · REACT · THE AGENT LOOP
Thought → Action → Observation
01 · THOUGHT
What do I need?
Find the exhibition start time.
02 · ACTION
Use a tool.
Search the allowed event page.
03 · OBSERVATION
What came back?
The page has no start time.
→
→
↓
04 · FINISHED?
YES → stop with a draft.
↑
NO → think again with new evidence
IHDE · POSTPHENOMENOLOGY AND TECHNOLOGICAL MEDIATION
Ihde: technology mediates our world.
Same camera. Different relations—depending on how you use it.
CLIP · CONTRASTIVE LANGUAGE–IMAGE PRETRAINING
CLIP brings words and images into a shared space.
Matching pairs move closer; mismatched pairs move apart. CLIP does not generate an image.
CLASSIC LATENT DIFFUSION · TRAINING
Training teaches a denoiser to remove noise.
Compare predicted noise with known added noise; use that error to update the denoiser, not the latent.
CLASSIC LATENT DIFFUSION · GENERATION
Generation turns new noise into an image.
The loop changes the latent, not the trained weights. Decode the final latent after the last update.
SOUND · A PICTURE THROUGH TIME
A spectrogram keeps the changing spectrum.
Time × frequency · colour encodes magnitude or energy, not human loudness directly.
REPRESENTATIONS · MIDI
MIDI describes events. It contains no sound.
A score is not a recording · the instrument or synthesizer supplies the sound.
MACHINE B · TWO DOCUMENTED APPROACHES
A picture of sound. Or tokens of sound.
Historical Riffusion v1 and MusicGen illustrate mechanisms—not Suno’s unpublished internals.
11:45-12:00 / BEFORE THE QUIZ / YOUR QUESTIONS
Ask me anything.
What would you like to clarify?
Send your questions on ClassPoint until 12:00.
ClassPoint · short answer — answer on the projector
CANVAS / WEEKS 1-6 + PLAYLIST / 10%
Mid-term quiz.
12:00-12:30.
AFTER THE QUIZ / AI + X / TEAMS OF 4-5
One concept. One product image.
Choose an area of life or practice, a person and a situation that matter to your team.
The product must incorporate AI: what does it recognise, recommend, adapt or generate during use?
Step 1: form and register your team; submit one short core-concept text.
Step 2: make and submit one image showing the product and a moment of use.
LIVE DEMO / CANVAS TEAM REGISTRATION
Watch Gio register a team.
Gio demonstrates the actual Canvas group set and registration route live.
In Step 1, form a team of 4-5 and make sure every member joins the same Canvas team.
Choose one representative to send both ClassPoint submissions, identifying the same team.
If you need a team or access help, tell a TA as Step 1 begins.
TEAM TEXT · 01 / CORE CONCEPT
20 min
Form your team. State the idea.
Form a team of 4-5 around a shared interest. Follow the demo to register on Canvas.
Choose one direction: for [person] in [situation], our [product] uses AI to [action], helping them [benefit].
Name what the person still decides and one assumption you need to investigate.
One representative submits one short core-concept text on ClassPoint, with the team name.
ClassPoint · short answer — answer on the projector
TEAM IMAGE · 02 / PRODUCT IMAGE
20 min
Show the product in use.
Sketch, mock up or generate one image showing your product in use, not just its logo.
Show the person and product. Label the AI action and what the person decides.
The same representative uploads one image on ClassPoint with the team name. Reopen it to check.
ClassPoint · image upload — answer on the projector
FINISH / TWO SUBMISSIONS PER TEAM
Check the text, image and team.
One core-concept text and one product image are visible in the two ClassPoint activities.
Both carry the same team name and describe the same provisional AI + X product.
Every member has joined the same Canvas team. Reopen it together to check.
If submission or registration failed, ask a TA before leaving.
GROUP PROJECT / 40% / TEAMS OF 4-5
Today is a seed. Week 13 is the showcase.
Design a product with AI inside it. Account for what its decisions do to people.
Week 8: develop the formal one-page group proposal: the decision, for whom, with what data.
Week 13: A0 poster, 3-5 minute video and a one-page mediation brief.
Research, ethical and social impact, communication and collaboration matter. See the syllabus rubric.
REMINDER / WEEK 7 / 20%
Your individual reflection is separate.
About 1000 words on AI in your creative process, especially Machine A versus Machine B.
Use at least three of your own weekly-challenge experiments from Weeks 2-6, with images.
End with a short AI-writing process note. Submit individually on Canvas in Week 7.
Use the existing brief and rubric. Check the actual deadline on Canvas.
Your team.
Your X.
One core concept. One image of an AI product.
SD2112 / WEEK 07 / VENETANJI.GITHUB.IO/SD2112-TEACHING