Unit 3

Human by Architecture 

Adetokunbo Ayoade

Brief

"We shape our buildings; thereafter they shape us”    - Winston Churchill 


We are drawn to this practice of architecture because we inherently sympathize with Mr. Churchill, understanding the buildings and cities we create are more than means to achieve a goal. Each line, whether as pencil marks in a sketch or polyline in Rhino, they are always a projection into the future of how the manner we inhabit this world can be different. We use architecture as a driver to create this change, for our physical and sensual comfort, as well as defining new social norms for engaging with one another. We are optimists, believing architecture will lead us collectively to a new state of being; evolved.


At a time when humanity’s mastery of technical and mechanical prowess is arguably unconstrained, we remain humbled in fully appreciating the human condition and its manifestations. What should concern us is the runaway development of techne without commensurate accounting of influence on our sense of being. The evolution of the city, manifested in changes in color palette, scale, infrastructure, and density should only be our point of departure, starting us on a path to investigate how we, as social beings, evolve in response to these new definitions of reality.

 

This workshop seeks to enlist the very techniques powering technical development in the effort to help us better understand our concurrent evolution. Using a subset of machine learning techniques called Reinforcement Learning, participants will train virtual actors that will be used in simulations to investigate how people respond to urban landscapes. The hope is these techniques can provide a means of gaining closer intimacy with the occupants in the city, and insights into how we inhabit the city.



Methodology

This project will focus specifically on working with a multiagent AI system within Unity. Using reinforcement learning, students will train AI agents that can respond aesthetically to a simulated urban landscape. Simulations will be used as vehicles to investigate the experiences of the different actors within the ever-evolving landscape of Seoul. In the process we can speculate the possible futures of Seoul.


Students will design and train virtual AI agents based on real world observations of people within the cityscape. Observations will be used to develop theories on how the urban features are influencing behaviors of the diverse set of occupants the city must support. These theories of human-building, human-city relationships will form the foundation for the virtual Agents that will be trained and used in simulated urban environments. 


Students will be required to collect the data by exploring Seoul city, observing space, and analyzing human interactions.  

This course will divide into three parts:

1. Data Collection: Explore the city, observe how spaces influences how different people behave

2. Reinforcement Training: design AI agents based on observations

3. Simulation and Design: design potential visions of Seoul and simulate how people will behave



Design Tools

Pencil, notebook

Cell Phone, Camera

Unity Engine

Visual Studio

3D software (Rhino or Blender)



Expected outcome

Course Objective 1: Observe the Social Interaction. Students will examine a space with a new point of view to immerse themselves in an open space in the city environment.

Course Objective 2: Learning Multi-Agent System. Students will have a better understanding of the multiagent system and be able to find the resource to write the AI code for agents.

Course Objective 3: Design public space options. Students will be able to propose design solutions based on the interaction with the simulated scenario.


The primary objective of this unit is to introduce to students how to incorporate reinforcement learning in the design process and provide an understanding of data collection, and machine vision in assisting architectural design. This unit will mainly focus on ML agents in Unity.


At the end of this course, students should be able to:

1. To understand how to digitalize analogue observations.

2. To explore ideas through simulation games.

3. To train their own AI agents with a particular behaviour.



Recommended reading:

Ann Sussman., Cognitive Architecture: Designing for How We Respond to the Built Environment, 1980

Sean Lally, The Air From Other Planets, 2013, Publisher: Lars Muller.

Tony Fry, Human By Design

UNIT TUTOR

Adetokunbo Ayoade

TheLittleThings AI