Wywiad ze sztuczną inteligencją w procesie zbierania danych do projektowania. Badanie możliwości ChatGPT – trzy przypadki

Ekrem Bahadır Çalışkan

doi:10.37190/arc230309

Streszczenie

Program architektoniczny ma oczywisty i zasadniczy wpływ na formę projektowanych budowli. Wiedza na temat oczekiwań i wymagań programowych jest pozyskiwana i weryfikowana w czasie procesu projektowania wstępnego. Złożoność definiowania potrzeb użytkowych i trudności w pozyskiwaniu informacji od stron zaangażowanych w powstawanie projektu są zróżnicowane. Wywiad jest podstawowym narzędziem mającym na celu tworzenie, udoskonalanie i zarządzanie wymaganiami dotyczącymi dyspozycji przestrzennych z perspektywy użytkowników końcowych. Sztuczna inteligencja (AI) oferuje wiele możliwości w różnych dziedzinach, w tym w projektowaniu, co od niedawna poddawane jest szczegółowym analizom. W niniejszym artykule przedstawiono badania nad możliwościami wykorzystania AI do programowania budynków z punktu widzenia ich użytkowników. Wykorzystano model językowy ChatGPT, który ma zdolność przewidywania oraz generowania tekstów i informacji. Podstawowym celem badania było ustalenie granic i zdolności/precyzji AI w formułowaniu wymogów przestrzennych dla trzech różnych typów budowli. Wyniki okazały się obiecujące pod względem łatwości prowadzenia ustaleń ma linii projektant–użytkownik na etapach przedprojektowych. ChatGPT pozwolił stworzyć jasny program architektoniczny przy zintegrowaniu kontekstu i ekonomicznym wykorzystaniu czasu. Postulowany wkład AI w projektowanie architektoniczne wymaga dalszych badań, jednak przedstawiona analiza wskazuje możliwe zastosowanie sztucznej inteligencji do definiowania wymagań programowych i przestrzennych. Kierunki dalszych ustaleń mogą obejmować walidację zdobytej wiedzy, procesy potrzebne do jej zastosowania, badania AI dla różnych typów budynków i interesariuszy projektu, a także dyskusje na temat wątpliwości prawnych i kwestii etycznych.

Podgląd pełnego artykułu jest możliwy wyłącznie na większych ekranach.

2023

3(75)

Ekrem Bahadır Çalışkan*

Briefing with artificial intelligence for requirement elicitation:

Three cases with ChatGPT for exploration of possibilities

Introduction

The brieng process amongst project stakeholders is

critical for the success and correct execution of construc-

tion operations under project objectives in construction

projects. It is a method of continuing communication and

collaborative work while the building stages are complet-

ed. It is especially useful for understanding the organiza-

tion’s needs and resources and connecting them to its ob-

jectives in the statement of project requirements [1]. One

of the primary goals of the brieng process is to ensure the

elicitation of requirements. As a result of dierent proce-

dures and report preparation, customer wants are identi-

ed and articulated as a client requirement. Unfortunately,

the construction sector has a poor track record of meeting

these standards due to the execution and complexity of the

project brief [2]. Moreover, collecting and transmitting

knowledge from clients to designers or designers to clients

is a critical challenge for successful requirement process-

ing, which may be hampered by a lack of time, structure,

experience, etc. Knowledge of space requirements must

be developed, processed, and utilized with verication

and validation, just like any other knowledge. Regard-

ing requirement elicitation throughout the brieng phase,

some knowledge capture approaches used in the domain

include brainstorming, storytelling, lesson-learned tools,

post-project reviews, workshops, design proposals, or in-

terviews [3]–[6]. The involvement of project stakeholders

is necessary to apply these techniques. The absences and

less inclusion of end-users make the requirement captur-

ing process unsuccessful.

Articial intelligence (AI) has signicantly aected

many elds and businesses in the last decades. It oers

systematic capabilities of reasoning based on inputs,

learning via dierences, and prediction [7]. Because of the

frequent release of AI tools on the web, the usage and re-

search about the capabilities, opportunities, and challeng-

es have increased. Studies on educational perspectives,

health, marketing, production, design, and automation

have risen [8]. A systematic review of the examination of

AI in higher education [9] and research showing the po-

tential use of MidJourney

in architectural practice [10]

may be noted as some contemporary works. Today, one

important debate is about what makes humans dierent

from, or better than, robots or intelligent machines [11].

AI can take diverse roles in many elds, such as humans.

The stated problem about less or no end-user inclusion for

requirement elicitation in architectural projects could be

evaluated and improved to dene the space better. Thus,

this study seeks opportunities for requirement elicitation

by putting AI in the end-user position. Opening the dis-

cussion and discovering the possibilities and limits are the

main objectives of the research. The debates on the us-

age of AI, property rights, and ethics are important issues

discussed continuously, which are out of the limitation of

the research. A semi-structured interview was dened and

executed for the present research objectives to re-create

similar brieng sessions with end-users. ChatGPT is used

as AI, which has recently received much attention. The

explanations and problems in the brieng process to elicit

the space requirements focusing on end-user involvement

and the recent research studies on ChatGPT are conducted

to state the research problem and shape the ground for

the research method. The brieng

records with AI are

AI script-based image generation tool.

DOI: 10.37190/arc230309

Published in open access. CC BY NC ND license

* ORCID: 0000-0002-5258-2976. Faculty of Architecture and

Fine Arts, Ankara Yıldırım Beyazıt University, Ankara, Türkiye, e-mail:

bahadir.caliskan@gmail.com

94 Ekrem Bahadır Çalışkan

presented and discussed, and the possible outcomes for

examining a framework for requirement elicitation at the

pre-project stage of construction projects are explored in

the nal part of the research.

Research method

The research examines AI–ChatGPT, considering cap-

turing requirement knowledge by putting it in the end-us-

er position. The knowledge required for the spaces before

design phases in pre-project design tried to be captured by

brieng sessions. Within the research, a literature survey

of brieng and requirement elicitation and barriers and

problems to state problems such as end-user involvement,

lack of clarity, and capturing the requirement knowledge

in the conventional brieng process was conducted. Be-

sides, the contemporary examinations of ChatGPT are

conducted to explore possibilities and capabilities. The

brieng framework with AI is continuously structured,

including the interview features, selection of building ty-

pology, and objectives of sessions considering the require-

ment capturing. The aim is to capture the requirement

knowledge from ChatGPT and evaluate the knowledge

and the process with discussion on ndings. Parallel to

the aim, the three case studies, kindergarten, single house,

and hotel, were executed with ChatGPT. The author re-

ned the outcomes of the brieng. The rened ndings

and the brieng session were discussed, and the outcomes

were stated. The conclusion for potentials, lacks, limits,

and contributions of research are presented at the nal.

Theoretical framework

Briefing and requirement elicitation

Space requirements are knowledge that represents

a project’s objectives, individual or group demands, and

project stakeholders’ viewpoints. This knowledge must

be recorded, processed through various actions, and man-

aged for future use. Brieng is the process that contin-

ues throughout the project with the interaction of clients

and other project stakeholders to capture and manage the

knowledge for the project’s success [12]. The brieng is

a mechanism for client, contractor, and designer collabo-

ration. Client and contractor participation in brieng aims

to collaborate with the contractor to promote innovation

and eciency in planning and production [13]. An essen-

tial purpose of the design brieng process is to gather cus-

tomer needs and deliver the design project accordingly.

The problem is establishing and closing the requirement

gap between the paying client, the user client, and the de-

signer through an eective brieng process.

The construction process encompasses all procedures

from a planned construction project, such as a new build-

ing, infrastructure, or refurbishment [14]. The develop-

ment and procedure may dier according to the charac-

teristics of a building, such as kind, size, and complexity.

Complicated projects may have a bigger ow of informa-

tion regarding the knowledge management cycle, which

involves many multi-disciplinary workers and may neces-

sitate additional brieng obstacles [15]. Brieng occurs

throughout the building process and is a critical tool and

procedure each stage requires. The brieng starts long be-

fore the project and continues long after and connects to

the beginning of a new project. The brieng process is seg-

mented into three principal stages for better understand-

ing and implementing brieng into a construction project:

pre-project, project, and post-project [1]. At the pre-project

stage and initial part of the project stages, the brieng is an

important process for requirement elicitation and statement

of needs with the involvement of the client and end-users.

Barriers and problems in briefing

The problems, gaps, and development areas have been

explored and studied for three decades for the phases of

the construction process. By implementing diverse tech-

niques, studies, and technologies, the industry has been

trying to develop and increase brieng performance for re-

quirement management. Barrett oered some suggestions

for improvement, including information must be presented

in a way that is acceptable to individuals; a brief individual

taker may be more appropriate than an architect; ensuring

client and user involvement; and, nally, a neutral com-

puter-based expert system to supplement professionals’

weak areas [16]. Blyth described the brieng framework

for building project stages and gave the key features and

requirements for managing requirement elicitation during

the brieng process [1]. Some obstacles and essential con-

cerns for successful requirement processing include a lack

of open and eective communication, a lack of clarity of

objectives, a lack of complete frameworks and formaliza-

tion, and an inability to assure the engagement of end users

[17]. Interviews, questionnaires, workshops, brainstorm-

ing, drawings, collaborative working environments, and

various research studies and technologies such as Client-

Pro, CBR, and BIM are also highlighted with their contri-

bution to the demand elicitation process [17].

The expertise and knowledge of clients and end users

are crucial [1], [12], [17]. It has a direct impact on inter-

personal communication and knowledge transfer. Project

stakeholders must be knowledgeable about deciphering

and commenting on the message [18]. As a result, a lack

of communication inuences the requirement elicitation

process. Furthermore, retaining the client’s engagement

at an exact level makes collecting knowledge about space

usage dicult. A clear and successful brieng with the

client and user is needed to capture the knowledge for

requirement elicitations. The critical factors that are rea-

soning problems for capturing the requirement knowl-

edge considering the role of client and end-user are lack

of communication, objectives and decision clarity, client

inexperience, and lack of proper participation of client /

end-user in the brieng process [1], [14], [17], [19], [20].

ChatGPT as artificial intelligence

ChatGPT (Chat Generative Pre-Trained

Transformer), an

OpenAI-developed big language model, is a type of arti-

cial intelligence that may be utilized as a chatbot [21]–[24].

Briefing with artificial intelligence for requirement elicitation: Three cases with ChatGPT for exploration of possibilities 95

A substantial amount of training data in the form of tuples

of input and output data must be provided [25]. ChatGPT

was created using deep language algorithms and training

sets to communicate conversationally [26], [27]. It per-

forms various tasks, including question-answering, ma-

chine translation, and language modeling, which can also

be used to develop virtual assistants and chatbots and

generate human-like discussions [28]. For years, the rm

has focused on research and development of AI through

training with the participation of platform-free consumers.

The most recent version was the 4.0 generation at the time

of this study, which included the ancestors of ChatGPT 2.0

and 3.0 [24], [29]. The web explorer chat box is intended to

enrich freely with registration. Conversation with AI with

a simple chat box is possible, and the replies are written

quickly. The conversation might continue by referencing

previous topics in the user’s speech. The menu on the left

could be viewed to access and manage recorded talks. An

upgrade payment is oered to receive unlimited use, faster

response times, and priority access to future features.

Tests with ChatGPT are ramping up following the initial

release in November 2022. Several working and research

articles investigate various subjects and approaches’ po-

tential capabilities and limits. Seeking the possibilities in-

cludes asking it to compose three tales [21], asking about

the best and worst US Presidents [30], and asking about

the function of ChatGPT [23]. Other topics covered in the

literature include the scope, boundaries, and repercussions

of AI [29], [31] and the possible application of ChatGPT

in global warming [22]. Besides, the ethics of usage, prop-

erty rights, and humanization of the outcomes of AI are

under discussion. Since this type of AI release began, the

number of developed research within the frameworks has

increased. AI will contribute more to all businesses and

scientic elds through advancements, experimentation,

and training. Landgrebe [25] identied many critical con-

cerns that learning activities must address to begin AI: ad-

equate training data, digital data storage, consistency, and

variation. As a result, published AI projects like ChatGPT

encourage people to engage in training by allowing them

to use it freely.

Case studies: briefing with ChatGPT

Based on the ndings from the literature survey of

brieng, requirement elicitation process, and barriers, it

could be stated that lack of inclusion of end-users, prob-

lems with clear statements, and capturing the requirement

knowledge from end-users are signicant considering the

roles of users in shaping the architectural program. On

the other hand, ChatGPT, a language model trained and

learned from diverse sources, can generate responses for

a wide range of areas. In parallel with the stated research

objectives of exploring the potential of AI for the brief-

ing process, the proposed interview structure is shown in

Figure 1.

First, the questions or explanations are said to ChatGPT

from the chat box to make AI understand the context and

subject. Then, the questions are asked to capture the space

requirement of the building typology. Material oers

and features are investigated, and electromechanical re-

quirements are asked for spaces. The thought of AI for

the site, budget, and construction time is tried to learn.

The records of the whole brieng process are taken and

rened to capture requirements. Three important issues

should be noticed for this experimental brieng process.

First, re-expressing to the same context may be necessary

to keep AI connected to context and subject. Secondly,

interviews and briengs should include two-way commu-

nication. However, the chat GPT cannot ask questions or

initiate new subjects. As the last issue, the questions in the

interview ow should be updated due to responses by the

interviewer/brief taker. It is normal in any semi-structured

interview [32].

Explanations of case studies: buildings as subject

Three dierent building types are selected to execute

the requirement capturing process. They can be expanded,

and variations can be conducted for further examinations.

One is kindergarten since it is an example of an education-

al building with dierent spaces. The space organization

Fig. 1. Briefing Process with ChatGPT

(elaborated by E.B. Çalışkan)

Il. 1. Proces programowania wywiadu z ChatGPT

(oprac. E.B. Çalışkan)

96 Ekrem Bahadır Çalışkan

of kindergarten is a complex issue aected by children’s

changing physical conditions [33, p. 130]. The second

building type is single housing, in which requirements are

dened due to the personal demands of the owners. The

spaces are generally known but dier due to the users’

intentions. Every community is a complicated issue with

socioeconomic, cultural, and demographic repercussions

[34, p. 71]. The last subject is a city hotel with accom-

modations and service spaces. Besides, the material selec-

tions and electromechanical requirements are important

for any hotel to sustain maintenance and facility services.

The assumptions are needed before executing the brieng

process since the objective is to capture the requirements

of these assigned buildings with strategic denitions. The

explanations of assumptions could be listed as:

• Kindergarten; 250 student capacity, location in Turkey,

• Single House; for a family of a married couple and

one child with 100,000-dollar yearly income. The jobs

are math professor and nurse, and the location is United

States of America,

• Hotel; 3–4-star city hotel with 200 beds, location is

Ankara.

Briefing sessions

Three brieng sessions were executed on 17.01.2023

with ChatGPT. The average time is twelve minutes per

interview. Table 1 shows the general information on com-

pleted sessions.

facility with classrooms, a playground, and possibly other

amenities such as a gym or library. It would be best to

check with the relevant authorities for specic guidelines

and requirements.

Author: Can you rewrite it due to this information? For

classroom 1.5 m

per student (After told requirements).

ChatGPT: Classrooms: 1.5 m

per student, so around

375 m

for 250 students, Playground: Can vary greatly

depending on the size and design, but a rough estimate

would be around 1000–1500 m

, Multipurpose Room:

This space can be used for activities such as gym, music,

or art and it usually around 500–800 m

, Library: Around

100–150 m

, Administrative Offices: Around 200–

300 m

, Kitchen and Dining Area: Around 200–300 m

Bathrooms and Changing Rooms: Around 200–300 m

Single house

Author: Can you assume that you are a mathematics

professor married to a nurse, have 1 child, live in America

and have 100,000 dollars a year?

ChatGPT: I am an articial intelligence and do not

have the ability to assume personal characteristics such

as being a professor, being married to a nurse, having

children, living in a specic location, or having a certain

income. I can provide information and answer questions

based on my programming and the data that I have been

trained on. Is there something specic you would like to

know?

Author: I am asking the space requirements: rooms for

a single house in which a married couple with 1 child can

live.

ChatGPT: Bedrooms: A minimum of three bedrooms

is typically recommended for a family of three – one for

the parents, one for the child, and one for guests or as an

oce, Living Room: A comfortable and spacious living

room for family time and entertaining, Kitchen: A func-

tional kitchen with enough counter space and storage for

meal preparation and storage, Dining Room: A separate

dining room for family meals, Bathrooms: A minimum of

two bathrooms, one of which should be an suite in the

master bedroom, Laundry: A dedicated laundry room with

space for a washer and dryer, Outdoor Space: A backyard

or outdoor living space for activities and relaxation.

Hotel

Author: List the space with area in m

and numbers,

and properties for this building.

ChatGPT: Guest rooms: 100–150 m

per room. (De-

pending on room type and amenities), Lobby and recep-

tion area: 50–75 m

, Dining area: 100–150 m

, Conference

room(s): 50–75 m

per room, Fitness center: 50–75 m

administrative oces: 50–75 m

, Parking: 100–150 m

Author: Guest rooms are so big. Are you sure?

ChatGPT: I apologize, the estimate I provided for guest

room size may have been too high. The average guest room

size in a hotel is typically around 20–30 m

per room.

However, it can vary depending on the type of hotel and

amenities oered in the room. Luxury hotels, for example,

may have larger guest rooms with more amenities, while

Kindergarten Single House Hotel

Duration 10 11 15

Number of

questions asked

9 12 15

Number of words

for responses

1998 2514 2808

Repeating and

refinement questions

2 3 4

Table 1. Information of Briefing Sessions

(elaborated by E.B. Çalışkan)

Tabela 1. Podstawowe dane na temat przeprowadzonych wywiadów

(oprac. E.B. Çalışkan)

All of the recordings are not shared since 7320 words in

total are recorded for three sessions. The owing sections

explain and evaluate the captured knowledge and the pro-

cess. Some samples from the sessions are presented below.

Kindergarten

Author: Can you tell the space requirements of kinder-

garten for 250 students in Turkey?

ChatGPT: The space requirements for a kindergarten

with 250 students in Turkey may vary depending on the

specic regulations and guidelines of the Turkish gov-

ernment or local education authorities. In general, a kin-

dergarten for 250 students would likely require a large

Briefing with artificial intelligence for requirement elicitation: Three cases with ChatGPT for exploration of possibilities 97

budget hotels may have smaller guest rooms. It’s important

to note that the space requirements will vary depending on

the hotel’s layout, design, and the amenities oered.

Evaluation of the briefing process

Evaluations are taken in two sub-parts. The rst one

presents the rened requirements captured from the brief-

ing process. Secondly, the ndings about the brieng pro-

cess are explained. Consequently, the merged ndings and

experiment are discussed to state the possibilities, lacks,

and limits.

Elicited requirements

Tables 2–5 show the brief requirements of kindergar-

ten, single house, and 3–4-star city hotel orderly. Because

of standing within the article’s limits, only the main head-

ings and features are presented in the tables. However, the

explanations, adjacency relations, design criteria, and fea-

tures of all spaces were given by ChatGPT.

Spaces

Area [m

]

(totals)

Dimensions

(w*d*h) [m]

Window Area Electromechanical Colors

Classrooms 375 8*6*3 10%

lighting, HVAC, fire alarm,

audio-visual

neutral colors

Play Ground 1250 25*50*4

outdoor

space

lighting, irrigation bright colors

Multipurpose Hall 625 15*15*4 10%

lighting, HVAC, fire alarm

and audio-visual, stage lighting

gray

Library 125 10*12*3 10% lighting, HVAC, audio-visual bule, pale yellow

Offices 200 10*10*2.7 10% lighting, HVAC, audio-visual neutral colors

Kitchen and Dining 250 20*20*3 10%

lighting, HVAC, fire,

kitchen equipment

red

Bathroom

and Services

200 2*2*2.7 unnecessary lighting, plumbing gray

Table 2. Requirements of 250-students kindergarten (generated from briefing) (elaborated by E.B. Çalışkan)

Tabela 2. Wymagania programowe dotyczące przedszkola na 250 dzieci (stworzone na podstawie wywiadu) (oprac. E.B. Çalışkan)

Spaces

Area [m

]

(totals)

Explanation

Bedrooms × 3 45

a minimum of three bedrooms is typically recommended for a family of three – one for the parents,

one for the child, and one for guests or as an office

Living Room 35 a comfortable and spacious living room for family time and entertaining

Kitchen 18 a functional kitchen with enough counter space and storage for meal preparation

Dining Room 12 a separate dining room for family meals at the main level

Bathrooms × 2 12 a minimum of two bathrooms, one of which should be in the master bedroom

Laundry 6 laundry room with a washer and dryer

Outdoor Space 60 a backyard or outdoor living space for activities and relaxation

Table 3. Requirements of the single house (generated from briefing) (elaborated by E.B. Çalışkan)

Tabela 3. Wymagania programowe dotyczące domu jednorodzinnego (stworzone na podstawie wywiadu) (oprac. E.B. Çalışkan)

Findings of the briefing process

• AI cannot assume itself directly in any position of

personal characteristics; however, by including the nec-

essary explanations.

• In an average of 20-minute sessions, brieng sessions

were executed with a considerable amount of response.

• The unit of dimensions and same space recommen-

dation like çardak, selamlık

changed due to project lo-

cation. However, there are no specific variations for

locations.

• Site recommendation is unavailable. The general lay-

out of the lands was explained.

• The budget and time of construction were estimated

in general. AI wanted further knowledge about material

selections, labor costs, etc.

Çardak and Selamlık are spaces from historical Turkish Houses.

The AI automatically relates the house issue to Turkish House historical

context.

98 Ekrem Bahadır Çalışkan

Spaces Area [m

] (totals) Colors

Guestrooms × 200 15–20 beige, gray

Lobby and Reception 50–75

• public spaces: brighter colors such as yellow

• bathrooms: neutral colors like white, gray

• corridors: neutral colors

Dining Area 75–100

Conference room *2 100

Fitness Center 25–50

Offices 25–50

Parking 100–150

Table 4. Requirements of the hotel (generated from briefing) (elaborated by E.B. Çalışkan)

Tabela 4. Wymagania programowe dotyczące hotelu (stworzone na podstawie wywiadu) (oprac. E.B. Çalışkan)

Table 5. Material and Electromechanical requirements of the hotel (generated from briefing) (elaborated by E.B. Çalışkan)

Tabela 5. Wymagania materiałowe oraz sprzętowe (elektryczne) w hotelu (stworzone na podstawie wywiadu) (oprac. E.B. Çalışkan)

Item Explanation

Flooring

carpet, tile, and hardwood are popular options for guest rooms and common areas;

ceramic tile is also commonly used in bathrooms and laundry rooms

Walls drywall or plaster is commonly used for interior walls, while exterior walls may be made of brick, stone, or stucco

Ceilings

drywall or plaster is commonly used for interior ceilings, while acoustic tile is commonly used in guest rooms

and common areas to improve sound insulation

Openings aluminum or vinyl windows and doors are commonly used for energy efficiency and durability

Lighting LED lighting is commonly used in guest rooms and common areas for energy efficiency and long life

HVAC

a central heating and cooling system, as well as individual thermostats in each guest room, to ensure a comfortable

temperature for guests. The system should also include air filtration and ventilation to maintain good indoor air quality

Plumbing

a central water supply and waste management system, as well as individual fixtures such as sinks, toilets,

and showers in guest bathrooms

Electrical

a central electrical distribution system, as well as individual outlets and lighting in each guest room and common areas;

the hotel may also require backup generators in case of power outages

Fire protection smoke detection and fire alarms in all areas of the hotel, as well as fire sprinklers and fire extinguishers in common areas

Security

CCTV cameras and card-key access control systems in all areas of the hotel, including guest rooms,

to ensure the safety and security of guests

• AI revised the responses due to newly given parame-

ters. For example, the areas were recalculated due to unit

area information.

• The proper responses were given when the author

warned about vital facts. For example, a guest room for

a standard city hotel could not be 150 m

• Explanations, design criteria, and features of spaces

were given in detail regarding contemporary consider-

ations.

• Quantitively, calculations and relations were executed

rapidly and accurately.

• Color and material recommendations are related to

the subject and context.

• Electromechanical requirements for building installa-

tions were given in coherence with general specications.

The diversity between spaces is promising.

• The dimensions of spaces and the proportion of open-

ings were given in detail.

• The levels and adjacency relations of spaces were ex-

plained.

• The context and subject need to be re-stated to get

more related responses.

• The responses were rst generated from typical ex-

amples and then developed by re-asking the questions

with further explanations.

• The structure of responses is formal and easy to un-

derstand.

Possibilities, lacks, and limits

For any brieng process, sustaining the involvement

of the end-user and client for capturing the requirement

knowledge in sucient time and clearly is vital. The in-

tentions and needs of clients specically enlighten the

design process with captured and rened requirement

knowledge. With problems and lack of these processes

Briefing with artificial intelligence for requirement elicitation: Three cases with ChatGPT for exploration of possibilities 99

underlined in the literature survey, AI was tried to be

a knowledge source. The outcomes are promising in a lev-

el of detail. ChatGPT can structure the response for gur-

ing out the general requirements of any building. When

the brieng process is taken further, it can generate the re-

lation of spaces, materials, and features. Knowledge gen-

eration is rapid, so the brief taker can arrange questions to

get requirements by evaluating answers. For this reason,

a brief taker’s experience is vital to govern the whole pro-

cess. However, it was also vital in the conventional brief-

ing process. The AI can stand in a position of integrity and

continuity for interviews; generally, sometimes, conduct-

ing explanations in a conversation is necessary.

The explanations, considerations, and features that

ChatGPT underlines are convenient to contemporary is-

sues in architecture. The structure of responses has value

considering the context. In this manner, AI also uses this

capability to respond by not giving specic responses

to questions. The typical requirement recommendations

should be repeated by adding new explanations to ques-

tions to capture valuable knowledge. The units of dimen-

sions or some details of spaces were generated due to

location dierences; however, the knowledge of require-

ments was found to represent general thoughts. Besides,

it cannot oer site, budget, and time of construction

work. Perhaps it will be better to ask for the references of

responses to make the requirement more related to con-

text and subject.

Any roles or personal characteristics could not be as-

signed to ChatGPT. It would be better and benecial if it

could be. Thus, the authors carefully executed the inter-

view for every step and session by stating the explanation

related to the building subject. Some misunderstandings

and illogical responses were observed, like space areas or

material recommendations. However, it was very success-

ful in giving a response and correcting the process after

the warning. Also, the rapid calculation capability makes

it advantageous for generating knowledge for revised sit-

uations. An important contribution is creating and listing

all the issues iteratively by conducting new inputs. Color

recommendations, material assignments, and electrome-

chanical requirements present the general layout of the

spaces. For every space of buildings, the brief taker can

rene and create the requirement specications by look-

ing at the records.

Discussions on the experiment

The study is an experiment using ChatGPT to elicit

the requirement of buildings by putting it in an end-user

position. Based on the interview records, rened require-

ments, and evaluations, this paper underlines the possible

utilization of AI in the pre-project stage of architectural

design, which has positive and negative contributions. On

the positive side, the usage of AI as a limitless knowledge

source of requirement in faster and more detailed time in-

tervals. The uncertainty of the captured knowledge repre-

sentation ratio for end-users must be stated as a negative

side. The studies and experiments need to be conducted to

validate the accuracy of captured knowledge.

Another important situation is the experience of the

brief taker on requirements and the trained level of AI

about the interview subject. ChatGPT can be further

trained by giving documents and reports about the in-

tended area, and the brief taker can expect the responses

accordingly. The experience and knowledge level of the

interviewer directly aect the evaluation of the response,

which should be later put in a validated framework to be

more objective. Although a level of experience is needed,

the study shows the potential to create valuable responses

for requirements.

The architectural design and construction industry have

many stages than requirement elicitation, such as design,

3D rendering, collaboration, or tendering. ChatGPT and

other AI tools may be implemented in these stages and

may have important implications. The application proce-

dures and authorship are important issues to be solved and

agreed upon. Many practical implications of AI in design

can be found online, and research on them has risen. By

involvement of experts and producing valuable studies,

the future directions can be cleared.

Conclusion

The presented experiment is a premise attempt to im-

plement articial intelligence in a brieng process for

requirement elicitation of spaces. The study argues that

AI releases like ChatGPT can be a project stakeholder,

especially representing end-users, to dene and devel-

op architectural programs and features of spaces. Archi-

tects always confront the problem of dening clear and

structured requirements to start and develop the design

of buildings at the pre-design process. Thus, knowledge

sources, like surveys, brieng with project stakeholders,

experience, and observations, contribute. Examining arti-

cial intelligence with semi-structured interviews shows

the potential requirement for capturing the environment in

the brieng process. The lack is related to the inability to

assure two-way communication and sustain the continuity

and integrity of communication. The challenge of com-

munication with AI diers from that of humans in trans-

ferring and getting emotions. However, it can be directed

to any level by implementing new prompts. Besides, the

responses are to be more elaborated than general state-

ments considering the explained subject.

On the other hand, the generation capability for re-

quirements, including asked items and the responses’

structure and value, uncover the possible integration and

implementation of AI in requirement elicitation. An im-

portant bottleneck of this study is to verify and validate

the utilization of outcomes of the brieng with AI. Two

possible frameworks must be considered to evaluate the

consistency of AI-generated demands and human beings’

actual wishes. One is to use recorded brieng reports and

re-create the process with AI, and the second is to exe-

cute the same possible cases with AI and Humans simul-

taneously. However, with underlined further progress in

briengs with articial intelligence, the study contributes

to the possible implementation and usage with revealing

ndings. Within the conduction of ethical concerns, the

100 Ekrem Bahadır Çalışkan

theoretical and practical usage foundation of AI in knowl-

edge production can radically change the processes of

requirement elicitation and validation, especially in the

cases of end-user involvement. Further examinations to

explore capacity and develop a convenient model for im-

plementation in the brieng process are needed; however,

this study underlines the utilization of articial intelli-

gence for requirement elicitation with possible contribu-

tions and limits.

Translated by

E.B. Çalışkan

[1] Blyth A., Worthigton J., Managing the brief for better design, Rout-

ledge, London 2010, doi: 10.4324/9780203857373.

[2]

Shahrin F., Johansen E., Lockley S., Udeaja C., Eective Capture,

Translating and Delivering Client Requirements Using Building In-

formation Modelling (Bim) Technology, [in:] Arcom Research Work-

shop On Decision-Making Across Levels, Time And Space: Exploring

Theories, Methods And Practices, EPSRC, Manchester 2010, 38–45.

[3] Kamara J.M., Anumba C.J., Carrillo P.M., Bouchlaghem D., Con-

ceptual framework for live capture and reuse of project knowledge,

“International Conference on Information Technology for Con-

struction” 2003, January, 178–185.

[4] Tan H.C., Anumba C.J., Carrillo P.M., Bouchlaghem D., Kamara J.,

Udeaja C., Capture and reuse of project knowledge in construction,

Wiley-Blackwell, 2010, doi: 10.1002/9781444315448.

[5] Al-Ghassani A., Improving the structural design process: A know l-

edge management approach, PhD thesis, Loughborough Universi-

ty, Leicestershire 2003.

[6] Pourzolfaghar Z., Ibrahim R., Abdullah R., Adam N.M., A tech-

nique to capture multi-disciplinary tacit knowledge during the con-

ceptual design phase of a building project, “Journal of Information

and Knowledge Management” 2014, Vol. 13, No. 2, doi: 10.1142/

s0219649214500130.

[7] Dwivedi Y.K., Kshetri N., Hughes L. et al., “So what if ChatGPT

wrote it?” Multidisciplinary perspectives on opportunities, chal-

lenges and implications of generative conversational AI for re-

search, practice and policy’, “International Journal of Information

Management” 2023, Vol. 71, doi: 10.1016/j.ijinfomgt.2023.102642.

[8] Mondal S., Das S., Vrana V.G., How to Bell the Cat? A theoretical

review of Generative Articial Intelligence towards digital disrup-

tion in all walks of life, “Technologies” 2023, Vol. 11, No. 2, 44,

doi: 10.3390/technologies11020044.

[9] Crompton H., Burke D., Articial intelligence in higher education:

the state of the eld, “International Journal of Educational Tech-

nology in Higher Education” 2023, Vol. 20, No. 22, doi: 10.1186/

s41239-023-00392-8.

[10] Jaruga-Rozdolska A., Articial intelligence as part of future prac-

tices in the architect’s work: MidJourney generative tool as part of

a process of creating an architectural form, “Architectus” 2022,

nr 3(71), 95–104, doi: 10.37190/arc220310.

[11] Bozkurt A., Xiao J., Lambert S. et al., Speculative Futures on

ChatGPT and Generative Articial Intelligence (AI): A Collec-

tive Reection from the Educational Landscape, “Asian Journal of

Distance Education” 2023, Vol. 18, Iss. 1, 53–130, doi: 10.5281/

zenodo.7636568.

[12] Barrett P., Stanley C., Better construction brieng, Wiley-Black-

well, Oxford 1999.

[13] Ryd N., The design brief as carrier of client information during

the construction process, “Design Studies” 2004, Vol. 25, Iss. 3,

231–249, doi: 10.1016/j.destud.2003.10.003.

[14] Olatokun E., Pathirage C., Importance of knowledge capturing

(KC) in the design brieng process in the construction industry,

[in:] 12

International Postgraduate Research Conference (IPGRC

2015), 2015, 1–16.

[15] Kamara J.M., Anumba C.J., Hobbs B., From brieng to client re-

quirements processing, [in:] W. Hughes (ed.), 15

Annual ARCOM

Conference, 15–17 September 1999, Liverpool John Moores Uni-

versity, Association of Researchers in Construction Management,

Vol. 1, 317–326.

[16] Barrett P.S., Hudson J., Stanley C., Good practice in brieng: the

limits of rationality, “Automation in Construction” 1999, Vol. 8,

Iss. 6, 633–642, doi: 10.1016/s0926-5805(98)00108-3.

[17] Pegoraroa C., Paula I.C., Requirements processing for building de-

sign: A systematic review, “Producao” 2017, Vol. 27, 1–18, doi:

10.1590/0103-6513.212116.

[18] Norouzi N., Shabak M., Bin Embi M.R., Khan T.H., The archi-

tect, the client and eective communication in architectural de-

sign practice, “Procedia – Social and Behavioral Sciences” 2015,

Vol. 172, 635–642, doi: 10.1016/j.sbspro.2015.01.413.

[19] Faatz S., Architectural programming: providing essential knowl-

edge of project participants needs in the pre-design phase, “Orga-

nization, Technology & Management in Construction: An Interna-

tional Journal” 2009, Vol. 1, No. 2, 80–85.

[20] Olatokun E.O., Requirement elicitation using knowledge captur-

ing (Kc) techniques during the client brieng process for improved

client satisfaction in the UK construction industry, University of

Salford, Salford 2017.

[21]

McGee R.W., Annie Chan: Three short stories written with ChatGPT,

2023, February 15, https://ssrn.com/abstract=4359403.

[22] Biswas S.S., Potential use of Chat GPT in global warming, “An-

nals of Biomedical Engineering” 2023, Vol. 51, Iss. 6, 1126–1127,

doi: 10.1007/s10439-023-03171-8.

[23] Biswas S., Prospective role of ChatGPT in the military: According

to ChatGPT, “Qeios” 2023, 1–19, doi: 10.32388/8wyyod.

[24] ChatGPT, 2023, https://openai.com/blog/chatgpt [accessed:

16.03.2023].

[25] Landgrebe J., Smith B., Making AI meaningful again, “Synthese”

2021, Vol. 198, No. 3, 2061–2081, doi: 10.1007/s11229-019-

02192-y.

[26] OpenAI, 2023, https://openai.com/ [accessed: 13.03.2023].

[27] Caucheteux C., Gramfort A., King J.R., Deep language algorithms

predict semantic comprehension from brain activity, “Scientic

Reports” 2022, Vol. 12, No. 1, 1–10, doi: 10.1038/s41598-022-

20460-9.

[28] Aydın Ö., Karaarslan E., Is ChatGPT leading generative AI? What

is beyond expectations?, “Academic Platform Journal of Engi-

neering and Smart Systems” 2023, 11(3), 118–134, doi: 10.21541/

apjess.1293702.

[29] Floridi L., Chiriatti M., GPT-3: Its nature, scope, limits, and con-

sequences, “Minds and Machines” 2020, Vol. 30, Iss. 4, 681–694,

doi: 10.1007/s11023-020-09548-1.

[30] McGee R.W., Who were the 10 best and 10 worst U.S . Presidents?

The Opinion of ChatGPT (Articial Intelligence), 2023, February,

doi: 10.2139/ssrn.4367762.

[31] Yang X., Li Y., Zhang X., Chen H., Cheng W., Exploring the limits

of ChatGPT for query or aspect-based text summarization, “arXiv”

2023, February, doi: 10.48550/arXiv.2302.08081.

[32] Fellow R., Liu A., Research methods for consrtuctions, Wiley-

Blackwell, Hoboken 2008.

[33] Yalçın M., Bozdayı A.M., Hakan E.M., Cultural determinants

within the design set up of kindergarten and preschool interiors:

Assessment of four typologies in terms of their spatial formation,

“MEGARON” 2017, Vol. 12, No. 1, 130–144, doi: 10.5505/mega-

ron.2017.49469.

[34] Sarioğlu-Erdoğdu G.P., Problems in housing research and compar-

ative housing studies, “MEGARON” 2011, Vol. 6, No. 2, 71–78.

References

Briefing with artificial intelligence for requirement elicitation: Three cases with ChatGPT for exploration of possibilities 101

Abstract

Brieng with articial intelligence for requirement elicitation:

Three cases with ChatGPT for exploration of possibilities

Architectural program is important knowledge of space requirements aecting the design besides context, concept, and intentions. The re-

quirement knowledge is elicited and validated between project stakeholders during the pre-design and design process. The complexity of dening

requirements and the diculty capturing knowledge from project parties vary. Brieng is a vital process to capture, rene, create, and manage space

requirements from end-users. Lacks of communication, the inclusion of user clients, and clarity of objectives are critical for eliciting requirements.

The AI tools oer many possibilities to diverse areas, and several studies have been conducted recently. The research seeks the usage of AI to capture

the building requirement by putting it in the position of the end-user. For this purpose, ChatGPT was used, a language model capable of generating

texts and predicting and creating knowledge. The study’s objective is to explore AI’s limits and capabilities for the brieng process to initiate the

requirement elicitation. Within the semi-structured interview process, the brieng session was executed for three cases by putting the AI in the end

user position. The results are promising in the requirement elicitation of pre-project stages considering the clarity of the architectural program, integ-

rity in context, and usage of time. The argued contribution of AI in architectural projects requires further research; however, the study underlines the

possible usage of AI for dening the requirements of the spaces. The avenues of further research may include validating captured knowledge pro-

cesses needed, examining the AI for dierent building typologies and project stakeholders, and seeking the bias, ethical concerns, and combinations

of humans and AI for various tasks.

Key words: articial intelligence, brieng, requirement capturing, ChatGPT, end-user

Streszczenie

Wywiad ze sztuczną inteligencją w procesie zbierania danych do projektowania.

Badanie możliwości ChatGPT – trzy przypadki

Program architektoniczny ma oczywisty i zasadniczy wpływ na formę projektowanych budowli. Wiedza na temat oczekiwań i wymagań pro-

gramowych jest pozyskiwana i werykowana w czasie procesu projektowania wstępnego. Złożoność deniowania potrzeb użytkowych i trudności

w pozyskiwaniu informacji od stron zaangażowanych w powstawanie projektu są zróżnicowane. Wywiad jest podstawowym narzędziem mającym

na celu tworzenie, udoskonalanie i zarządzanie wymaganiami dotyczącymi dyspozycji przestrzennych z perspektywy użytkowników końcowych.

Sztuczna inteligencja (AI) oferuje wiele możliwości w różnych dziedzinach, w tym w projektowaniu, co od niedawna poddawane jest szczegółowym

analizom. W niniejszym artykule przedstawiono badania nad możliwościami wykorzystania AI do programowania budynków z punktu widzenia ich

użytkowników. Wykorzystano model językowy ChatGPT, który ma zdolność przewidywania oraz generowania tekstów i informacji. Podstawowym

celem badania było ustalenie granic i zdolności/precyzji AI w formułowaniu wymogów przestrzennych dla trzech różnych typów budowli. Wyni-

ki okazały się obiecujące pod względem łatwości prowadzenia ustaleń ma linii projektant–użytkownik na etapach przedprojektowych. ChatGPT

pozwolił stworzyć jasny program architektoniczny przy zintegrowaniu kontekstu i ekonomicznym wykorzystaniu czasu. Postulowany wkład AI

w projektowanie architektoniczne wymaga dalszych badań, jednak przedstawiona analiza wskazuje możliwe zastosowanie sztucznej inteligencji do

deniowania wymagań programowych i przestrzennych. Kierunki dalszych ustaleń mogą obejmować walidację zdobytej wiedzy, procesy potrzebne

do jej zastosowania, badania AI dla różnych typów budynków i interesariuszy projektu, a także dyskusje na temat wątpliwości prawnych i kwestii

etycznych.

Słowa kluczowe: sztuczna inteligencja, wywiad, uchwycenie wymagań, ChatGPT, użytkownik