DE-BOT
An AI-aided Design System
The project DE-BOT is a research project for a question that how AI being integrated into a design process. The research began with the existing machine learning technologies then focused on a Human-Machine-Interaction model. The purpose is to build up a general principle system for designing an AI product.
Projrct type /
Diplom project
Role /
Product designer
Agency /
Hfg-Offenbach
Year /
2022
Research & Froschung
The whole system treated as an AI assisting design service circuit consist of 3 main sub-concepts: The Conversational User Interface, The Text to Form Generation Function and The Package Recycling System. Each part connected with others as a whole functional system proposing on benefiting the professional and amateur designers.
The current industrial design process is a labor oriented workflow. The designer is in the same position as a machine, because most of the work in the process is repeated. The creation process can be separated in Evaluation, Strategies and Mechanical works, which are comparable to the working process of the machine learning process: Discrimination, Input Data and Generation.
Our tools are extensions of our purposes, and so we find it natural
to make metaphorical attributions of intentionality to them.
-- John R Searle
System & Constructions
The AI system of the project DEBOT was designed to achieve not only the function of text to form generation, but also the fluent and intuitive interaction, which brings the user an ideal interaction method to communicate with AI. The system consists of 4 main parts namely the User Interface, the Semantic Filter, the Generator and the Final Filter. The user interface was integrated with the GPT-3 chatbot, which has the ability to have customized fluent communication with humans. The semantic filter was based on the GPT-3 technology and can abstract the key ideas from a complex text with the guide of preset examples.
User-Interface Design
The user interface contains a GPT-3 Chatbot from the company OpenAI, and a visual figure.
The chatbot allows users to have a natural conversation with the program and it applies a design guide about the product details.
The GPT-3 Chatbot was developed by OpenAI with 175 billion machine learning parameters to achieve a semantic understanding of human language and fluent human-like replies in the same context. The chatbot was integrated into the conversational user interface to enhance the feeling of intuitive interaction.
AI Figure Test
A test program was developed to the question: what should an AI user interface look like?
The tested user was arranged to talk with the GPT-3 chatbot about any topics while facing different AI figures, from an abstract geometry to a real-time video human figure.
The feeling evaluation is based on the idea of Calibration of Trust. After each section of talking with different figures, the user should record their feeling of trust during the conversation. From T0 to T5 means from extreme untrusted to extreme over trusted.
The test results show different phenomenons, which were highly related to the feeling of the AI figures, and became important guides for the design of the user interface. As for the idea of calibration of trust, the ideal feeling status is located in T3, which points to the figure in the test 3 phase.
AI Figure Design
Three key words for the design principles were defined: continuity, initiative and reliability. Based on the principles the ideas of the AI interface were generated. The figure concepts were developed from a 2 dimensional reactable geometry to a 3 dimensional environment interactable metaball based form.
The design of AI visualization was defined in 4 status as shown above. Based on the human interactions, the transformation of bubble form represented the Additional Reactions and the color transform represented the Status Symbols.
Prototype video, please turn on the sound
Semantic Filter and Form Generator
The semantic filter catches the key descriptions from the conservation and sends them to the generator.
After the user interface, comes the semantic filter, which is based on the GPT-3 language processing technology from OpenAI. The filter judges the conversation contents via setting the preset examples for comparison. For example hier shows the judgment result as positive while the input text has the high semantic similarity with the preset text.
The main technology implemented in this project is called Text to Form which is based on the functions of generative adversarial neural networks. The program processed the points cloud data in a 3D model database and connected via GAN training with the pretrained VAE data from a description text database. The forms were finally generated following the input descriptions with data of cloud points and then were transformed into voxel groups, and finally became specific forms.
The program is based on the pretrained project of Tyler Habowski in the following link.
Logistic
Logistics is another important part of the service system. Regarding the thinking of sustainable economy and autonomous production , packaging technologies should be developed and fit the trend. With the additive manufacturing technologies and community production, the package could no longer be as today, but integrated together with the product itself with the same material, which highs up the efficiency of the autonomous manufacturing as well as the recycling.
The usage scenario is based on the logistical autonomous technologies in the coming future, with which the package generation was set and under 4 basic standards:
•Good for Transportation and Storage;
•Good for Protection of Product;
•Easy for Additive Manufacturing;
•Easy for Users to Use.
Different forms of the integrated package were generated and evaluated under the 4 standards: Transportation and storage, Protection of the product, Easy to manufacture, Easy to break.
The program of package generation was developed in the grasshopper. The final package form was selected with the balance of evaluation factors.
The advantage of the integrated package lies in the reduction of using extra materials. The net structure applies a full protection as well as an easy-to-break user experience. The broken package material can be recycled and reproduced in the community production circuit.
The recycling process of the package was based on the additive manufacturing and the removable package structure. Users can easily take the package off the product by themselves and transport it back to the manufacturing circuit.
With the generation program the package structure fits all different sizes of products
