1. Admit there's a problem → Designers need help.
There is an increasing need to support designers at the early conceptual design stage of the engineering design process. Why do designers need help?
- Risk aversion — The majority of resources spent bringing a product to market are committed by decisions made in the first 10% of the design activity. This adds significant risk to design decisions made at the concept development stage.
- Accelerating marketplace demand — Designers are under pressure to develop innovative products and position them accurately in marketplace.
- Increasing complexity — The systems that support the process of product design, production, and distribution have become more complex.
2. Accept the root of the problem → It's hard to isolate, express, and manage inspirational ideas.
It's hard to filter signal from noise — to focus inspiration into a great idea. Designers feel this pain every day. To find inspiring idea nuggets in mountains of garbage. Even worse, if you have a great idea, but struggle to find the right way to bring it to form, position it in the market, or communicate it to others. This is where designers need help.
3. Explore potential solutions → Generative AI's are part of the solution.
The term 'Generative AI' refers to software that can use existing data to generate new data. For example, you can train a generative AI program using various types of content like text, audio, images, or video to create new content that is similar to the content fed into the system. Generative AI assistants, powered by Generated Adversarial Networks (GANs), are slowly entering the creative practitioner's toolbox.
GANs are a deep-learning-based type of generative model based on a game theory. A generator network produces data that it 'thinks' are similar to its training data. A discriminator network tries to distinguish between generated samples and samples taken from the training data. The output of this game is synthetic data that is similar to the training data.
Think of the generator like a counterfeiter, trying to make fake money, and the discriminator as being like police, trying to allow real money and catch counterfeit money. To succeed in this game, the counterfeiter must learn to make money that is indistinguishable from genuine money, and the generator network must learn to create samples that are drawn from the same distribution as the training data.— NIPS 2016 Tutorial: Generative Adversarial Networks, 2016.
GANs have gained significant media attention for their ability to generate synthetic data (specifically images and sounds) of increasing believability. As their ability to generate compelling synthetic data improves, we'll see them applied to more and more tools. Integration of GANs-based generative models into practical creative tools is an extremely active area of AI research. We believe the application of GANs to practical design tools is one of the most exciting (and potentially impactful) areas of AI development. The potential to enhance designers' creativity, productivity, and design horizons are enormous.
4. Filter solutions → Find the right tool for you
Developments in GANs are enabling incredible advancements to computer graphics, computer vision, and user-aided design research. As this research is translated into professional design tools, the capabilities of designers to make magic happen will explode! A designer’s primary purpose is to generate an optimal artifact given a set of constraints. Depending on the artifact to be created and the constraints on development — GANs-driven tools are already very capable. For example, in the automotive design space, three GAN-driven tools are:
- Concept rendering: GAN learns from past renders to accelerate render generation (see Vizcom).
- 3D surfacing: GAN learns from past 3D models to accelerate 3D concept modelling (see BlankAI!!!)
- Aerodynamic optimization: GAN learns from simulation data to present quick simulation results (see Neural Concept).
5. Understand limitations → Don't panic! AI is not taking your design job.
Human designers create solutions while keeping in mind a generally applicable model of the world and how it works. They are agents of empathy that create designs via skills and context learned through diverse and wide-ranging experiences and training. Generative AI's don't 'understand' design concepts - they learn statistical representations of them using precisely curated and limited data sets. Their generated 'ideas' are interpretations of these statistical models that can fail in unpredictable ways. They will require humans to understand and edit their outputs for a long time. Remember, how you frame and communicate the nuances of a design solution can't (at least not yet) be communicated statistically. Product design and development require significant common sense, so it will likely be a long time before the full process can be automated. Breath easy!
As a designer, you should not feel threatened by AI. Your power as a designer is not just your ability to create renders, create 3D forms, or perform a CFD simulation. Your superpower lays in your ability to understand and define constraints within the context of your design challenge. AI-design assistants can search, generate, and optimize well-structured design information but they can't make empathetic and intuitive leaps between design concepts. This is a fundamental truth about AI systems that are often overlooked.
Fundamentally, machine learning provides a mechanism for conveying experiential knowledge about real-world designs into a format that is tangible to machines. As the field of AI evolves, its application to design workflows will expand. So stay sharp!
6. Accept reality → Don't fight them, join them.
AI as a research field is quite mature and is developing rapidly. There's lots of progress in the research, but effectively linking this to high-impact professional design applications is a non-trivial challenge. At the root of this challenge is the need to develop collaborative approaches to human–AI interaction. There is a significant knowledge gap between ML researchers and design practitioners when it comes to the practical limits of what ML systems can and cannot achieve. So no stress, AI is not taking your job any time soon! Similarly, because a GANs-based systems learn about design from their training data, the challenge for tool developers is to create interfaces that encourage designers to explore new and novel ideas rather than replicate old ones. With BLANK we address this problem by enabling design teams to blend generated ideas together to discover unique form possibilities.
To learn more about how BlankAI can accelerate your automotive design process, please reach out for a demo!