Photo by Keijiro Takahashi
The time of human beings as the only entity that can think creatively is fast coming to an end, and in fact, we may no longer be the top dogs when it comes to creative thinking. This is because machine learning and artificial intelligence are now creating new and novel ideas for works of art, videos, and product solutions using generative models, allowing datasets to grow independently.
This article will discuss a few different types of creative machine learning categories and examples of each.
The first use case for creative models and GANs (generative adversarial networks) is in the case of natural language processing. NLP generative models are commonly used to identify a specific type of communication pattern in one or multiple types of languages.
Human behavioral signal processing is a major topic right now for study and growth and has been a focus for generative models. Known as Behavioral Signal Processing, this technology is being used by educators to determine if a student might have a learning disability by identifying ways that a student understands and communicates with others.
Not only have these BSP models been able to identify disabilities, but they have also found patterns in socioeconomic statuses and the way financial statuses change the way people talk. More advanced computer support teaching has also come from BSP by changing how a chatbot responds to the individual who needs specifically curated communication.
Each person has a unique writing style, and reading this can sometimes be hard. Generative models are used in OCR (optical character recognition) to identify and create a computer’s unique handwriting, not just the font but the individual way words are written. Each generated letter is tested against a confirmed version of that letter until it has created something close enough to the correct letter that it can be identified properly.
Most famously, generative models can be used to create unique art pieces. Once thought to be only a human capability, there is now a multitude of publicly available software to code or play with to generate art pieces. NightCafe is a publicly available art-generating program that allows you to type in random words and generate a painting based on each term. The works of art are never the same, even when using the same words.
AI can be used to make videos, real images (not just painted), and even music. But there’s no need to worry – using generative models is not the end of human creativity, even though AI tools can probably make better paintings and photos than people. These tools just make it easier for creatives to make their visions a reality. And if they can code a little bit, there are lots of open source programs that can be used to make their own software that generates art based on their style and tastes.
Sometimes it’s hard to determine the value of a product you just made. So let machine learning do it for you. If your product fits a specific industry with very few references, you can use GANs to generate data based on that small information you have. For example, when determining how to price a new car that uses hydrogen and battery power, you can use a GAN with the input data of gas, electric, and hydrogen cars to generate the value that your vehicle should best be priced at.
Another example is businesses that create medical software. Since buying medical images is expensive, these companies can use a GAN model to generate their own. Marketing has seen the use of generative models in making better advertisements as well. If you only have one photo of a product, like a shirt or a burger, GANs can generate new images for you based on that single one, allowing for a quick turnaround on requests for more content.
Similar to the case of generative art mixed with medical images, architects and engineers may soon find that the demand for their services is dropping due to ML-created houses.
Getting data to train a model to design a house is expensive, costing more than $1000 per floor plan. But it’s possible to create cheap data by having a few floor plans and using those to generate new plans. This can also be used for engineering plans.
When a customer needs a new house or building, GAN software can generate a house design, floor plans, and construction plans based on descriptions given by that program. For example: “Northwest modern home with a large kitchen, 4 bedrooms and a pool on a 1-acre flat plot.” The program can create a house plan that matches all the specifications using this description. Of course, the plan would have to be altered by a professional to fit the specific land better – a 17-degree limestone hill has more engineering requirements than a flat acre Andisol.
There are also many cases in which AI engineering creates the most efficient patterns for structural components.
Generative adversarial networks are a fantastic and fascinating way of creating cheap data and understanding how computers create. Da Vinci would be proud of what is being produced today on a massive scale by non-organic creatives. Still, humans will continue to prefer the art made by other humans to appreciate the experience and story behind a masterpiece. Although software can make better art than most artists, the computer cannot convey the life they had that led to the creation we gaze upon today.
Nahla Davies is a software developer and tech writer. Before devoting her work full time to technical writing, she managed — among other intriguing things — to serve as a lead programmer at an Inc. 5,000 experiential branding organization whose clients include Samsung, Time Warner, Netflix, and Sony.