The Desire Engine: Machine Learning and 3D Printing

Machine learning will impossibly transform our world. I say “impossibly” because we can not foresee the implications and changes that artificial general intelligence will bring to bear on society. The workplace, the economy, and information will always be irrevocably different. These modifications will also be so unpredictable as to make prediction silly.

Will AI or machine learning be used to write all the software in the world? Or will it just raise the abstraction level of software so that everyone can now describe a website orally and this site is then made by software? Or will all software engineers simply not have to write code anymore, but instead become more like software architects? Or will the profession become extinct? Or maybe nothing will change but software engineers will simply become more productive. We can’t yet begin to understand the implications of the general availability of artificial general intelligence before the general availability of this intelligence actually occurs. However, we can begin to speculate about what progress in machine learning and AI can mean for 3D printing in some cases.

Example image of the 3D printer nozzle used by a machine learning algorithm to detect and correct errors in real time. Highlighted regions show aspects of the image that the system focuses on, providing potential insights into how the algorithm makes predictions. Credit: Douglas Brion

AI and Additive

A “high quality” 3D-printed bioscaffold as designed with help from a machine learning algorithm developed at Rice University. Scale bar equals 1 millimeter. Courtesy of the Mikos Research Group

As has been made clear through recent research, additive manufacturing (AM) will be deeply influenced by AI and machine learning. Companies like AI Build develop control software that optimizes toolpaths for 3D printers. Machine learning has been used to speed up a great deal of research in 3D printing and bioprinting, as well. The technology has been used to turn images into 3D printable models and to generate many designs. We will use it to perform quality assurance (QA)  and to divine optimal system settings. Machine learning, machine vision, and neural networks will eliminate a lot of noise and reduce a lot of tedium. Finding islands of optimization in a sea of noise will lead to better parts, faster print times, fewer errors, higher yield, and better designs. All the things in the world will become better eventually.

There is a beautiful marriage of 3D printing and machine learning, however, that I think will also have some implications that are not as straightforward

The Fountain of All Want

Seven years ago, I came up with the idea of the Fountain of All Want, which I also called the desire engine. The idea is that we all have desires, wants, needs, and requirements from products. Some of these are conscious needs powered by rational thought. Others are subconscious desires running through us that we hardly understand.

Impulse-driven purchases or long-chased dreams in thing form all have to be designed now. They then have to be made in large numbers many months in advance. Then, money has to be spent distributing these items and marketing them to get people to want them and to get people near these objects. This system is inefficient.

There is a million of everything and nothing is made specifically for you or me. A perfusion of delayed landfill clutters the earth and much of it has to be sold at knock down prices as the seasons move on. Seasons used to mean just the falling of leaves or reappearance in full of the sun. Now, it is the dumping of prices as stocks need to be cleared. Leaves do litter the ground, but so many tons of commercial hope is dug into landfills, littering the earth. Many more millions of tons will be discarded soon, as they have outlived their usefulness. We can come up with a more efficient, profitable, and sustainable system than this.

PrintFixer is a new machine learning driven software package that will make 3D printing 50 percent more accurate (Image: PXHERE)

What is a Desire Engine?

No more of this “reduce” stuff. No. Let’s consume more cheerfully. Let’s make better and have better things. It will look like this:

We manufacture goods out of recycled and recyclable materials. We set up recycling systems to optimally preserve part properties and use as little energy as possible. We incentivize consumers to return their used items to manufacturers. And we make a desire engine. We make a bag, a t-shirt, and a mug. We then define the boundaries of our design space. Then, we simply try each and every combination of design and new design. We test it by sharing them on social and putting them online. We offer better scores to things that are clicked, shared, and bought. In this way, we will learn continuously what is popular, what is working, what is needed, what is interesting and ultimately what sells.

Snakes on a Plane

However, we have to watch out for the Snakes on a Plane effect, wherein excitement and lots of chatter on the internet coalesces around a cultural artifact. This could make the item seem popular. However, in reality it is the absurdity, stupidity, jumped the shark-ness or irony present in a thing that drives attention. Sarcasm and cynicism travel widely on the web, but is often misinterpreted.

So, we will have to accurately delineate what is excitement stemming from utter ridiculousness and what is genuine interest in the value of an object. We will of course be able to also sell the plainly silly at times and people like the absurd. Generally, however, we are looking for machine learning to find good products. These are rendered more accurately all the time and will be produced on demand using 3D printing or other digital manufacturing technologies.

Additive and Machine Learning

The combination is a valuable one. 3D printing will only make what is needed once a customer wants it. AM will spread tens of thousands of designs inexpensively. We will make an item once a customer gives us money and that item may exist in a series of one of just a few items. On the whole, this will give us a new way to more accurately and more sustainably earn money without inventory and without trying to predict demand and preference. We will simply make what the customer orders when it is ordered.