Meet DABUS: Artist, Inventor, Artificial Intelligence
The world’s first software program to be denied inventor credit on a patent has a winding past, and immense goals set before it. Stephen Thaler is hoping DABUS can make him immortal.
We have all contemplated at least once what it might mean to control someone’s mind. Steven Thaler, however, has been experimenting with the idea for the last 30-plus years.
His keystone project over that time has been in pursuit of an advanced AI that can reason and create autonomously. A polymath of the highest order, Thaler’s AI has trained hard to excel at stock picking, product design, and even creating art. To achieve these outcomes, Thaler relied on a discovery made early on in his work with neural networks: seeded with mental instability and pushed to the edge of death-inducing madness, these software applications produced some of their best creative association right before locking up and ‘dying.’
Mental Illness has served as a model for AI development for decades
As Thaler shifted focus from corporate work to independent consulting and his own company, he had a rich well of experience to draw from. In the mid ‘80s, he was already working with neural networks to solve complex problems, such as how to use lasers to create diamond from simpler carbon structures.
The first long arc of his solo career under the banner of Innovation Engines Incorporated featured an invention named the ‘Creativity Machine” as the company’s standard-bearer. His model for artificial intelligence, invented in the early 90s, relied on the software equivalent of near-death experiences and diagnosable personality disorders to achieve gains in ‘creative output.’
The analogy of these mental instabilities has become commonplace in the past decade especially, as big tech companies have adopted “adversarial generation” as one of the main architectures for creative output by AI. While previous iterations of Thaler’s AI have depended on the episodic cycling of manic and depressive methods, DABUS has evolved into higher levels of sophistication.
Thaler’s been building complicated machines for a long time. For him, it’s personal. It’s his path to immortality.
The inventor’s motivation for this work comes from a deeply personal place. Thaler has shared his vision of a future where machines progress enough to host man’s sentience forever- or, “as long as there’s electricity.” While there have been many applications for his expertise along the way, DABUS takes on a new importance when put in context — it marks a step forward for Stephen on his march towards never-ending life.
Imagination Engines version 2: How DABUS Differs
As his craft has matured, Thaler has played with increasingly complex interrelations of smaller artificial intelligence models. In 2008, he received a patent for DABUS, which is an evolutionary step beyond the generative adversarial network-based Creativity Machine.
Unlike many AI applications that have received news coverage in the past few years, DABUS does not specialize in one medium or domain. Adaptable to a variety of the humanities, math, science, physics, Thaler likens this AI to something like a human brain in operation. The main difference between DABUS and other AIs is that DABUS has progressed from a ‘manic’ and ‘depressive’ succession of processing across two neural nets to a model more closely patterned to the human brain.
What’s in DABUS’ head?
Where the original Creativity Machine was architected on the principle of successive manic and depressive cycles for idea generation, DABUS’ “mind” follows more organic patterns of thought. It is composed of many smaller neural nets, each optimized for processing linguistic, visual modeling, or even auditory impressions.
Perhaps the best analogy for understanding DABUS is that found in brewing beer. In the course of yeast working during fermentation, complex molecules such as sugars and essential oils break down and recombine into brand new compounds. DABUS uses a similar process; it harnesses the breakdown, synthesis, and recombination of perceptions and consequences to create memories.
Feedback Loops and Governance
These smaller AI models bump into one another, generate new associations on the subject under scrutiny, and continue those “trains of thought”. As associations build, a sort of “judge” presiding over all interactions determines which are stronger or weaker when compared to the problem statement being evaluated and provides feedback accordingly on those stronger or weaker ideas.
As feedback is generated, the network of nets reacts, continuing their work on promising idea chains, and abandoning work on less useful ones. The outcome is twofold: “memories” form and can even be shared across the network, and ever-promising ideas evolve. Thaler has argued that this difference from epochal thinking in other AI models means it takes much less training to get to useful outcomes, and the system overall becomes more self-sufficient.
In 2008, as DABUS matured to the point of patent-readiness, networking and clustering were a bottleneck to the speed at which the software could run. Because of the need for speedy pattern recognition at a meta-level across many interactions running on multiple computers, Thaler turned to computer vision for higher cognitive functions. Cameras take in the progress made on all computing nodes in use. Those images are processed and interpreted by the “judge” in DABUS, which can provide direction back out to the nodes.
Where DABUS has been / where DABUS is going
While focused on art (and particularly creating surrealist art) as one early use case for DABUS, Thaler has gone on to refine the AI model for many more applications. The AI has enjoyed a varied career, from ideating for the US Department of Defense to learning how to trade stocks, and, more recently, to physical object invention.
More recently, DABUS has focused on inventing. The AI was credited with invention for a drinking cup and an emergency beacon light in 2019 on two patent applications placed to offices in the United States, United Kingdom, and European Union. All have denied the initial filing (and an appeal) since. While DABUS is not yet accepted by these governing bodies as a full-fledged, legally protected inventor, Thaler is certain to press on in both building towards his vision of computing sentience and pressing for his work to be recognized.
With any luck for Thaler, that is not where the story ends. At the end of the journey he has charted, there is never-ending life. Recent events have shown the world what progress is being made on the path to machines that can think for themselves. Those same events have shown Thaler that there is more work to do. Not only does he have to keep improving his creativity machines; now, there is a fight for their rights to tackle.