Biomimicry: A new foundation for knowledge engineering and a striking new path in recombinant technology.
This article is not about the next big thing devoted to AI. It‘s more than naive evangelism. It’s a humble view of how we, and other species model our internal worlds. Imagine, what great opportunities may arise by bridging the natural and artificial world of data processing, without praying for the next upcoming technology.
Artificial data models, that fit with natural data processing.
Our brains are so damn fast because they don’t rely on structured data models, but rather on organically grown semantics, concepts, patterns, empathy, and phenomenology. AI is not about relationships. So maybe it’s time to break with a long history of tech-based approaches. It‘s time to develop an understanding for data models, that discover the meaning behind specialized, interconnected biological systems.
In 2006, I started investigating biology-inspired data models, while building one of the very first real native, most flexible, and fast XML databases in the market.
You may not remember the hype around XML at the beginning of the 2000s. But, in those days NoSQL databases were not yet at the forefront and XML looked like the only alternative to models based on rows and columns. I sensed that there is a significant chance to make things better, since 80% of the data we produce is unstructured, which is utterly human. And what is a huge problem for organizations, is that many of them still try to press unstructured data into structured models. Of course today we have way more crazy opportunities that come up with Machine Learning. But for a too long time now, many of us still take an extreme accounting perspective on data processing. And guess what, this is where semantic gaps even happen.
Take a look at this simplified representation of real-world data and how it still gets stored in many organizations.
Today’s organizations struggle to keep up with an exploding increase of heterogeneous data. Data that comes in many sizes and types and grows in volume and shape. So it’s evident that less structured databases can’t be the only answer to these challenges. We need solutions that emerge from new paradigms.
The beginning of something new: Graph databases.
Our perceived world is extensively interconnected. And graph data models mimic our whimsical and irrational real-world actions and relationships. They try to mirror how we think, which is fantastic and lays a firm ground to use with AI. But today we’re still at the beginning of graph technology. We haven’t yet exploited its full underlying power.
So if we solely rely on human data processing, we may get blinded by science. It’s way too complicated to reproduce our brains. We yet can’t grasp it in its entirety. Why do we dream, how do we dream, and what effects does it have on processing new information. There are a plenty of simpler biological systems. Ones that perform specific complex tasks deadly better than we do — due to natural specialization. And specialization could be the pivot for next-gen data models.
It was 2005; I was still studying at the State University of Arts and Design Karlsruhe, immersed in the theories of Heinz von Förster, Foucault, and Wittgenstein. While additionally working for the Market Intelligence Department at SAP Walldorf, I got a more in-depth look at how big companies manage large amounts of data through relational databases. Complex customer data records got meticulously pressed into rows and columns. Customers got treated as a bunch of numbers, and meaningful connections were partly re-established post hoc to ensure some context for the analytics teams. It only served a narrow framework of knowledge creation, not for scaling purposes.
I wondered, why we couldn’t use data models that may echo interhuman relationships, are scalable and still serve the business units. Back at my studio desk, I started to execute my first ideas.
I first pinned down the most important things I would need to reach my goals.
- Do not make use of a relational database.
- Use a flat file system.
- Use a language-oriented data model
- Make use of nodes, paths, attributes, and key-value pairs
- Rule out broken links between nodes
I saw that XML could probably meet all my requirements, but the use of XML at that time was focused on documentation of knowledge with strict schemas. So, I started to design a schemaless data structure.
My journey with many failures
In 2006 it was an unusual attempt to start directly by Design. The truth is that I was not willed to waste time by thinking and acting inside silos (first the model, then the logic, and then the view). It led to numerous, no, countless iterations in the first month, but the development speed kept on increasing week by week, followed by database crashes, blown up routines, incorrect data allocations, and finally at the tipping point, when I thought that I would win the game, a tremendous performance bottleneck. And so it seemed that the whole project would be condemned to fail.
Turning point and final approach
I was totally frustrated and saw no way out. I put all my eggs in one basket, but the database couldn’t scale to more than 200,000 data records, full-blown with nodes, attributes, labels, ids, etc. I forced the server, not at today’s speed, literally to its knees.
I was about to quit the project. Then I saw a TV documentation about the colonial organization of a particular type of jellyfish, the siphonophores, and since I, as a Sicilian, grew up partly under the sea, I sharply recognized, that I’ve found the missing piece of the jigsaw. Siphonophores are physiologically integrated zooids, structurally similar, but each with individual functions and structures. Here I saw the big chance to build an extendable, scalable schemaless database based on intimate, and contextual relationships.
Outcome and impact
- Handling of over 4,000,000 full-blown data records
- Faster parsing along smaller data buckets
- Absolute structural flexibility in each zooid
- No broken links due to unambiguous allocation
- Generation and its zooids can manage specific tasks
For me, it was inspiring at that time to bridge two apparently unrelated worlds, but it was more overwhelming to see what impact the act of bridging could have. What’s important to me is to give trends a chance, by trying to get to the very bottom of it. XML was a trend in 2006 like Blockchain is today. Blockchain as an example is not about finance, and it’s not even a financial product. It’s a technology that still hasn’t found its way, except being intended for trust building within larger communities.
The project was finally closed in 2010, but similar approaches are now found in modern graph databases, who can profit from bridging the natural and the artificial world of biological systems.