Evolution has carefully preserved two distinct hemispheres in our brain – the right side pays wide open attention to the world, seeing the whole, whereas the left is adept at focusing on a single detail. The right hemisphere sees things in context and connected, recognizing the implicit meaning in vast amounts of information. In contrast, the left hemisphere isolates what it sees, focusing on what is unambiguous and logical.

Advances in computer technology and communications have given us machines with tremendous “left brain” capability, but even the most powerful systems struggle with the simplest right brain tasks. To solve the big problems facing our scientists, governments and financial institutions, a new line of machine intelligence is needed.

ai-one inc., based in San Diego and Zurich, has developed an adaptive holosemantic data space with semiotic capabilities (“biologically inspired intelligence”) that allows users to quickly analyze and discover meaningful patterns of interleaved text, time related data, and images. With the ai-one™ SDK, developers create intelligent applications that deliver better sense-making capabilities for semantic discovery, knowledge collaboration, sentiment analysis, image recognition, data mining and more. It also provides complex AI with reasoning and learning capabilities that get smarter with each use.

As has happened in the past, the breakthough comes not from the giant technology companies but from some brilliant, independent minds at a small company in Switzerland.

What is it and why does it matter?
The amount and types of information generated today are enormous and the access we enjoy due to the Internet are unprecedented. From an EMC/IDC report on the Digital Universe: “At nearly 500 billion gigabytes the Digital Universe, if converted to pages of text and assembled into books, would stretch to Pluto and back 10 times. At the current growth rate, that stack of books is growing 20 times faster than the fastest rocket ever made.

The Digital Universe is also messy… more than 95 percent of the data in the Digital Universe is unstructured, meaning its intrinsic meaning cannot be easily divined by simple computer programs…..”

This is creating an “information overload”. While for some the issue might be a nuisance, in the realm of research, knowledge management and global finance, “information overload” is a major and compounding problem for the experts and professionals we depend on to keep our physical and financial lives healthy and safe.

According to ai-one’s CEO Walt Diggelmann, conventional computing, i.e. “left brain” capability, cannot be forced to deliver “right brain” functions like pattern recognition and “sense making”. The behaviour of financial markets, terrorist organisations, and biological systems cannot be reduced to the elegant mathematical models of the physical world. ai-one’s new line of evolution can help “connect the dots”, whether we are trying to find a terrorist, build safer financial markets or understand the human genome.”

Understanding ai-one technology
One of the big visions of mankind is to build an equivalent to the human brain. At ai-one inc. the goal is to emulate the ability of the brain to think and to recreate that same ability in a computing system. By understanding the functions of the elements and pathways in the neocortex, technology could have the capability to process information and solve problems like the human brain.

Computers are generally based on the theories of Konrad Zuse and John von Neumann. These are formal data handling concepts. In order to build a biologically inspired machine with the same abilities as our brain, ai-one needed to replicate the biological structure and processes.

Cracking the neural code
The first key was to discover the genetic algorithm. This algorithm not only describes why and how the neurons fire, but also successfully interprets the actual meaning of the stimuli. The discovery of this algorithm and the creation of a data space within which the information could be stored lead to the development and introduction of ai-one, the name used to describe it’s technology.

ai-one consists of several genetic algorithms which enable the generic detection of recurrent patterns in all types of data, including but not limited to binary data, text, and specified formats. These patterns can be recursively nested in the input data and still found by the algorithms.  The recognition of these basic patterns is not based on a linear algorithm that tries to detect significant changes in the input data stream, instead the detection of these patterns is an inherent characteristic of the way the data is stored in ai-one.

ai-one uses a new kind of neural network, called a holosemantic data space (HSDS) where the incoming data is transferred into its generic representation as neural cells and synapses. By introducing the data into the HSDS, the inherent patterns are recognised by the interconnection of the resulting neural cells. The smallest amount of data processed by ai-one is called a data-quant. Depending on the application, a data-quant can be a binary digit, a pixel, a character in text, an atomic data value of a particular protocol/file format, etc.

The HSDS does not contain weights or thresholds within its neural cells, therefore the semantic context of the input data is established only by the interconnection over synapses of the cell cores. After data is introduced into the HSDS, the system produces no human viewable structure of the data, as it self-organizes the resulting cells and synapses. The communication with the HSDS is done by the stimulation of the neural cells (with a question or command), which then produces a human viewable structure that can be used for further processing of the data or used by an application.

ai-one employs a combination of formal and neural computing. This environment has specific features such as recognizing concepts, discovering intrinsic information structures and concept patterns with the concept of language, understanding Boolean logic.

As in the brain, time is an inherent dimension in the data, providing the connection to pragmatics. When something was learned by the system is a factor in understanding the semantic. As the meanings of concepts and structures evolve, ai-one learns from the new content but can also discern and track those changes over time.

Teaching, not programming
“ai-one, like a child, has to be fed information and taught what to do, not programmed”, says Manfred Hoffleisch, its inventor and Head of R&D. For ai-one to converse with a user, it has to learn basic knowledge and goals. In order to understand the semantic meaning and semantic association of words, where something is said and by whom is considered. The location where something is written defines the meaning and importance of the content. Once a given set of information has been loaded, ai-one learns from fulfilling specific tasks through interaction with experts and applications. The resulting knowledge base can be extracted and transferred to the next ai-one instance or a new generation of products.

Comparison with known ai approaches
Since there are no weights or thresholds for the calibration of the HSDS, it differs fundamentally from today’s known neural approaches. The main advantage is that the system doesn’t suffer under so called over-learning or under-learning conditions.

The quality of the result is no longer limited by the intelligence of the developers and the rules they create, but by the data itself. Prof. Ulrich Reimer, University of St. Gallen says “the HSDS uses a different paradigm based on context resulting in directed associations that provide better metrics, faster performance, and incremental learning from a smaller set of inputs”.
 
Product and value proposition
ai-one inc.’s (ai = autonomic intelligence) is used as an application programming interface (API) for the development of “best in class” products. It is incorporated into three API libraries, Ultra-Match™, Graphalizer™ and Topic-Mapper™. Ultra-Match is specialized for the field of imaging, Graphalizer for time related value chains and Topic-Mapper is specialized in the area of data and language (semantics). 

The libraries are delivered as a Software Development Kit (“SDK”) with training and custom development. Once an application embedded with ai-one is marketed, the Company receives licensing fees for the life of the product. Adoption of a powerful new technology must have a strong value proposition for the developer.
– Short learning curve & fast application development
– Binary data handling for speed (proprietary chip version for more speed)
– Architectural design flexibility using small (500KB) core program

Success with early adopters
After some dramatic results using ai-one on a laptop in a NIST competition, the Company was approached by the BKA (German counterpart to the FBI) to develop an application for crime scene shoe print matching and analysis. The outcome of this project was the commercial product, ASTIS™, which began selling in the EU and US forensic lab markets in 2010.

In a project with Swissport, ai-one was used to solve the problem of name verification and matching between airline passenger manifests and the no-fly list from the Department of Homeland Security. This application is being used by airports in Zurich, Frankfurt, and New York’s JFK.

Aggressive and innovative companies have worked with ai-one on a variety of new projects. Brainup, a semantic search tool for enterprise search engines, is hitting the market now. The CENDOO “butler” will launch later this year. A tool for searching, analysing, modeling and trading on financial patterns correlated with events monitored with text analytics is under development. Additionally, ai-one has partnered with universities on studies to assess the performance of ai-one technology.

The road ahead
The latest developments in the industry referred to as the “Semantic Wave” or Web 3.0, are a clear validation of this new direction. Microsoft’s Bing search engine, Googles acquisition of Metaweb, and Apple’s acquisition of SIRI at huge multiples underscore the value of effective semantic technology. With ai-one’s revolutionary technology, the best may be yet to come.

For more information contact Tom Marsh. Email: tm@ai-one.com; www.ai-one.com