The Company

Cortical.io AG is an Austrian-based software company that has developed an innovative artificial intelligence (AI) technology based on a natural language understanding (NLU) approach to interpret and process human language text. The software can search, extract, annotate, and analyze key information from unstructured text and is language-agnostic. Current business uses include search and classification of information found in contracts, messages, and other forms of unstructured text.
Cortical.io is headquartered in Vienna, with additional offices in New York and San Francisco. Founded in 2011 by Francisco Webber (CEO) and Daniel Schreiber (CFO), Cortical.io has raised a total of $15.3 million in funding over six rounds. The latest was a debt financing round of $9.1 million from the European Investment Bank (EIB) announced March 3, 2021. Company officers say the funds will be used in part to expand R&D efforts to deliver intelligent document processing solutions.

The Technology

On the surface, Cortical.io sells cognitive capture software into two mainstream business applications: contract analysis and message processing. AI textual analysis methods have been in use in these markets for at least 15 years, but a deeper dive into Cortical.io reveals a novel technology platform that underpins these products and likely future applications. Rather than embracing traditional machine learning (ML) or even advanced deep learning techniques, Cortical.io takes a very different and patented approach. It leverages semantic folding and hierarchical temporal memory (HTM) techniques for natural language understanding (NLU), inspired by the legendary Jeff Hawkins’ groundbreaking research on neuroscience.

This is a very complex topic. In the simplest of terms, Cortical.io software implicitly mimics the way the human brain, specifically the neocortex, stores and creates memories. Think of it this way: Cortical.io mimics biological processes, as compared to more common AI deep learning methods that work on the basis of mathematical processes. In practical terms, the software is pre-trained on a corpus of data relevant to an industry or topic (insurance claims, for example). This training exercise builds a library of digital semantic fingerprints that denote relationships, context, and content. Incoming captured data is then matched against these fingerprints to ascertain relative semantic similarity.

Using such a method, Cortical.io claims, one can quickly train the system to recognize elements and actions within a document. Intriguingly, as the incoming data is essentially converted to a semantic fingerprint, it can run across any language and – in theory at least – meet any text-based cognitive capture need. This report can’t go much deeper into this novel approach, but if you are interested in how it differs from other methods we recommend reading more on the topic, as we have recently been doing.¹

Cortical.io architecture consists of sparse distributed representations (inspired by Numenta, the company founded by Jeff Hawkins) with Cortical.io’s own semantic folding technology (SFT) sitting atop to enable NLU. The bridging point is an API that converts data/words, etc. into digital fingerprints. Cortical.io software requires a relatively light computing footprint, which is the opposite of the massive processing and data resources required by deep learning, the current preferred approach. Nevertheless, Cortical.io has partnered with Xilinx to develop a Semantic Supercomputing platform.

As mentioned above, there are two products running atop this platform. The first, Contract Intelligence, is a product to extract insights – or more specifically, concepts – rather than keywords from large volumes of contracts or other unstructured documents. It has been designed for use by subject matter experts to automatically identify such things as clauses and provisions within the document, and it comes with a REST API to integrate with existing contract lifecycle management systems. The second is Message Intelligence, which has been designed to read through large volumes of incoming and outgoing emails and social media posts and apply pre-defined filters and classifiers. By applying NLU to these messages, the product can be used to prioritize and triage incoming opportunities or, conversely, to manage risk and compliance.

In summary, Cortical.io technology has the potential to manage text and language-based analysis with shorter training times and higher accuracy than previous approaches. Moreover, the Cortical.io approach can run unsupervised and needs relatively small amounts of data to learn.