AI is Not Magic.
The current super buzzword is AI, or Artificial Intelligence. Many people imagine AI as a program that behaves like a human, but in reality, AI has been around for nearly 70 years, encompassing a wide range of algorithms and processes. The essence of AI is any program that can solve an abstract problem. Historically, the AI space has faced a peculiar challenge: whenever an algorithm within this domain demonstrates utility, it tends to be reclassified as an algorithm outside of the AI space. This has happened with technologies like expert systems and fuzzy logic.
The current algorithms popularly branded as AI are more accurately known as large language models. These models use various forms of machine learning to generate text and images based on initial input. To laypeople, this process may appear magical, and while it is indeed impressive, it is fundamentally based on statistics. These models lack true understanding or reasoning, which often leads to flawed and illogical outputs.
At Auspex Labs Inc., we have explored but not embraced 'AI' (Large Language Models) for our products, as they are simply too unreliable at this time. Instead, we are utilizing unsupervised machine learning and actively building supervised machine learning capabilities into our products. These algorithms can handle abstract inputs and provide much narrower, more determinative results that meet an acceptable margin for human use.
The machine learning, 'ML', algorithms that we are using are divided into two types: Supervised Machine Learning and Unsupervised Machine Learning.
Unsupervised Machine Learning algorithms are straightforward. These algorithms organize data into groups based on the characteristics of the data. There are several algorithms in this space that can sort data into a predefined number of groups or allow the algorithm to dynamically determine the number of groups to create. We use these to assign more subjective properties to network devices.
Supervised Machine Learning algorithms are more complicated. In our case, they provide classification to answer the question, "Is this network traffic normal or anomalous?" The challenge with Supervised Machine Learning is that it requires data that we can label as normal or abnormal to train a model to perform the classification. We have put significant effort into obtaining and labeling data for this effort, which is ongoing and crucial for improving our models.
These AI/ML algorithms are sufficiently predictable in their outcomes that we can trust their output for risk detection. By leveraging these algorithms, we are able to identify potential threats and anomalies in network traffic with a high degree of accuracy. This allows us to proactively address security issues and ensure the integrity of our clients' network infrastructures. Our commitment to refining these models and continually improving our data labeling processes underscores our dedication to providing reliable and effective machine learning solutions for our customers.
Community Deviation™ is one of Auspex Labs Inc.’s proprietary Unsupervised Machine Learning algorithms.
