Not long ago, the state of voice recognition was quite primitive, and interacting with it was painful. Call management systems using voice recognition were fraught with errors, making it difficult to navigate to a desired destination or produce the correct result.
But that was then, and this is now. Voice recognition has come a long way. It’s now generally effortless and even enjoyable to ask things of Alexa, Siri or Google Assistant, and call systems work reasonably well. No one can deny the progress made in voice recognition.
With advances in voice recognition — and also image recognition — many are wondering if the same evolution can occur with enterprise security. Certainly, enterprise security needs advancement. Today, a data breach seems inevitable for most organizations with valuable assets, and few seem to be impervious. Many security professionals acknowledge that perimeter defense alone is insufficient, and the task is to find active attackers before they can steal or damage data or resources.
The generally touted answer for pinpointing an attacker is machine learning. Finding signs of an attack or attacker is a big data problem involving finding meaningful signs and an accurate understanding. Machine learning is seen as a way of finding significant data points and patterns and disregarding the superfluous. A visit to any major security exposition over the past two years confirms that machine learning is everyone’s buzzword and the supposed secret sauce to new security solutions.
Like voice and image recognition, machine learning can be greatly advanced to produce accurate, useful results. In particular, graph machine learning shows promise in being able to make sense of network data to find confirmation of abnormal and malicious activities.
Contrary to popular belief, advancing machine learning is not simply a matter of perfecting algorithms. Solving the problem is a classic “chicken or egg” conundrum, involving data and algorithms. Both are essential: Great algorithms but bad data will yield bad results. Great data but bad algorithms are equally flawed.
Even the data requirement is not so simple. The old computer adage of “garbage in, garbage out” applies, but not in one dimension. Getting the right data is a factor of many issues. First, there is data content and data format. Let’s say that an artificial intelligence (AI) challenge we want to solve is finding people you know who are not connected with you on Facebook. Data format would be people on Facebook, as opposed to organizations and other information. It is a bit like looking for data in the right place. Data content would be honing in on those people who you likely know.
Paring the data to this level makes the task of determining who you should friend much easier. Computation using algorithms then takes over to find the best matches. A medical analogy for data format is the time of day when a blood pressure measurement is taken, and content would be akin to how the measurement is made.[…]
Thank you for reading this post, don't forget to subscribe to our AI NAVIGATOR!