In many ways, it feels like artificial intelligence (AI) has burst from the world of science fiction into almost every aspect of our lives. Yet for many years, people have discussed Alan Turing’s Test for machine intelligence, or Imitation Game, where a human asking questions via a text interface would try to identify from two sets of answers which was the human and which the machine.
Achieving this task has been much of what has driven development in the world of machine, or artificial, intelligence, and this test is largely responsible for creating the likes of ChatGPT and other large language model (LLM) approaches.
Our desire as humans to develop artificial intelligences that reflect our human characteristics can be traced back thousands of years. Homer tells of the Greek god Hephaestus, who built female servants of gold who had minds, hearts, intelligence, vocal chords and strength, not to mention Talos, a mechanical being capable of circling Crete three times a day (running at about 250 miles per hour if you were wondering) throwing rocks at potential intruders.
However, it can also be argued that the desire to reflect human characteristics in our pursuit of machine intelligence has been a distraction and wasted significant effort that could have been applied elsewhere. At the very least, it has led to a focus on seeing those human characteristics as somehow more valuable and important in AI solutions than the many other areas we use.
Naturally (no pun intended), information technology is closely linked to the development of AI, and cybersecurity is an area that has willingly embraced its potential. When considering AI within a solution, the challenge is understanding what that means in real terms. Consider for a moment why the likes of ChatGPT or Midjourney have grasped the public’s imagination versus the developments made with playing Chess or Go that seem much more mundane and the outcomes much less applicable.
It’s important to remember though, that far less exciting methods of machine learning can be more effective and bring far greater value. If you are of an older generation, you will remember the amazement of using GPS for the first time. In general, the focus was on the satellites sat in geosynchronous orbit above the earth and how they could tell us exactly where we were. You may not have known that the behaviors of ant colonies, their natural intelligence if you will, was one of the methods used in deriving the algorithms that calculate the most efficient route.
AI in Security Solutions
It’s the same in cybersecurity. Vendors are rushing to splash AI across their marketing without any real consideration for whether that’s a good or bad thing. If you want to see a great example of how poorly AI can interpret everyday life, search online for ‘Synthetic Summer,’ an advert created using AI by Helen Power and Chris Boyle from privateisland.tv. Based on the horrors in that video, it’s fascinating to imagine what decisions AI could be making with network and endpoint data.
Of course, this is the point. AI is an extremely generic term and can apply to a whole range of different technological approaches. Simple algorithms can be dramatically more effective than an entire cloud-based LLM when applied to the task it was designed for. When considering updating or replacing your security solutions, ensure that you are making decision-based outcomes, not marketing-led ones.
It may sound great that a solution includes AI, but understanding what that means is important. It’s also important that the vendor can explain and clarify in simple terms how they use AI and how it will improve your security posture. Simply casting AI as a magic solution, coated in buzzwords like neural networks, machine learning or LLMs, is all well and good, but outcomes need to be specific, meaningful and measurable. Whilst you don’t need to understand the various AI techniques in depth, understanding the role and its outcomes is essential.