From the course: Artificial Intelligence for Cybersecurity

Disciplines of artificial intelligence

From the course: Artificial Intelligence for Cybersecurity

Start my 1-month free trial

Disciplines of artificial intelligence

“

- [Instructor] Let's say someone comes up to you and says, "I built this machine and it's AI capable." How would you know if that claim is true? Fortunately, academics have proposed a framework to address this. For a machine to be considered generally artificial intelligent, it ought to have at least six foundational capabilities. In fact, these six capabilities are required to pass the well-known Turing test proposed by scientist Alan Turing. And this six capabilities or disciplines are: Ability to understand the natural language spoken or written by other humans. Ability to store and process information. Ability to reason. Ability to learn from new information. Ability to see and perceive objects in the environment. And lastly, ability to manipulate and move the physical objects. I'm sure like me, you are also surprised that these are too high expectations for a human made machines. Now, some advanced agents may possess all of these six capabilities, but in order to reap the benefits of AI for security, you actually don't need all of them in intelligent agent. For example, if you are building or deploying an AI-based network intrusion detection system, the system doesn't need the ability to see or physically manipulate objects, but it must be able to store, process and learn from enormous amounts of network logs. On the other hand, an AI powered security surveillance robot watching the entrance of a data center must have the ability to see and distinguish a person from a wandering deer. A chat bot such as OpenAI's chat bot, Google's Bard or Meta's LLaMA must be able to understand natural language and reason. That said, at minimum, artificial intelligence systems employed in the field of security possess some common capabilities, capabilities that make them intelligent. They store and process large volume of data. In the field of security, this data is usually in the form of logs from network devices, workstations, and API calls and so on. They identify patterns in the data. For example, they can determine that a user logs in every workday between eight and 9:00 AM from the Pacific time zone. They can learn from new information. For example, they can notice a change that the same user now connects via VPN from a home location every Friday. And lastly, they can make a recommendation or take action. For example, a cloud provider's identity and access management system calculates a sign-in risk score and based on the security policy, automatically blocks the login or enforces the two factor verification.

Contents