Security Confidential S2 E0 Luis Martin

Manoj Tandon: Welcome to another episode of the Dark Rhino Security Podcast. We are joined again today by AI architect and designer Luis Martin to discuss the origins of artificial intelligence and its capabilities in the modern era. Welcome everybody. This is the second part of our series on artificial intelligence. We are once again grateful to have Luis Martin join us all the way from Madrid, Spain, for this second part in the series. We’re going to continue our conversation on artificial intelligence and various aspects of it in terms all of us can understand. The goal of this initial series of podcasts with Luis is to give everybody an understanding of artificial intelligence at a level where we can all become conversant with it and talk relatively well about it. With that, Luis, welcome. How are you today?

Luis Martin: I’m fine. It’s very good weather in Spain, and I’m very happy to talk again about the main principles of artificial intelligence with you and share this with our audience. It’s very important that potential customers clearly understand the theoretical and fundamental principles of AI. To be consistent with history, the term AI was introduced from a proposal by John McCarthy to the Rockefeller Foundation called A Study on Computational Models of Simulation of All Aspects of Learning and Other Characteristics of Human Intelligence. McCarthy used the term artificial intelligence to differentiate it from his colleagues in cybernetics, who focused more on communication and control between man and machine. The fundamental scientific element was the use of symbolic logic and inductive reasoning.

Manoj Tandon: Symbolic logic was prevalent in the early days of AI. What changed there?

Luis Martin: Many things changed. During the 70s and 80s, symbolic logic was the fundamental element of AI. From the 80s onward, the neural approach appeared strongly and involved radically different computational paradigms. Currently, symbolic and neural computation are mixed into what we call neurocognitive computing.

Manoj Tandon: Has symbolic computing in AI been superseded by neural networks?

Luis Martin: In a way, but it is still very much alive, especially in intelligent planning and problem-solving strategies that require formulating a series of steps to achieve a specific objective. Examples include assisted driving, logistics planning, scientific hypothesis testing, cooking recipes, automatic manufacturing, and air traffic control. In cybersecurity specifically, heuristic reasoning is extremely efficient for simulating warning and response scenarios against threats and attacks, risk analysis, and planning cyber strategies. This current symbolic approach is often referred to technologically as GOFAI: Good Old-Fashioned AI.

Manoj Tandon: Good Old-Fashioned AI—that’s a real acronym. Listening to you describe this, it seems like AI is really just sets of algorithms. Is that what sets it apart?

Luis Martin: In the first stage of artificial intelligence, John McCarthy raised the problem that the training system for learning is inseparable from the problem of how to represent knowledge and how to transform that representation when errors occur so it becomes more adequate for the task at hand. This involves the generation of new truths from old ones and directly relates to intelligent development of representation in the mind. It includes abilities like creativity, learning, and problem-solving. From the start, it was clear that knowledge representation and processing are the fundamental principles on which AI is established. Apart from that, they are algorithms of varying complexity applied to different problems.

Manoj Tandon: This seems like a very complex problem set to solve. The representation of knowledge, processing, learning models, and mimicry of human intelligence are definitely non-trivial. We’ve also heard terms like “intelligent systems.” Can you differentiate artificial intelligence from intelligent systems?

Luis Martin: Artificial intelligence is a discipline of computer science, while intelligent systems are explicit systemic models or AI technologies in the form of complete or incomplete systems with cognitive, rational, neural, or biomimetic approaches. These intelligent systems will form the basis of the technological revolution and the transformation of organizations that will support 21st-century society.

Manoj Tandon: So artificial intelligence is the overarching discipline, whereas intelligent systems are explicit system models?

Luis Martin: Correct. Intelligent systems are the best design strategy for implementing AI technologies.

Manoj Tandon: Can you give us insight into the capabilities of these systems?

Luis Martin: Some capabilities include capturing and integrating information from multiple sources and formats while evaluating reliability and accuracy; detecting external stimuli and generating concepts and meanings; situational awareness and prediction; processing natural language and understanding meaning; inferring new information through analytical and synthetic processes; learning from experience; modifying objectives according to environmental context; maintaining associative memory; cooperating with other entities to solve complex objectives; and triggering actions to achieve goals.

Manoj Tandon: To our listeners, those descriptions sound almost human. But these are still models, not actual human intelligence.

Luis Martin: Exactly.

Manoj Tandon: Otherwise people start thinking we’re building Skynet from Terminator.

Luis Martin: I agree completely.

Manoj Tandon: In what areas of AI research and technological design are you most involved right now?

Luis Martin: At Dark Rhino Security we have a very extensive research and development program. The core technologies I’m most involved with include multi-source information evaluation, sensory detection technologies, situational awareness in real time, evidence-based analysis technologies, automatic cooperative analysis of complex situations, smart big data analysis, strategic information production, strategic planning technologies, machine learning, technologies for adversary reasoning analysis, dual knowledge technologies for intelligent cybersecurity agents, and natural language processing technologies. Some of these lines of research are extremely critical for us.

Manoj Tandon: We hear “machine learning” and “deep learning” constantly, but not terminology like “automatic cooperative analysis of complex situations.” Have systems already been built around that concept?

Luis Martin: Yes. The theoretical approach I’m using to apply this to cybersecurity and national corporate security problems is based on the idea that the best way to resolve complex problems is through cooperative approaches.

Manoj Tandon: That almost sounds like an artificial neural network. Could you explain the difference between artificial neural networks and deep learning?

Luis Martin: Learning involves two processes. The first is assimilating information from experience into memory by storing and relating it to existing information. The second is extracting knowledge patterns that transform cognitive abilities such as classification, recognition, and reasoning. Artificial neural networks are structured in nodes simulating neurons and synaptic connections organized in layers. The lower layer receives external information, such as image pixels, and the system trains through repeated samples. The weights propagate through the hierarchy until the network can precisely recognize and classify images. The interesting part is that the system can recognize dogs or buildings without explicit programming. This is called unsupervised learning. Neural networks detect complex relationships between patterns and dynamically classify them.

Manoj Tandon: What you’re describing fundamentally sounds like linear algebra with weighted connections between neurons. A certain pattern fires and determines whether something is a dog or not.

Luis Martin: Exactly. It is very important to design the neural network structure correctly in order to reduce training time and minimize the amount of required data.

Manoj Tandon: That brings us to big data. It seems like a tremendous amount of time in deep learning systems is spent creating the training data set. Is the quality of the data the weakest link?

Luis Martin: Not only the data itself, but also the network structure. Organizations often don’t have the specific data they actually need. In my opinion, the first step is identifying the real data available within the organization. Then the network structure and training methods must be designed according to that reality.

Manoj Tandon: Do you think machine learning is similar to how humans learn?

Luis Martin: Honestly, we still don’t understand how the human brain truly works. But current AI systems are already demonstrating greater capabilities than humans in areas such as detection, recognition, and classification. Combined with computational capacity, this allows us to glimpse a revolution in creating new forms of non-human intelligence that are artificial in a strict sense.

Manoj Tandon: That distinction between human and machine intelligence is important because these systems are excellent at narrow tasks where sufficient data exists, but they don’t possess consciousness or morality.

Luis Martin: I completely agree. Human intelligence is too complex to compare directly with artificial intelligence. The most powerful AI today is not comparable even to an insect in terms of general intelligence. AI has been designed to enhance our cognitive abilities in areas such as pattern recognition, planning, analysis, and decision-making.

Manoj Tandon: Since we’re a cybersecurity company, how can deep learning and automatic reasoning be applied to cybersecurity?

Luis Martin: The applications are immense because cybersecurity environments generate millions of data points in real time. AI can detect and classify situations instantly and launch automatic responses against threats and attacks. Some examples include automatically analyzing vulnerabilities, detecting new cyber weapons from the earliest stages, identifying stealth attacks and information poisoning, reducing cognitive overload for SOC analysts, implementing intelligent virtual operators capable of automatic reasoning and decision-making, and simulating attack scenarios to test defensive or offensive responses.

Manoj Tandon: You mentioned detecting sophisticated cyberattacks involving contaminated information and erroneous decision-making. In the United States, election tampering and fake news dominate headlines. Could AI-based cybersecurity systems detect or stop that type of misinformation?

Luis Martin: It is very important to detect what is currently called fake news. The objectives in social media deception campaigns are different from cybersecurity attacks. In social media, the goal is broad propagation. In cybersecurity, the goal is stealth and silence. Successful deception attacks integrate poisoned information into organizational systems without detection. This is a very important issue for organizations because many attacks are designed so the organization never realizes the true objective.

Manoj Tandon: What you don’t know, you can’t defend against. That sounds like a great topic for another podcast. Luis, we are at the end of our time here, but thank you once again for sharing your insights as we continue this educational series on artificial intelligence.

Luis Martin: Thank you. It is very important for me as an architect and designer at Dark Rhino Security to explain the main lines of research and experimental design that we are pursuing within our organization.

Manoj Tandon: Thank you, Luis. And thank you to everyone for joining us. Be sure to hit that subscribe button and visit us at darkrhinosecurity.com.