Making sense of AI: the explainer

The common problem

If like me, your head is spinning with AI-related acronyms and definitions, this chart will help you put the puzzle together.

Did I find it online? The web is not short of AI-related charts, but nothing can be found of this level of abstraction.

Finding the solution

The chart above was painstakingly assembled by sifting through the results of dozens of web searches, and ChatGPT and Bard prompts.

Why dozens? Because so many of the responses were ambiguous or contradictory. Some may argue that this chart is incorrect, but few people will have expended as much effort in realising that there may be no correct answer!

Listed below, are short descriptions of each term that should be comprehensible to anyone unburdened with IT knowledge. My focus is on generative AI, the red and pink parts of the chart. They grey items are there for context.

Artificial intelligence (AI)

Artificial intelligence describes the ability of a machine to simulate human intelligence. We are familiar today with Artificial Narrow Intelligence (ANI), also known as weak AI, which is designed to perform specific tasks, such as playing chess or translating languages. ANI is typically very good at these tasks, but it is not able to generalize to other tasks.

Artificial General Intelligence (AGI), also known as strong AI, is a hypothetical type of AI that does not yet exist, but it is a major goal of AI research by companies like OpenAI . AGI would have the ability to understand and reason like a human being.

Generative AI

Here is a list of simple definitions of each of the terms:

• Generative AI:?A type of artificial intelligence (AI) that can create new content, such as text, code, images, and music.
• Large Language Model (LLM):?A type of AI model that is trained on a massive dataset of text and code to learn patterns and relationships. LLMs?are designed for understanding and generating language.?LLMs can be used to generate text, translate languages, write different kinds of creative content, and answer your questions in an informative way.
• PaLM: A large language model developed by Google AI that is capable of generating different creative text formats, like poems, code, scripts, musical pieces and email.
• Bard: Developed by Google.?It can answer questions in a human-like way by accessing up-to-date information from the internet.?It’s based initially on the LaMDA family of large language models and later PaLM. Bard references and accesses the internet in real time, whereas ChatGPT does not.
• Generative Pre-trained Transformer (GPT):?A type of LLM that uses a transformer architecture. Transformers are a type of neural network architecture that is well-suited for natural language processing tasks.
• ChatGPT:?A chatbot application that is built on top of GPT-3 and GPT-4 LLMs. Unlike Bard, which generates responses using data from the web, it is trained to generate human-like text in response to a wide range of prompts and questions, enabling users to refine and steer a conversation towards a desired length, format, style, level of detail, and language.
• Diffusion Model:?A type of AI model that can be used to generate realistic images and videos.?
• DALL-E 2 and Imagen are diffusion models that can generate realistic images and videos from text descriptions.
• AlphaFold 2:?A deep learning model that can predict the 3D structure of proteins from their amino acid sequence.

Machine Learning

Machine Learning is a subset of AI that focuses on developing algorithms that can learn from data and improve their performance over time.?

• Supervised learning is the most common type of machine learning, and it is used to train machines to predict outputs based on known inputs. For example, a supervised learning algorithm could be used to train a machine to predict whether a customer will cancel their subscription, based on their behaviour.
• Unsupervised learning is used to train machines to find patterns in data without being given any labeled inputs. For example, an unsupervised learning algorithm could be used to cluster customers into different groups based on their purchase history.
• Reinforcement learning is used to train machines to learn how to behave in an environment by trial and error. For example, a reinforcement learning algorithm could be used to train a self-driving car to learn how to navigate the road.

Other types of AI

Apart from generative AI and machine learning, there are several other types of AI that have emerged in recent years. These include:

• Expert Systems: These AI systems are designed to solve specific problems in a domain where experts have a deep understanding. They use a knowledge base and inference rules to make decisions and solve problems. Examples of expert systems include medical diagnosis systems, financial planning tools, and fraud detection systems.
• Robotic Process Automation (RPA): RPA software mimics human actions to automate repetitive tasks, such as data entry, processing invoices, and customer service interactions. This type of AI is particularly useful in industries with high volumes of repetitive tasks, such as banking, insurance, and healthcare.
• Fuzzy Logic: a system of reasoning that deals with imprecise or incomplete data. It is based on the idea that there can be degrees of truth between 0 and 1, unlike traditional Boolean logic which only has true or false values. Fuzzy logic is used in a variety of applications, such as controlling appliances, making decisions in robotics, and developing expert systems.
• Evolutionary Computation: a type of AI that is inspired by natural selection. It uses algorithms that mimic the process of evolution to solve complex problems. This type of AI is used in a variety of applications, such as optimizing designs, scheduling tasks, and training neural networks.

Cyber security considerations for AI

AI security matters to AI system vendors, buyers and users in different ways. Listed below are the most important considerations.

Vendors

1. Data security and privacy.
2. Model security, explainability, and susceptibility to bias and discrimination.
3. Infrastructure security and resilience.
4. Supply chain security.
5. Governance and regulatory compliance (eg EU AI Act).

Buyers

1. Vendor security posture and reputation.
2. Data security and privacy practices through the AI lifecycle.
3. Model security and explainability.
4. Infrastructure security and deployment.
5. Continuous monitoring of cyber threat exposures and risk management.

Users

1. Familiarise yourself with the security of the ai system. Understand how the vendor addresses data security, model robustness, and infrastructure protection.
2. Employ strong authentication methods, such as multi-factor authentication (MFA), to access the AI system.
3. Watch out for anomalies and report any suspicious activity to the vendor or it security.
4. Keep AI software up-to-date as you should any software.
5. Report security vulnerabilities responsibly.

Whether you are a vendor, buyer or user, as AI continues to evolve, it is important to carefully consider these risks and develop strategies to mitigate them. This is the focus of research at WithSecure and many other companies.