What is AI? All you need to know about artificial intelligence

What is artificial intelligence?

Artificial intelligence is “a technical and scientific field devoted to the engineered system that generates outputs such as content, forecasts, recommendations or decisions for a given set of human-defined objectives” [ISO/IEC 22989:2022]. While this definition of artificial intelligence is accurate from the technical perspective, how does it translate for the average person?

In truth, AI is just a practical tool, not a panacea. It’s only as good as the algorithms and machine learning techniques that guide its actions. AI can get really good at performing a specific task, but it takes tonnes of data and repetition. It simply learns to analyse large amounts of data, recognize patterns, and make predictions or decisions based on that data, continuously improving its performance over time.

The machine learning has become so “competent” as to generate everything from software code to images, articles, videos and music. This is the next level of AI, the so-called generative AI, which differs from traditional AI in its capabilities and application. While traditional AI systems are primarily used to analyse data and make predictions, generative AI goes a step further by creating new data similar to its training data.

A very brief history of artificial intelligence

Today’s AI loosely stems from the 19th-century invention of Charles Babbage’s “difference engine” – the world’s first successful automatic calculator. British code-breaker Alan Turing, who was a key figure in the Allies’ intelligence arsenal during WWII, amongst other feats, can also be seen as a father figure of today’s iterations of AI. In 1950, he proposed the Turing Test, designed to assess a machine’s ability to exhibit intelligent behaviour indistinguishable from that of a human.

The exponential growth of computing power and the Internet brought with it the concept – and the reality – of machine learning, the development of AI algorithms that can learn without being programmed, by processing large datasets. Over the past decade, AI has become integral to everyday life, influencing how we work, communicate and interact with technology.

How does AI work?

In essence, AI analyses data to extract patterns and make predictions. It does this by combining large datasets with intelligent AI algorithms – or sets of rules – that allow the software to learn from patterns in the data. The way the system accomplishes this is through a neural network – an array of interconnected nodes that relay information between various layers to find connections and derive meaning from data.

To grasp how this works, we must unpack the following concepts:

• Learning: AI’s machine learning feature enables machines to learn from data, identify patterns and make decisions without explicit programming. Going one step further, advancements in deep learning empower AI software to understand more complex patterns using millions of data points.
• Reasoning: The ability to reason is crucial to AI because it allows computers to mimic the human brain. AI can make inferences based on commands it is given, or other available information, to form hypotheses or develop strategies for addressing a problem.
• Problem solving: AI’s problem-solving capability is based on the manipulation of data through trial-and-error techniques. It involves using algorithms to explore various possible paths to find the most optimal solution to complex problems.
• Processing language: AI uses natural language processing – or NLP – to analyse human language data in a way that is meaningful to computers. What is NLP? It refers to the ability of computers to understand, interpret and generate human language, using text analysis, sentiment analysis and machine translation.
• Perception: AI scans the environment through sense captors such as temperature sensors and cameras. Known as computer vision, this field of AI enables machines to interpret and understand visual data and is used in image recognition, facial recognition and object detection.

Strong AI vs weak AI

Artificial Intelligence (AI) encompasses a diverse spectrum of capabilities, which can be broadly classified into two categories: weak AI and strong AI.

Weak AI, often referred to as narrow AI, embodies systems meticulously crafted to excel at specific tasks within well-defined parameters. These systems operate within a confined scope of expertise and lack the capacity for general intelligence. Think of them as specialists trained to perform particular functions efficiently.

Here are some examples of narrow AI applications, characterized by their specialized algorithms designed for specific tasks:

• Meta’s (formerly Facebook) Newsfeed: Utilizes narrow AI through algorithms that analyze your past interactions and similarities with other users to personalize content, showing how AI is tailored for the specific task of content curation based on user interests.
• Amazon’s Suggested Purchases: Employs task-specific AI to recommend products by analyzing your browsing and purchase history, alongside data from similar users, demonstrating AI's ability to make predictions within the confined context of shopping preferences.
• Apple’s Siri: A voice-activated virtual assistant powered by narrow AI, Siri uses natural language processing to understand and respond to specific spoken requests, illustrating AI's application in performing dedicated tasks such as setting reminders or sending messages.
• Recommendation Systems (e.g., Netflix): These systems leverage narrow AI to suggest movies or shows by comparing your viewing history with patterns from millions of users, showcasing AI's role in providing personalized entertainment recommendations.
• Chatbots for Customer Service: Powered by narrow AI, chatbots handle specific customer inquiries through pre-defined patterns and data-driven responses, reflecting how AI can be specialized to improve customer service efficiency.
• Virtual Personal Assistants (e.g., Amazon Alexa, Apple Siri): These assistants use narrow AI for a range of specific tasks like setting alarms, answering questions, and controlling smart home devices, based on voice commands and natural language understanding.

Each of these applications demonstrates the strength of narrow AI in executing well-defined tasks by analyzing large datasets and following specialized algorithms, without possessing the broader, adaptable intelligence akin to human cognition.

In contrast, the concept of strong AI, also known as general AI, aspires to develop systems capable of tackling a wide array of tasks with a level of proficiency that satisfies human standards. Unlike their narrow AI counterparts, strong AI systems aim to possess a form of general intelligence, allowing them to adapt, learn, and apply knowledge across various domains. Essentially, the goal is to create artificial entities endowed with cognitive abilities akin to those of humans, capable of engaging in intellectual endeavors spanning diverse fields.

And since you asked, here are some examples of strong AI:

• Human-Level Chess Programs: Programs like Deep Blue and AlphaZero can play chess at a grandmaster level, analyzing complex positions and making strategic decisions.
• Natural Language Understanding and Generation: Systems like OpenAI’s GPT-4 can understand and generate human-like text, write essays, poems, and even engage in conversations.
• Self-Driving Cars: Autonomous vehicles use strong AI to perceive their environment, make real-time decisions, and navigate safely without human intervention.
• Robotics with General Capabilities: Robots equipped with strong AI can perform tasks beyond repetitive actions. For example, Boston Dynamics’ Spot robot can navigate complex environments, open doors, and carry objects.
• Creativity and Art: Some AI models can compose music, create visual art, and write original stories or poems. For instance, AIVA generates classical music compositions.
• Turing Test Passers: These AI systems can convincingly simulate human conversation. Passing the Turing test involves indistinguishable interactions between an AI and a human evaluator.

While researchers and developers continuously strive to push the boundaries of AI capabilities, achieving true general intelligence comparable to human cognition poses immense challenges and remains an elusive goal on the horizon.

Having said that, seeing the significant advancements in AI technology and machine learning, the question is not if but when. What do you think? Will we be surrounded by strong AI in our daily lives in 5, 10 or 20 years?

What are the four types of AI?

Artificial Intelligence (AI) encompasses a diverse array of capabilities, each serving distinct functions and purposes. Understanding the four types of AI sheds light on the evolving landscape of machine intelligence:

Reactive Machines:

These AI systems operate within predefined rules and lack the capacity to learn from new data or experiences. They excel in executing specific tasks efficiently but remain static in their abilities. For instance, chatbots used to interact with online customers often rely on reactive machine intelligence to generate responses based on programmed algorithms. While they perform well within their designated functions, they cannot adapt or evolve beyond their initial programming.

Use Cases: These machines excel in specific tasks but lack adaptability. For instance, chatbots that provide scripted responses to online customers fall into this category.

Example: Imagine a chess-playing program that evaluates each move based on predetermined rules without learning from past games.

Limited Memory:

Unlike reactive machines, AI systems with limited memory possess the ability to learn from historical data and past experiences. By analyzing and processing information from previous interactions, these systems can make informed decisions and adapt to some extent based on their training. Examples include autonomous vehicles equipped with sensors and machine learning algorithms that enable them to navigate through dynamic environments safely. Similarly, natural language processing applications utilize historical data to improve language understanding and interpretation over time.

Use Cases: Applications include autonomous vehicles, natural language processing, and recommendation systems.

Example: Self-driving cars analyze real-time sensor data and past driving experiences to navigate safely and make decisions on the road.

Theory of Mind (An Area of Ongoing Research)

This realm of AI represents a complex area of ongoing research, delving into the concept of artificial agents capable of understanding the beliefs, intentions, and emotions of other entities. While humans intuitively infer mental states to navigate social interactions, replicating this cognitive ability in machines presents significant challenges. Developing AI with a theory of mind could revolutionize various fields, including human-computer interaction and social robotics, by enabling more empathetic and intuitive machine behavior.

Challenges: Developing AI with theory of mind is complex and remains an active research area. It requires modeling human-like empathy and understanding.

Potential Applications: If achieved, such AI could enhance human-computer interactions, empathy in virtual assistants, and social understanding.

Self-aware AI (Hypothetical and Speculative)

The notion of self-aware artificial intelligence explores the hypothetical scenario of AI systems possessing consciousness and the ability to comprehend their own existence.

Have you watched the movie Blade Runner? Well, that's exactly it.

While often portrayed in science fiction and philosophical discourse, achieving genuine self-awareness in AI remains speculative and far from practical realization. The concept raises profound questions about the nature of consciousness, identity, and ethical considerations surrounding AI development. Despite its speculative nature, the exploration of self-aware AI prompts reflection on the evolving relationship between machines and human cognition.

In summary, the four types of AI illustrate the diverse spectrum of intelligence exhibited by artificial systems, ranging from reactive rule-based machines to speculative notions of self-aware consciousness. As AI continues to advance, exploring the capabilities and limitations of each type contributes to our understanding of machine intelligence and its impact on society.

Machine learning vs deep learning

Machine learning relies on algorithms which automatically improve performance through experience. These have three categories:

• Supervised learning: The algorithm is trained on a labelled dataset where each example has an input and a corresponding output, learning from this labelled data to make predictions on new, unseen data.
• Unsupervised learning: Without any predefined labels or outputs, the algorithm learns to discover hidden structures or groupings within the data.
• Reinforcement learning: Trained to interact with an environment and learn through trial and error, the agent receives feedback in the form of rewards or penalties as it performs actions allowing it to learn and improve performance over time.

Deep learning is a subset of machine learning, focused on training artificial neural networks with multiple layers – inspired by the structure and function of the human brain – consisting of interconnected nodes (neurons) that transmit signals.

By automatically extracting features from raw data through multiple layers of abstraction, these AI algorithms excel at image and speech recognition, natural language processing and many other fields. Deep learning can handle large-scale datasets with high-dimensional inputs, but requires a significant amount of computational power and extensive training due to their complexity.

Examples of artificial intelligence

So what can AI do? Most people are familiar with it through smart speakers and smartphone assistants like Siri and Alexa, but new AI technology constantly makes our lives easier and more efficient in many other ways.

Here are some examples of artificial intelligence applications:

• Healthcare AI can process and analyse vast amounts of patient data to provide accurate predictions and recommend personalized treatment for better outcomes.
• Business and manufacturing benefits from automation in every field, from fraud detection, risk assessment and market trends analysis to AI-powered robots on production lines. AI systems can also predict equipment failures before they occur and detect anomalies in network traffic patterns, identifying cybersecurity threats. And in retail, AI offers inventory management, personalized shopping experiences, chatbots to assist customers and analysis of customer preferences, increasing sales through better targeted adverts.
• Education AI includes intelligent tutoring systems which adapt to students’ needs, providing tailored feedback and guidance. AI also offers automated grading, content creation and virtual-reality simulations.
• Transportation AI optimizes traffic flow, predicts maintenance needs, and improves logistics in shipping companies, while in agriculture it can optimize crop yield and reduce resource wastage. Drone technology monitors soil conditions, identifies crop diseases and assesses irrigation requirements, and AI systems can recommend efficient pesticide usage and crop management.
• Entertainment: By analysing user preferences, AI can recommend movies, music or books. Virtual and augmented reality create immersive entertainment environments. Realistic CGI and “special effects” AI enhances the visual experience of movies and games.

AI governance and regulations

With increasing integration across various industries, the importance of ensuring the quality and reliability of AI software cannot be overstated. Despite the risks involved, AI still suffers from a lack of regulation. This is where International Standards can help.

Standards, such as those developed by ISO/IEC JTC 1/SC 42 on artificial intelligence, play a pivotal role in addressing the ethical and responsible development and use of AI technologies. They help to bridge the gaps in regulation, giving decision makers and policymakers the tools to establish consistent and auditable data and processes.

• ISO/IEC 42001 AI management systems

  ISO/IEC 23894 AI – Guidance on risk management

  ISO/IEC 23053 Framework for AI Systems Using ML

These standards can bring long-term value to a business, particularly in areas such as environmental reporting. Standards build credibility with stakeholders, ensuring the benefits of artificial intelligence outweigh the associated risks through aligning with existing regulations and governance tools.

How will AI change our world?

As it becomes more sophisticated, we can expect to see artificial intelligence transform the way we work and live. In addition to the many applications outlined above, AI will play a crucial role in addressing global challenges and accelerating the search for solutions.

But the ethical implications of AI will develop alongside it. As AI becomes more powerful and pervasive, we must ensure it is developed and used responsibly, addressing issues of bias, privacy and transparency. For this to be achieved, it is crucial to stay informed and be proactive in shaping its development, to build a future that is both beneficial and ethical for all.