What is Artificial Intelligence?

Artificial Intelligence (AI) has evolved from a niche scientific pursuit to a cornerstone of modern technology, used in various aspects of daily life and industry. AI can be broadly defined as the simulation of human intelligence processes by machines, particularly computer systems.

These processes include learning (acquiring information and rules for using the information), reasoning (using rules to reach approximate or definite conclusions), and self-correction. AIs are algorithms and systems that can perform tasks that typically require human intelligence, such as visual perception, speech recognition, and decision-making.

Throughout its history, AI has experienced hype and disappointment cycles, often referred to as "AI winters" and "AI springs." Each wave of interest brings new advancements and applications, pushing the boundaries of what machines can achieve.

Despite periodic setbacks, AI consistently re-emerges as a focal point of innovation. Today's AI technologies are not just theoretical concepts but practical tools embedded in everyday devices and services.

Many businesses rely on AI for operational efficiency and innovative solutions. To be able to craft strategies to use it to its fullest while being aware of its limitations, you need to understand its scope, the various types, and its applications.

Types of AI

AI can be broadly classified into three types based on capabilities and functionalities:

• Narrow AI:

Narrow AI, or Weak AI, is artificial intelligence designed to excel at specific, well-defined tasks. Unlike the hypothetical concept of Artificial General Intelligence (AGI), which aims to replicate human-level intelligence across a broad range of domains, Narrow AI focuses on mastery within a particular area.

Think of Narrow AI as a specialist. An accountant is an expert at handling financial data, a chess program is a master strategist within the game, and facial recognition software excels at identifying faces in images. These are all examples of Narrow AI excelling in their targeted domains.

There are different techniques to achieve this specialization:

• Simple regressions

Simple regression, including linear regression, is a statistical method used to model the relationship between a dependent variable and one independent variable by fitting a linear equation to the observed data. It is often visualized as drawing a best-fit line through a scatter plot of data points.

The line, also known as the regression line, represents the best estimate of the dependent variable for given values of the independent variable. This method is used to make predictions or infer causal relationships between variables.

A simple linear regression could predict a house's price (the dependent variable) based on its size (the independent variable). The model learns the average relationship between size and price, allowing it to estimate the price of a new house based on its size alone.

• Machine learning and neural networks

Machine learning is a field where algorithms learn patterns and make predictions from data. Many modern AI applications are neural networks, which are complex structures inspired by the interconnected neurons in the human brain.

Source: Paper by Tekerek and Bay via ResearchGate

Neural networks are composed of layers of interconnected nodes, each of which processes and transforms information. By adjusting the strength of connections between nodes (known as weights), neural networks can learn to recognize patterns in data and make accurate predictions.

There are three main types of machine learning used in Narrow AI:

• Supervised learning: This approach involves training a model on labeled data where the desired output is known. For instance, to build a spam filter, you could train a model on thousands of emails labeled as "spam" or "not spam." The model learns to recognize patterns in the email content that indicate spam and can then classify new emails accordingly. Read more here.
• Reinforcement learning: In this scenario, an AI agent learns by interacting with an environment and receiving rewards or penalties based on its actions. This is how AI has mastered games like chess and Go, where the agent learns winning strategies through trial and error. Read more here.
• Unsupervised learning: This type of learning involves finding patterns and structures in unlabeled data. It's often used for tasks like customer segmentation, where AI groups customers based on similar purchasing behavior or interests. Read more here.

Generative AI and Large Language Models (LLMs) fall under the umbrella of Narrow AI. Despite their impressive capabilities in generating text, images, or code, they are still specialists. A generative AI for creating realistic portraits like midjourney cannot write a financial report, and an LLM like Gemini can write different kinds of creative text formats but cannot control a robot or diagnose a disease.

Here's why they're considered Narrow AI:

• Focused training: Both generative AIs and LLMs are trained on massive datasets specific to their function. A portrait-generating AI learns from millions of images, and an LLM learns from vast amounts of text data.
• Limited scope: While impressive in their areas, they cannot apply their knowledge to entirely new tasks or contexts. The portrait AI cannot use its image knowledge to write a poem, and Gemini cannot use its writing skills to perform surgeries.

Narrow AI's specialization allows it to achieve high accuracy and efficiency within its domain. This is why Narrow AI is the most used form of AI today, powering applications like:

• Language translation apps: Translate between languages with impressive accuracy.
• Recommendation systems: Suggest movies, music, or products based on your preferences.
• Voice recognition technology: Understand and respond to spoken commands.

By understanding Narrow AI's focus and limitations, you can appreciate its capabilities in specific tasks while recognizing that it's not a substitute for general human intelligence.

• General AI:

General AI, also known as General Artificial Intelligence (AGI), remains the holy grail of AI research. It represents a hypothetical type of AI that transcends the limitations of Narrow AI by exhibiting human-like cognitive abilities.

Unlike Narrow AI, which thrives on specialized tasks and predefined datasets, AGI aspires to continuously acquire knowledge from various sources (text, images, experience) and integrate it into a comprehensive understanding of the world.

It would then apply its learned knowledge to solve complex problems in novel situations. This would entail analyzing information, identifying patterns, drawing inferences, and making logical deductions.

AGIs should have the cognitive flexibility to adapt strategies and approaches based on new information and changing contexts. This requires overcoming the brittleness of Narrow AI, which often struggles outside its specific domain.

Finally, it should be able to transfer learning to effectively use knowledge from one domain to solve problems in another. This highlights the need for a generalized learning mechanism that transcends specific datasets.

Developing AGI presents formidable challenges that push the boundaries of current AI techniques. Some key hurdles include:

• Symbolic vs. subsymbolic representations: How can we bridge the gap between the symbolic representations humans use for knowledge (concepts, rules) and the subsymbolic representations employed by machine learning models (neural network weights)?
• Causal reasoning: Understanding cause-and-effect relationships in the real world is crucial for intelligent decision-making. Current AI often struggles with this aspect, relying heavily on correlations within data.
• Common sense reasoning: AGI should possess a common-sense understanding of the physical and social world, allowing it to make inferences and fill in missing information. This is a complex task that requires integrating vast amounts of background knowledge.
• Explainability and transparency: As AI systems become more complex, understanding their decision-making processes becomes paramount. This is crucial for ensuring trust and ethical considerations in AGI development.

The race towards AGI is fueled by cutting-edge research in deep learning, natural language processing, and reinforcement learning. While a true AGI might still be distant, continued research in these areas will undoubtedly lead to further breakthroughs in artificial intelligence capabilities.

• Super AI:

The concept of Super AI, also known as Artificial Superintelligence (ASI), accurately reflects a level of AI that surpasses human intelligence in all respects, including creativity, general wisdom, and problem-solving abilities. This form of AI is purely hypothetical and speculative at present, as it means AI achieving capabilities far beyond the current state of technology.

Super AI represents a potential future in which machines mimic and exceed human cognitive abilities in all aspects. The development and implications of such technology raise significant ethical and safety concerns. These include control issues, the implications of decision-making surpassing human understanding, and the potential societal impacts of such AI systems.

The ethical and safety considerations surrounding Super AI focus on the need for robust governance and ethical guidelines to ensure that advancements in AI technology are developed and deployed responsibly​​.