What is artificial intelligence?

Artificial Intelligence (AI) is a technology that enables computers and machines to simulate human learning, understanding, problem-solving, decision-making, creativity, and autonomy.

AI-equipped applications and devices can see and recognize objects. They can understand and respond to human language. They can also learn from new information and experiences, providing tailored recommendations to users and experts. Furthermore, they can act autonomously, eliminating the need for human intelligence or intervention (a classic example being the self-driving car).

However, in 2024, most AI researchers, industry professionals, and related headlines are focusing on breakthroughs in Generative AI (Gen AI)—a technology capable of creating original text, images, videos, and other forms of content. To fully understand Generative AI, it is essential to first grasp the underlying technologies upon which its tools are built: machine learning (ML) and deep learning.

Machine Learning

A simple way to think about artificial intelligence is to view it as a series of overlapping or derivative concepts that have emerged over the course of more than 70 years:

Directly under the umbrella of artificial intelligence lies machine learning , which involves creating models by training algorithms to make predictions or decisions based on data; it encompasses a wide range of techniques that enable computers to learn from data and derive insights from it without being explicitly programmed for specific tasks.

There are many types of machine learning techniques or algorithms, including linear regression, logistic regression, decision trees, random forests, Support Vector Machines (SVM), K-Nearest Neighbors (KNN), clustering, and others. Each of these methods is suitable for different types of problems and data.

One of the most well-known types of machine learning algorithms is the neural network (or artificial neural network); these networks are modeled after the structure and function of the human brain. A neural network consists of interconnected layers of nodes (acting as neurons) that work together to process and analyze complex data. Neural networks are particularly well-suited for tasks involving the identification of complex patterns and relationships within large volumes of data.

The simplest form of machine learning is called supervised learning , which involves using labeled datasets to train algorithms to accurately classify data or predict outcomes. In this type of learning, humans associate each training example with a labeled output. The goal is for the model to learn the relationship between inputs and outputs within the training data, enabling it to predict outputs for new, unseen data.

Deep Learning

Deep learning is a subset of machine learning that uses multi-layered neural networks—known as deep neural networks—which closely mimic the human brain's complex decision-making ability.

Deep neural networks comprise an input layer, at least three hidden layers (often numbering in the hundreds), and an output layer, unlike the neural networks used in classical machine learning models, which typically contain only one or two hidden layers.

These multiple layers enable unsupervised learning : they can automate feature extraction from large, unlabeled, and unstructured datasets, and generate their own predictions regarding what the data represents.

Because deep learning does not require human intervention, it enables machine learning on a massive scale. It is highly suitable for natural language processing (NLP), computer vision, and other tasks involving the rapid and accurate identification of complex patterns and relationships within vast amounts of data. Most artificial intelligence applications in our daily lives rely on some form of deep learning.

Deep learning also enables:

1. Semi-supervised learning , which combines supervised and unsupervised learning by using labeled and unlabeled data to train AI models for classification and regression tasks.
2. Self-supervised learning , which generates implicit labels from unstructured data, rather than relying on labeled datasets for supervisory signals.
3. Reinforcement learning , which learns through trial and error and reward functions rather than by extracting information from hidden patterns.
4. Transfer learning , where knowledge gained from a task or dataset is used to improve a model's performance on another related task or a different dataset.

Generative AI

Generative AI—sometimes referred to as "Gen AI"—refers to deep learning models capable of creating original, complex content, such as long-form text, high-quality images,realistic video or audio , and more, in response to a user prompt or request.

In general, generative models encode a simplified representation of their training data, then use this representation to create new work that is similar to, but not identical to, the original data.

Generative models have been used for years in statistics to analyze numerical data. However, over the past decade, they have evolved to analyze and generate more complex data types. This evolution coincided with the emergence of three advanced types of deep learning models:

1. Variational Autoencoders (VAEs) , introduced in 2013, enabled models to generate various forms of content in response to a prompt or instructions.
2. Diffusion models , which first appeared in 2014, add "noise" to images until they become unrecognizable, then remove the noise to generate original images in response to prompts.
3. Transformers—models trained on sequential data to generate extended sequences of content (such as words in sentences, shapes in an image, frames in a video, or commands in code)—form the core of most generative AI tools making headlines today, including ChatGPT, GPT-4, Copilot, Gemini, and others.

How Generative AI Works

In general, generative AI operates in three stages:

1. Training , to create a foundation model.
2. Fine-tuning , to adapt the model to a specific application.
3. Generation, evaluation, and further tuning to improve accuracy.

Training

Generative AI begins with a "foundation model"—a deep learning model that serves as the basis for various types of generative AI applications.

The most common foundation models today are Large Language Models (LLMs) , which were created for text generation applications. However, there are also foundation models for generating images, video, audio, or music, as well as multimodal foundation models that support multiple types of content.

To create a foundation model, practitioners train a deep learning algorithm on vast amounts of relevant, raw, unstructured, and unlabeled data—such as terabytes or petabytes of text, images, or videos from the internet. The training process yields a neural network with billions of parameters representing encodings of entities, patterns, and relationships within the data, capable of autonomously generating content in response to prompts. This is the foundation model.

This training process requires immense computing power, is time-consuming, and is extremely costly. It entails the use of thousands of clustered GPUs and weeks of processing, typically costing millions of dollars. Open-source foundation model projects—such as Meta’s Llama-2—enable generative AI developers to bypass this step and avoid the associated costs.

Settings

Next, the model must be fine-tuned for a specific content generation task. This can be done in various ways, including:

1. Fine-tuning , which involves providing the model with application-specific labeled data, questions or prompts the application is likely to receive, and the corresponding correct answers in the required format.
2. Reinforcement Learning from Human Feedback (RLHF) , where human users evaluate the accuracy or relevance of a model's outputs, enabling the model to improve itself. This can be as simple as people typing or speaking corrections to a chatbot or virtual assistant.

Generation, evaluation, and further tuning

Developers and users regularly evaluate the outputs of their generative applications and continue to fine-tune the model—sometimes as often as once a week—to enhance accuracy or relevance. In contrast, the foundation model itself is updated less frequently, perhaps every year or 18 months.

Another option for improving the performance of a generative AI application is Retrieval-Augmented Generation (RAG) , a technique that extends a foundation model to utilize relevant sources beyond its training data, thereby enhancing outputs for greater accuracy or relevance.

AI Agents and Agentic AI

An AI agent is an autonomous artificial intelligence program capable of performing tasks and achieving goals on behalf of a user or another system without human intervention, by designing its own workflow and utilizing available tools (such as other applications or services).

Agentic AI is a system composed of multiple AI agents whose efforts are coordinated to accomplish a task or achieve a goal that is more complex or larger than what any single agent within the system could achieve on its own.

Unlike chatbots and other AI models that operate within predefined constraints and require human intervention, AI agents and agentic AI demonstrate autonomy, goal-oriented behavior, and the ability to adapt to changing circumstances. The terms "agent" and "agentic" refer to the "agency" of these models—their capacity to act independently and purposefully.

One way to view agents is as a natural next step following generative AI. Generative models focus on creating content based on learned patterns, whereas agents utilize this content to interact with one another and with other tools to make decisions, solve problems, and complete tasks. For instance, a generative AI application might be able to tell you the best time to climb Mount Everest based on your work schedule, but an agent could provide that information and then use an online travel service to book the best flight and reserve a room at the most suitable hotel in Nepal.

Benefits of Artificial Intelligence

Artificial intelligence offers numerous benefits across various industries and applications. Some of the most common benefits include:

1. Automating repetitive tasks.
2. More insights from data, faster.
3. Enhancing the decision-making process.
4. Fewer human errors.
5. Available 24/7.
6. Reducing financial risks.

Automating repetitive tasks

Artificial intelligence can automate routine, repetitive, and often tedious tasks—including digital tasks such as data collection, entry, and processing, as well as physical tasks like warehouse inventory picking and manufacturing operations. This automation frees up workers to focus on higher-value, more creative work.

Enhancing the decision-making process

Whether used for decision support or fully automated decision-making, artificial intelligence enables faster, more accurate predictions and reliable, data-driven decisions. When combined with automation, AI empowers companies to seize opportunities and respond to crises as they arise—in real time and without human intervention.

Fewer human errors

Artificial intelligence can reduce human error in various ways, ranging from guiding individuals through the correct steps of a process and flagging potential errors before they occur, to fully automating processes without human intervention. This is particularly important in industries such as healthcare, where, for instance, AI-guided surgical robots enable consistent precision.

Round-the-clock availability and consistency

Artificial intelligence operates continuously, is available around the clock, and delivers consistent performance every time. Tools such as chatbots or virtual assistants can alleviate the workload on customer service or support staff. In other applications, such as material handling or production lines, AI can help maintain work quality and production levels when used to perform repetitive or tedious tasks.

Reducing financial risks

By automating hazardous tasks—such as animal control, handling explosives, and operations in deep ocean waters, at high altitudes, or in outer space—artificial intelligence can eliminate the need to expose workers to the risk of injury or worse. Although not yet perfect, self-driving cars and other vehicles offer the potential to reduce the risk of injury to passengers.

Artificial Intelligence Use Cases

There are numerous real-world applications of artificial intelligence. Here is a small sample of use cases across various industries to illustrate its potential:

Customer Experience, Service, and Support

Companies can implement AI-powered chatbots and virtual assistants to handle customer inquiries, support tickets, and more. These tools utilize natural language processing (NLP) and generative AI capabilities to understand and respond to customer questions regarding order status, product details, and return policies.

Fraud detection

Machine learning and deep learning algorithms can analyze transaction patterns and flag anomalies—such as unusual spending or unfamiliar login locations—that indicate fraudulent transactions. This enables organizations to respond more quickly to potential fraud and mitigate its impact.

Personalized marketing

Retailers, banks, and other customer-facing companies can use artificial intelligence to create personalized customer experiences and marketing campaigns that delight customers, boost sales, and prevent customer churn.

Human Resources and Recruitment

AI-powered recruitment platforms can streamline the hiring process by screening resumes, matching candidates with job descriptions, and even conducting initial interviews using video analysis.

Application Development and Updating

AI-powered code generation and automation tools can streamline repetitive coding tasks associated with application development and accelerate the large-scale migration and modernization of legacy applications.

Predictive maintenance

Machine learning models can analyze data from sensors, IoT devices, and operational technology (OT) to predict when maintenance is needed and forecast equipment failures before they occur.

Challenges and Risks of Artificial Intelligence

Organizations are racing to leverage the latest artificial intelligence technologies and reap their numerous benefits. This rapid adoption is essential, yet implementing and maintaining AI workflows entails challenges and risks.

Data risks

Artificial intelligence systems rely on datasets that may be vulnerable to data poisoning, data manipulation, data bias, or cyberattacks that could lead to data breaches.

Model risks

Malicious actors can target AI models for theft, reverse engineering, or unauthorized manipulation.

Operational risks

Like all technologies, models are susceptible to operational risks such as model drift, bias, and governance structure failures.

Ethical and legal risks

If organizations do not prioritize safety and ethics when developing and deploying artificial intelligence systems, they risk committing privacy violations and producing biased results.

AI Ethics and Governance

AI ethics is a multidisciplinary field that examines how to maximize the beneficial impact of artificial intelligence while minimizing risks and negative outcomes. AI ethics principles are implemented through a governance system that ensures AI tools and systems remain safe and ethical.

Common values associated with AI ethics and responsible AI include the following:

Explainability and Interpretability

As artificial intelligence advances, humans face the challenge of understanding how an algorithm arrives at a result. Explainable AI is a set of processes and methods that enable users to interpret, understand, and trust algorithmic outcomes.

Fairness and Inclusion

Although machine learning is inherently a form of statistical discrimination, such discrimination becomes unacceptable when it grants privileged groups a systematic advantage and places underprivileged groups at a disadvantage.

Robustness and Security

Robust artificial intelligence effectively handles exceptional conditions—such as input distortions or malicious attacks—without causing unintended harm.

Accountability and Transparency

Organizations must implement clear responsibilities and governance structures. Furthermore, users should be able to see how the AI service operates, evaluate its functionality, and understand its strengths and limitations.

Privacy and Compliance

Many regulatory frameworks, including the General Data Protection Regulation (GDPR), require organizations to adhere to specific privacy principles when processing personal information.

Weak AI vs. Strong AI

To contextualize the use of artificial intelligence across varying levels of complexity, researchers have identified several types of artificial intelligence:

Weak AI (also known as "Narrow AI") refers to artificial intelligence systems designed to perform a specific task. Examples include "smart" voice assistants like Alexa and Siri, or chatbots on social media.

Strong AI ( also known as Artificial General Intelligence, or AGI) possesses the ability to understand, learn, and apply knowledge across a wide range of tasks at a level equal to or exceeding human intelligence. This level of AI is currently theoretical, and there are no known AI systems that approach this level of sophistication.

History of Artificial Intelligence

The idea of a "thinking machine" dates back to ancient Greece. However, since the advent of electronic computing, significant events and milestones in the development of artificial intelligence include the following:

1. 1950 : Alan Turing published the paper "Computing Machinery and Intelligence." In it, he posed his famous question—"Can machines think?"—and proposed a test now known as the "Turing test."
2. 1956 : John McCarthy coined the term "artificial intelligence" at the first artificial intelligence conference.
3. 1967 : Frank Rosenblatt built the Mark 1 Perceptron, the first computer based on a neural network that "learns" through trial and error.
4. 1980 : Neural networks that use the backpropagation algorithm to train themselves became widely used in artificial intelligence applications.
5. 1997 : IBM's Deep Blue computer defeated the then-World Chess Champion Garry Kasparov.
6. 2011 : IBM Watson defeated the champions on the game show Jeopardy!.
7. 2015 : Baidu's Minwa supercomputer used a specialized deep neural network to identify and classify images with accuracy exceeding the human average.
8. 2016 : DeepMind's AlphaGo program defeated Go world champion Lee Sedol.
9. 2022 : The emergence of Large Language Models (LLMs), such as OpenAI’s ChatGPT, brought about a massive shift in the performance and capabilities of artificial intelligence.
10. 2024 : The latest AI trends point to the continued resurgence of artificial intelligence. Multimodal models capable of processing various types of data are delivering richer, more powerful experiences. Smaller models are also making strides.

Now that you have become an expert in the fundamentals of this exciting world, it is time to put them into practice.

Are you ready to transform your words into a voice that captivates hearts and minds?

Start your journey into the world of audio content today.

Try the " Makina " platform for free and convert your first 1,000 characters into a professional voiceover in less than 5 minutes!

Start your free trial now.

Source: IBM.COM