Understanding Artificial Intelligence
Artificial intelligence (AI) is a technology that enables computers and machines to mimic human intelligence and problem-solving abilities.
AI, whether operating independently or in combination with other technologies (such as sensors, geolocation, or robotics), can carry out tasks that would normally require human intelligence or intervention. Digital assistants, GPS navigation, self-driving cars, and generative AI tools like OpenAI's ChatGPT are just a few examples of AI in the news and in our daily lives.
As a branch of computer science, AI includes areas like machine learning and deep learning. These fields focus on creating AI algorithms modeled after the human brain's decision-making processes, allowing systems to learn from data and improve their predictions or classifications over time.
Artificial intelligence has experienced many cycles of hype, but even skeptics recognize the release of ChatGPT as a significant milestone. The last major advancements in generative AI were in computer vision, but now, the progress is in natural language processing (NLP). Today, generative AI can not only learn and synthesize human language but also other data types, including images, video, software code, and even molecular structures.
AI applications are expanding rapidly. However, as AI tools become more prominent in business, discussions about AI ethics and responsible AI practices are becoming increasingly important. To learn more about IBM's stance on these issues, please refer to "Building Trust in AI."
Types of Artificial Intelligence: Weak AI vs. Strong AI
Weak AI, also known as narrow AI or artificial narrow intelligence (ANI), is designed to perform specific tasks. This type of AI powers most of the systems we encounter today. The term "narrow" might be more fitting, as these AI systems are anything but weak—they enable robust applications like Apple's Siri, Amazon's Alexa, IBM's watsonx™, and self-driving vehicles.
Strong AI includes artificial general intelligence (AGI) and artificial superintelligence (ASI). AGI is a theoretical form of AI that would possess intelligence equal to that of humans, with self-awareness and the ability to solve problems, learn, and plan for the future. ASI would surpass human intelligence. Although strong AI remains entirely theoretical with no practical examples today, AI researchers are exploring its development. For now, the best examples of ASI come from science fiction, like HAL, the superintelligent, rogue computer in 2001: A Space Odyssey.
Deep Learning vs. Machine Learning
Machine learning and deep learning are subfields of AI, with deep learning being a subfield of machine learning.
Both machine learning and deep learning use neural networks to learn from vast amounts of data. These networks are programmatic structures inspired by the human brain's decision-making processes. They consist of layers of interconnected nodes that extract features from data and make predictions.
The difference between machine learning and deep learning lies in the types of neural networks they use and the level of human intervention required. Traditional machine learning algorithms typically involve supervised learning, where data needs to be labeled by humans. Deep learning, however, uses deep neural networks with multiple layers, enabling unsupervised learning from large, unstructured datasets without human intervention, making machine learning scalable.
The Rise of Generative Models
Generative AI refers to deep-learning models that can take raw data—like all of Wikipedia or the works of Rembrandt—and learn to generate new, statistically probable outputs based on that data. At a high level, generative models create simplified representations of their training data and use them to produce new works similar to, but not identical to, the original data.
Generative models have been used in statistics for years, but deep learning has expanded their applications to include images, speech, and other complex data types. One of the earliest classes of AI models to achieve this was variational autoencoders (VAEs), introduced in 2013, which became widely used for generating realistic images and speech.
Early examples of generative models, such as GPT-3, BERT, and DALL-E 2, demonstrate what’s possible. In the future, models will be trained on vast, unlabeled datasets for a variety of tasks with minimal fine-tuning. This shift from narrow AI systems to broad AI systems capable of learning and working across multiple domains is driven by foundation models. These models are trained on large, unlabeled datasets and fine-tuned for various applications.
Looking ahead, foundation models are expected to significantly accelerate AI adoption in enterprises. By reducing the need for data labeling, businesses will find it easier to implement AI, leading to more widespread use of AI-driven automation in critical areas. IBM hopes to bring the power of foundation models to every enterprise through a seamless hybrid-cloud environment.
Applications of Artificial Intelligence
AI has a wide range of real-world applications today. Some of the most common include:
- Speech Recognition: AI uses natural language processing (NLP) to convert spoken language into text. This technology powers voice search on mobile devices and enhances accessibility in various languages.
- Customer Service: Virtual agents and chatbots are replacing human agents in customer interactions, providing personalized advice, answering FAQs, and transforming customer engagement across websites and social media.
- Computer Vision: AI enables computers to interpret and act on visual data from images and videos. Applications include photo tagging, medical imaging, and self-driving cars.
- Supply Chain Management: AI-driven tools like adaptive robotics and predictive analytics optimize supply chain operations, enabling autonomous decision-making and improving visibility and transparency.
- Weather Forecasting: AI enhances traditional weather models with machine-learning techniques, making forecasts more precise and relevant to weather-sensitive industries.
- Anomaly Detection: AI models analyze large datasets to identify unusual patterns, helping to detect faulty equipment, human error, or security breaches.
History of Artificial Intelligence: Key Milestones
The concept of a "thinking machine" dates back to ancient Greece, but key milestones in AI’s evolution include:
- 1950: Alan Turing publishes "Computing Machinery and Intelligence," introducing the idea of a test to distinguish between human and machine intelligence, now known as the Turing Test.
- 1956: John McCarthy coins the term "artificial intelligence" at the first AI conference at Dartmouth College, marking the birth of the field.
- 1967: Frank Rosenblatt develops the Mark 1 Perceptron, the first computer based on a neural network that learned through trial and error.
- 1980s: Neural networks using backpropagation algorithms become widely adopted in AI applications.
- 1997: IBM's Deep Blue defeats world chess champion Garry Kasparov, demonstrating AI's growing capabilities.
- 2011: IBM Watson wins "Jeopardy!" against champions Ken Jennings and Brad Rutter, showcasing AI’s ability to understand and process natural language.
- 2016: DeepMind's AlphaGo defeats Go champion Lee Sedol, highlighting AI's potential in complex strategic games.
- 2023: The rise of large language models (LLMs) like ChatGPT marks a significant advancement in AI, with deep-learning models trained on vast datasets revolutionizing AI's potential in various fields.