What is Artificial Intelligence?

What is Artificial Intelligence?

As of 2021, researchers have employed the following classification of AI types:

1. Artificial Super Intelligence (ASI): This is a hypothetical AI that not only replicates human capabilities but surpasses them. Believers in ASI anticipate that it will have the power to penetrate human thoughts and emotions in order to subjugate them. See Superintelligence: Futurologists' Nightmare or the Real Future of Artificial Intelligence?

2. Artificial General Intelligence (AGI): AGI, also hypothetical, stands a step below ASI in terms of reasoning abilities. Advocates of AGI believe in the potential to create machines capable of at least performing the same actions as humans but are limited in their convictions compared to ASI.

3. Artificial Narrow Intelligence (ANI): ANI, or weak AI, exhibits subtle hints of intelligence in machine behavior. It is designed for a strictly defined narrow range of applications. In the case of ANI, there is no possibility of autonomous behavior or independent development beyond human control. Systems equipped with ANI can only exist in the form created by humans and theoretically cannot go beyond human control.

Outdated general definitions of artificial intelligence include:

• (J. McCarthy) AI develops machines with rational behavior.
• (Britannica) AI is the ability of digital computers to solve tasks typically associated with high human intellectual capabilities.
• (Feigenbaum) AI develops intelligent computer systems with capabilities traditionally associated with human intelligence: language understanding, learning, reasoning, problem-solving, etc.
• (Elaine Rich) AI is the science of teaching computers to do something that humans are currently better at.

The term "intelligence" refers to the quality of the mind, consisting of the ability to adapt to new situations, learn and remember based on experience, understand and apply abstract concepts, and use knowledge to manage the surrounding environment. Intelligence is a general capacity for knowledge and problem-solving that unites all cognitive abilities of humans: sensation, perception, memory, representation, thinking, and imagination.

In the early 1980s, researchers in the field of computation theory, Barr and Feigenbaum, proposed the following definition of artificial intelligence:

"Artificial intelligence is a branch of computer science that deals with the development of intelligent computer systems, i.e., systems with capabilities traditionally associated with human intelligence - language understanding, learning, reasoning, problem-solving, etc."

Later, AI came to encompass a range of algorithms and software systems distinguished by their ability to solve some tasks as a human would contemplate their solutions.

The fundamental properties of AI include language understanding, learning, and the ability to think and, importantly, act. AI represents a complex of related technologies and processes evolving rapidly, such as natural language processing, machine learning, expert systems, virtual agents (chatbots and virtual assistants), and recommendation systems.

Artificial Intelligence (AI)

Artificial Intelligence (AI) is a technology that enables a system, machine, or computer to perform tasks requiring intelligent thinking, thereby simulating human behavior through gradual learning using acquired information and solving specific questions.

A hypervisor is software designed for creating, launching, and controlling virtual machines. Different operating systems (OS) can be installed on them. These virtual machines are isolated from hardware systems and utilize the resources of the virtual computer on which they are running.

Integrating AI into mechanisms and systems allows for the automation of routine, labor-intensive, or complex processes, including enhancing their accuracy and productivity. Therefore, this technology serves as a crucial business resource.

Advantages of Implementing AI

The use of artificial intelligence (AI) and solutions based on it provides several advantages to businesses.

1. Elimination of Human Factor: Utilizing programmable, self-learning algorithms eliminates the risk of human error and allows for the discovery of solutions not readily apparent to humans.

   
   

   Risk Reduction: AI-powered machines can be employed in situations involving risk to human safety. For example, AI robots can replace humans in specific production areas or work in conditions related to natural disasters.

   
   24/7 Availability: Intelligent machines can be utilized continuously without breaks or weekends, as they are not affected by distracting factors.

   
   Adaptability: Within defined conditions, the application of AI solutions enables the discovery of fast solutions. For instance, AI in chatbots helps better understand the "natural" language of customers, find answers to intricately formulated questions, and manage a high volume of simultaneous inquiries.

   
   

   Quick Decision-Making: Applications, machines, devices, and other tools based on AI make decisions faster than humans. This capability is beneficial in production processes, data analytics, creating predictive models, calculations, and other tasks.

Challenges in AI Implementation

Several reasons impede the adoption and use of artificial intelligence.

1. Manual Data Labeling: Controlled learning (supervised) of neural networks requires manual data labeling, a time-consuming process.

   
   

   Data Volume Requirements: Training models demand a substantial volume of data gathered from various sources, structured, cleaned of irrelevant information, and brought to a common format. This requires an organized system and a team of specialists.

   
   Interpretability: Understanding the results obtained from AI algorithms in terms of decision-making logic can be challenging.

   
   Task-Specific Models: AI models are oriented towards solving specific tasks. For instance, an AI algorithm designed for detecting a particular type of fraud may not recognize other types — each task and condition require a distinct model.

   
   Data Quality Impact: If the initial dataset for training is distorted or insufficient, the results of AI work may be compromised. For example, using only red objects in the training set may lead to errors or discrepancies when a blue object is encountered during self-learning.

   
   Expertise Requirement: Adequate competency in working with AI and developing projects based on it is crucial for evaluating risks and making decisions at each stage of algorithm implementation.

Applications of Artificial Intelligence

• AI technologies find application in various fields.

Communication Information Systems: Recognizing voice commands, searching for relevant answers, and vocalizing them using generated human voices.

Transportation and Logistics: Developing unmanned vehicles and drones for automated delivery of goods and parcels to remote areas.

Financial Sector: Predicting risks, identifying fraudulent activities, assessing customer creditworthiness, and detecting and blocking malicious attacks. By 2023, it is projected that 90% of credit applications at Sberbank will be processed using AI.

Medicine: Diagnosing diseases, detecting early-stage abnormalities, and making long-term predictions about a patient's condition.

Military Industry: Developing new weapons, personal protective equipment, and devices for recognizing opponents in complex conditions.

Business: Conducting analytics, customer segmentation, developing personalized offers, optimizing routine work processes, and identifying risks and fraud.

Three Types of Artificial Intelligence

There are three types of artificial intelligence distinguished:

For AI to perform designated tasks, it must be pre-trained on real or similar tasks. Two methods are used for this:

Machine Learning: A method of teaching a system or machine through programming and strict control.

Machine learning is divided into three types:

Machine learning requires a relatively small volume of data. The training process is divided into small stages, and the results of each stage are consolidated into a block of output data. Features must be precisely defined and created by the user. Output data is presented as a number, such as ratings or classifications.

Deep learning differs from machine learning in that part of the process is hidden and lacks obvious logic. The user receives results that can be presented in any format: text, audio, number.

Deep learning requires a large volume of labeled source data and powerful computing equipment. The model learns not from the original dataset but from the training results obtained in the previous stage — each completed cycle serves as a teacher.

AI Methods: NLP, CV, Data Science

Natural Language Processing (NLP): Speech Technologies

Computer Vision (CV)

Applied for:

Data Science

Utilizing methods such as:

Narrow AI (Weak AI): This is the only type of AI currently available, applied in voice assistants, virtual reality systems, recommendation mechanisms, and other solutions.

General AI (Strong AI): AI with self-awareness and capabilities approaching those of humans. According to experts, strong AI will not be fully developed and available for use until at least the year 2075.

Super AI (Superintelligence): AI with complete self-awareness and formed thinking surpassing human capabilities. Super AI is presumed to be able to self-reprogram, create new directions, and algorithms without human intervention.

Supervised Learning: Training on labeled data sets with obvious patterns.

Unsupervised Learning: Training on unlabeled data sets without explicit patterns.

Reinforcement Learning: Sequential training on labeled and unlabeled data sets.

Deep Learning: A type of machine learning using neural networks. The training process is divided into stages and has a structure with multiple input, hidden, and output layers.

Texts: Recognize, automatically translate.

Speech: Recognize, generate.