What Is Artificial Intelligence? Definition, How It Learns, and Types
Artificial intelligence is the ability of computer systems to carry out tasks associated with human intelligence, such as learning, problem solving, and decision making. The article explains how these systems learn from data, their types, and the work they take on in companies.
Nurullah Okatan · Content Writer
· Updated: · 13 min read

According to the definition on Wikipedia, artificial intelligence is the ability of computers to perform tasks specific to human intelligence, such as learning, problem solving, perception, and decision making. These systems make predictions by extracting patterns from data, and today they are used in customer support, content generation, and forecasting.
The definition of artificial intelligence and the tasks it performs
Artificial intelligence transfers certain mental tasks that humans perform to a computer. The definition on Wikipedia lists these tasks as learning, reasoning, problem solving, perception, and decision making. Their practical counterparts are familiar tasks. A system sorts an incoming email by its subject, recognizes an object in a photo, or gives a written answer to a customer's question. Another forecasts next month's sales by looking at past data.

TÜBİTAK defines artificial intelligence as the general name for computer systems that imitate human intelligence. The word imitate has a narrow meaning here. The system does not think or understand the subject; it produces an output resembling a human decision. A translation tool does not translate correctly because it understands grammar. It translates correctly because it has learned from many sentence pairs which equivalent is more likely.
Midas, on the other hand, defines artificial intelligence as a branch of computer science that carries out targeted tasks through intelligent machines and applications. The phrase "targeted task" in the definition draws the boundary of the field. Every system is built for a specific job and is evaluated on that job. A list of chatbots, process automation, and content generation tools used by companies can be seen on the artificial intelligence services page.
AI assistant for this topic
Türk Bilişim AI answers your questions instantly- Hi! Ask whatever you're curious about.
How does artificial intelligence learn?
Artificial intelligence learns by finding recurring relationships in example data. TÜBİTAK calls this process machine learning and defines it as the most fundamental method for enabling artificial intelligence to learn. A model is shown many customer reviews labeled "positive" or "negative." It calculates which words appear together with which label. When a new unlabeled review arrives, it makes its decision based on that calculation.
In classical machine learning, humans often decide what the model should look at. For a house price model, features such as square meters, number of rooms, and location are selected by hand. Deep learning leaves this selection to the model. Midas defines deep learning as a subfield of machine learning and one of the building blocks of artificial intelligence. Multilayer artificial neural networks extract their own features from raw images, sound, or text. For this reason, deep learning requires more data and computing power. Face recognition and speech-to-text work with this method.
According to TÜBİTAK, the learning done by artificial intelligence is in fact a statistical prediction process. The model does not know the truth; it calculates the most likely answer. A chat tool, while writing a sentence, selects the word most likely to come next at each step. That is why it can produce a fluent but wrong answer. Such errors increase on topics that are scarce in the training data. In business use, grounding the model's answers in the company's own documents reduces this risk. Examples of business applications appear on the artificial intelligence page.
Looking for Automatic Social Media Visual Generation?The difference between narrow, general, and super artificial intelligence
All artificial intelligence systems in use today fall into the narrow artificial intelligence type. The classification compiled by Kapadokya University groups the types under three headings: narrow (weak), general (strong), and super artificial intelligence. The same source notes that all systems today are in the narrow category.
| Type | What it does | Current status |
|---|---|---|
| Narrow (weak) artificial intelligence | Works on a single task, such as translation, face recognition, or chat | All systems in use today are of this type |
| General (strong) artificial intelligence | Carries what it learns in one field over to other fields and takes on different tasks like a human | Has not yet been developed in practice |
| Super artificial intelligence | Surpasses human intelligence in every field | A theoretical concept |
According to Kapadokya University, general artificial intelligence has not yet been developed in practice. A narrow model cannot transfer what it has learned to another field. A model that plays chess cannot read a contract with the same knowledge. General artificial intelligence, by contrast, must be able to use knowledge gained in one area on a new problem. A system that carries out this transfer reliably does not yet exist.
The same source reports that some experts classify artificial intelligence into four categories based on functionality. Limited memory artificial intelligence is one of these categories. According to Midas, this type makes predictions, performs complex classification tasks, and is the most widely used type today. It keeps recent data for a short time and makes decisions based on it. A self-driving car that monitors the speed of nearby vehicles and decides when to change lanes is an example.
Milestones of artificial intelligence from 1950 to 2016
According to the timeline compiled by Bilişimle, in 1950 Alan Turing raised the question "Can machines think?" and developed the Turing Test. The test does not directly measure whether a machine thinks. A human interrogator converses through written messages. If the interrogator cannot tell whether the counterpart is a human or a machine, the machine is considered to have passed the test.

According to the same source, the term "artificial intelligence" was officially used for the first time at the Dartmouth Conference in 1956. Wikipedia also writes that the research field was founded in 1956 at a workshop at Dartmouth College. According to Midas data, the US Department of Defense showed interest in this technology in the 1960s. At that time, computers began to be programmed to imitate basic human reasoning.
Bilişim's timeline begins with two dates from games. IBM's Deep Blue defeated world chess champion Garry Kasparov in 1997. Google DeepMind's AlphaGo surpassed the world champion of Go in 2016. Deep Blue showed that a machine that rapidly calculates possible moves can beat a human master at chess. In Go, the number of possible moves is far greater than in chess, and brute-force calculation is not enough. AlphaGo showed that a model that learns from games can surpass humans in a field that requires intuition. New developments in the field can be followed in articles on artificial intelligence and automation.
Which tasks are companies using artificial intelligence for today?
Companies mostly use artificial intelligence for repetitive tasks. Answering customer questions, generating text and images, classifying documents, and forecasting demand are at the top of the list. According to a PitchBook study cited on Wikipedia, in 2024, 22% of startups receiving new funding described themselves as artificial intelligence companies.
A chatbot answering customer questions forwards the incoming message to a language model. The model first identifies the intent of the question: order status, return, or price? It then finds the relevant section in the company's own documents and writes the answer based on that section. For operations such as order inquiries, the bot connects to the company's system via an API. When it is not sure of the answer, it hands the conversation over to a customer representative. Türk Bilişim, described as Turkey's best AI Chatbot & Customer Assistant company, has been developing artificial intelligence solutions in chatbots, automation, and content generation since 2022. AI chatbots and customer assistants are part of these solutions.
A company wanting to start with artificial intelligence should begin with the process, not the tool. A workable sequence is:
- Choose a time-consuming and repetitive task, for example the most frequently received customer questions.
- Extract the data you have for this task: past correspondence, product documents, FAQ texts.
- Determine in advance how you will measure success, for example the rate of conversations handed over to a representative.
- Start with a small pilot and monitor the results for a few weeks.
- If the pilot reaches its goal, expand its scope to other processes.
If it is unclear which process to address first, AI strategy and consultancy work can help make that choice based on the data available. In companies with customers who speak multiple languages, the languages in which the bot will respond should be decided during the pilot phase. Preparing separate test questions for each language reveals translation-related errors before going live.
Looking for Custom AI Models & Integration?Frequently asked questions
What type of artificial intelligence does ChatGPT fall under?
ChatGPT is narrow artificial intelligence, because according to the classification of Kapadokya University, all systems in use today belong to this category. As a language model trained with deep learning, it generates text, and each answer is based on probability calculations.
Does using artificial intelligence require large amounts of data?
Training a model from scratch requires a lot of data. If a ready-made language model is used, the company's product documents, FAQ texts, and past correspondence are enough as a starting point for most tasks.
Will artificial intelligence completely take over jobs from humans?
No. Narrow artificial intelligence takes on specific steps of a task, while humans still manage decisions and exceptions. Because it works on a predictive basis, its outputs need to be checked regularly.
Are artificial intelligence and automation the same thing?
No. Automation works with pre-written rules and stops when a situation has no rule. Artificial intelligence learns from data and produces predictions even for inputs whose rules were never written; both methods can be used together in the same process.
Türk Bilişim · Free Quote
Get a free quote for your Automatic SEO Article Generation project
Leave your number and we will get back to you right away. The call is free.
Your request is saved on this device
We can't reach our server right now. Your request will be sent automatically once the connection is back. To talk right away:

Share this article
Comments
Comments
No comments yet.
















