The progress in artificial intelligence is actually a turning point in the life of humanity. Over time, artificial intelligence performs many functions of the human brain, sometimes faster and more accurately. In recent years, technological progress in artificial intelligence has progressed from the level of machine learning to the level of deep learning, which is used in many fields such as disease diagnosis, stock market forecasting, and other fields.

The architecture of deep learning models consists of a structure called “Neural Network,” which is also known as “Artificial Neural Network(ANN).” So the relationship between artificial intelligence, machine learning, and deep learning is like nested circles. Perhaps the easiest way to imagine the relationship of the triangle of artificial intelligence, machine learning, and deep learning is to compare them to Russian matryoshka dolls. That is, in such a way that each one is nested and a part of the previous one. That is, machine learning is a sub-branch of artificial intelligence, and deep learning is a sub-branch of machine learning, and both of these are different levels of artificial intelligence.

Machine learning actually means the computer learns from the data it receives, and algorithms are embedded in it to perform a specific task. They identify patterns in the data and perform a preliminary analysis on it. Deep learning algorithms can be considered a complex evolution of machine learning algorithms. Deep learning describes algorithms that analyze data in a logical structure, similar to how the human brain reasons and makes inferences. To achieve this goal, deep learning uses algorithms with a layered structure called Artificial Neural Networks. The design of such algorithms is inspired by the biological neural network of the human brain, which leads to the learning process. As human decision-making is based on the combination of calculation and feeling, it has been tried to design new artificial intelligence algorithms in this direction. A clear example that can be presented in this field is the translation machine. If the translation process from one language to another is based on machine learning, the translation will be very mechanical, literal, and sometimes incomprehensible. But if deep learning is used for translation, the system involves many different variables in the translation process to make a translation similar to the human brain, which is natural and understandable. The difference between Google Translate 10 years ago and now shows such a difference.

One of the capabilities of machine learning and deep learning is stock market forecasting. Today, in modern ways, predicting price changes in the stock market is usually done in three ways. The first method is regression analysis. It is a statistical technique for investigating and modeling the relationship between variables. For example, the relationship between inflation rate and stock price fluctuations. In this case, the science of statistics is utilized to calculate the potential stock price based on the inflation rate. The second method in forecasting the stock market is technical analysis. In this method, by using past prices and price charts and other related information such as volume, the possible behavior of the stock market in the future is investigated. Here the science of statistics and mathematics (probability) are used together, and usually linear models are applied in technical analysis. However different quantitative and qualitative variables are not considered at the same time in this method. If a machine only performs technical analysis on the developments of the stock market, it has actually followed the pattern of machine learning. But another model of stock price prediction is the use of deep learning artificial intelligence, or ANN. Artificial neural networks can be very useful in modeling the non-linear processes that lead to stock prices and trends. Also, the percentage of neural network error is much lower than regression and technical analysis.

Neural networks operate in two phases: training and prediction. During the training phase, the neural network is fed with up-to-date market data, including cryptocurrency prices, trading volumes, economic indicators, and other relevant information. The network learns from this data and adjusts its internal parameters to recognize patterns and correlations. This human-supervised learning process forms the basis of the network's ability to predict the future. After sufficient training of the neural network, it enters the prediction phase. In this section, it receives new and immediate data and processes it through interconnected network layers. The network's internal parameters, which are formed during the training phase, enable it to make predictions such as future cryptocurrency prices or stock market trends. In the past, a multilayer perceptron in a neural network was used. The main shortcoming of this type of neural network was the lack of memory. In other words, it could not use the information in its memory and previous experiences to analyze the current situation because basically, there was no such memory. But today, a recurrent neural network with long short-term memory (LSTM) is applied. The recurrent neural network has the ability to store the input information and can use this information in the next stages of learning. By this way, it increases the network's ability to analyze the complex structure of relationships among stock price data. Today, many market applications such as Sigmoidal, Trade Ideas, TrendSpider, Tickeron, Equbot, Kavout are designed based on the second type of neural network and are considered to be the best applications based on artificial intelligence for predicting the stock market.

However, it is important to note that relying solely on artificial intelligence to predict the stock market may not be reliable. There are various factors involved in predicting stock prices, and it is a complex process that cannot be easily modeled. Emotions often play a role in the price fluctuations of stocks, and in some cases, the market behavior may not follow a predictable logic. Social phenomena are intricate and constantly evolving, and the effects of different factors on each other are not fixed or linear. A single event can have a significant impact on the entire market. For example, when former US President, Donald Trump withdrew from the Joint Comprehensive Plan of Action (JCPOA) in 2018, it resulted in unexpected growth in Iran's financial markets and a significant decrease in the value of Iran's currency. Iranian national currency depreciated %1200 since then. Such incidents can be unprecedented and have far-reaching consequences.

Furthermore, social phenomena are always being constructed and do not have a predetermined form in the future. The behavior of humans in some situations is not linear and just like the past, but humans may show behavior in future situations that is fundamentally different from the past. While artificial intelligence only performs the learning process based on the past or current data. Yet, artificial intelligence in order to learn and predict the stock market requires a lot of accurate and reliable data, which is usually not available to everyone. If the input data is sparse, inaccurate, or outdated, it loses the ability to produce the correct answer. After all, maybe the artificial intelligence will be inconsistent with the new data it acquires and will eventually reach an error. Updating and correcting artificial intelligence errors also requires high expertise and technical knowledge, which must be managed by an expert human. Another point is that artificial intelligence may do its job well, but humans do not fully trust it, simply because it is a machine. As passengers get into driverless cars with fear and trembling. In fact, someone who wants to put his money at risk in the stock market trusts human experts more than artificial intelligence. Therefore, although artificial intelligence technology can help reduce human errors and increase the speed of decision-making in the financial market, it is not able to make reliable decisions for shareholders alone. Therefore, to predict stock prices, the best result will be obtained if the two expertises of finance and data science are combined with artificial intelligence. Of course, in the future, with advances in artificial intelligence, its error rate may decrease, but the future of social phenomena, including the stock market, is always accompanied by uncertainty.

no