We are proud to have prepared and compiled the best and most complete article about artificial intelligence or AI on the Persian web for you science lovers.
What is AI?
Going back to the 1950s, the fathers of this branch, Minsky and McCarthy, defined artificial intelligence as any action performed by a program or machine. Undoubtedly, such a definition is very general and therefore sometimes we see discussions about artificial intelligence. Artificial intelligence systems usually have some behaviors and actions similar to human intelligence, learning, reasoning, problem solving, knowledge presentation, perception, movement and in a range Less show social intelligence and creativity.
Applications of artificial intelligence
Today, artificial intelligence is everywhere, and to guide your online shopping, understand what you need to know about virtual partners such as Google Alexa and Apple Siri, photo detection, spam detection and fraud detection and card misuse. Credit used.
Different types of artificial intelligence
At a very high level, artificial intelligence can be classified into two types: limited artificial intelligence and general artificial intelligence. Limited artificial intelligence is the kind of thing we see in computers today. Intelligent systems that have learned or learned to do specific things without clear planning for how to do them.
This type of AI is evident in Apple’s Siri virtual language and speech recognition, in automated visual recognition systems, in engines that deliver products based on your purchase history. Unlike humans, these systems only learn how to do certain things and are therefore known as limited artificial intelligence.
Limited AI functions
Limited AI has many uses: interpreting video feedback from drones inspecting infrastructure such as oil pipelines, organizing work or personal calendars, answering simple customer service questions, and assisting radiologists in Identify tumors in photos, diagnose elevator malfunctions using data collected by IoT devices, and more.
General artificial intelligence functions
General artificial intelligence is very different and is a type of adaptive intelligence found in humans. A flexible form of intelligence that can learn how to do a lot of things (from cutting hair to making spreadsheets). General AI is more common in movies (like HAL in 2001 and Skynet in Terminator), but there is no such thing as external intelligence at the moment, and AI experts are working to make it happen.
A study conducted by four groups of experts in 2012/13 concluded that there is a 50 percent chance of realizing general artificial intelligence between 2040 and 2050, adding that by 2075 the probability is as high as 90 percent. will receive. The group went so far as to predict that “superior artificial intelligence” (intelligence with a power beyond human perceptual function) would be invented thirty years after the realization of the dream of general artificial intelligence.
Extensive research has been done on artificial intelligence, all of which are considered complementary. Machine learning is a factor in which a large amount of data is presented to a computer system, using which it learns how to perform a specific action (such as comprehending speech or photographing a scene).
The key to the machine learning process is neural networks. They are brain-inspired networks with interconnected layers of algorithms called neurons. They transfer data to each other and can perform special operations. A subset of machine learning is called deep learning, in which neural networks expand and form multiple layers to allow the transfer of large volumes of data. These are neural networks that enable computers to perform functions such as speech comprehension.
There are different types of neural networks, with different strengths and weaknesses. Return neural networks are a type of neural network that is particularly suitable for language processing and speech recognition, while annular neural networks are commonly used in image recognition.
Neural network design is also evolving, with researchers recently modifying a more efficient form of deep neural network called short-term memory, or LSTM, that can be used in demand systems such as Google Translate with unparalleled speed.
Another area of artificial intelligence research is evolutionary computing, borrowed from Darwin’s famous theory of natural selection, which finds that genetic algorithms undergo mutations and random combinations between generations in an attempt to find an optimal solution to a given problem.
This method has even been used to help design artificial intelligence models, the effective use of artificial intelligence to build AI. This type of use of evolutionary algorithms to optimize neural networks is known as neuroevolution and can play an important role in helping the efficient design of artificial intelligence as the use of intelligent systems becomes more common. Especially since scientists need data so much.
The technique was recently demonstrated by Uber AI Labs, which published articles on the use of genetic algorithms to train deep neural networks to solve learning amplification problems.
Finally, there are expert systems in which computers They are programmed with rules that allow them to make a set of decisions based on a large number of inputs, which allows the device to mimic the behavior of a human expert in a particular field. An example of such a knowledge-based system could be an autopilot system in a flying aircraft.
Stimulus factor of artificial intelligence
The greatest advances in artificial intelligence research in recent years have been in machine learning, especially in deep learning. This has been done somewhat with easy access to data, but even more so with the explosion in parallel computing power in recent years. During this time, the use of GPU clusters to teach machine learning systems has become more common.
These clusters not only provide much more powerful systems for teaching machine learning models, but now as cloud services through The Internet is also widely available.
Over time, large technology companies, such as Google and Microsoft , have begun to use specialized chips tailored to machine learning models as well as training.
An example of one of these chips is Google’s Tensor Processing Unit (TPU), the latest version of which is faster and uses useful machine learning models built using Google’s TensorFlow software library.
They can infer information from the data and also have the ability to interpret the amount of training. These chips are not only used to teach models for DeepMind and Google Brain, but also models that are the basis of Google Translate and Image recognition is used in Google Photo, and also provides services to the general public that can build machine learning models using Google TensorFlow Research. .
The second generation of these chips was unveiled at Google I / O conference in May last year, TPU is able to teach a Google machine learning model for faster translation.
Elements of machine learning
As mentioned, machine learning is a subset of artificial intelligence and is generally divided into two main categories: supervised and unsupervised learning.
A common way to teach artificial intelligence systems is to teach them using a large number of labeled or titled examples. These machine learning systems feed on huge amounts of data that are used to perform specific tasks. They can be pictures that show a picture or text sentences in a footnote to indicate whether the word “bass” is related to music. After training, the system can apply these titles or tags to new data, such as dogs in a newly uploaded photo.
Training in these systems typically requires extensive data, so that Some systems need to analyze millions of samples to learn how to do a task – although this is very possible in the age of big data and massive data mining.
There is a very large set of educational data – Google’s Open Image Dataset contains about nine million images, while the YouTube-8M video source contains seven million titled videos. ImageNet is one of the primary databases with more than 14 million classified images.
In the long run, access to large datasets with titles may also be less important than access to high computing power.
In recent years, GANs have shown how machine learning systems fed by small amounts of labeled data can generate huge amounts of fresh data for their training. This approach can lead to the emergence of learning. It almost leads to monitoring, where systems can learn how to perform their tasks using less labeled data than is necessary today for supervised learning systems.
Learning without supervision
In contrast, unsupervised learning uses a different approach, where algorithms try to identify data patterns and look for similarities that can be used to classify that data.
For example, you can collect garden fruits of the same weight or cars with the same engine size. This algorithm is not pre-configured to select specific types of data, but searches for data that can be based on its similarities. Be categorized, for example Google News collects stories with similar topics every day.
In reinforcement learning, the system tries to maximize rewards based on its input data, essentially going through a process of trial and error until it achieves the best possible result.
An example of reinforcing learning is the Deep Q network in Google DeepMind Is used for the best human performance in a variety of classic video games. The system feeds from each pixel and determines various information, including the distance between objects on the screen.
The system also creates a pattern by looking at the score obtained in each game, which maximizes performance in different situations. For example, in the case of the video game Breakout.
Leading companies in the field of artificial intelligence
By playing an essential role in artificial intelligence in modern software and services, each of the major technology companies is working to provide robust machine learning technology for home use and to expand their sales through cloud services to the general public. Each of them tries to go beyond the boundaries of this field, although Google with DeepMind AI AlphaGo has probably had the greatest impact on the public awareness of artificial intelligence.
Artificial intelligence services available
All major cloud operating systems – Amazon Web Services, Microsoft Azure and Google Cloud Platform – provide access to GPU arrays to train and implement machine learning models, and Google is using this array to enable users to Tensor processing units to use.
All necessary infrastructure and services are available from three major databases, cloud-based stores are able to hold large amounts of data needed to teach machine learning models, data conversion services In order to analyze them, visualization tools to display transparent results and software that simplifies model making.
These cloud platforms make it even easier to create custom machine learning models. Recently, Google introduced a service that makes it possible to create artificial intelligence models called Cloud AutoML automatically. These services create custom image diagnostic models that do not require any machine learning expertise.
Cloud-based services and machine learning are constantly evolving, and in early 2018, Amazon unveiled a new AWS for The simplification process is designed to teach machine learning models.
These services are for those firms that do not want to build their own machine learning models but instead want on-demand AI-based services – such as voice, vision and language recognition. – Suitable to use.
Leading countries in the field of artificial intelligence
Countries that cross the road?
It is a great mistake to think that the US tech giants have created the AI. Chinese companies Alibaba, Baidu and Lenovo are investing heavily in AI. China is pursuing a three-step plan to make AI a major industry for the country, a program worth 150 billion yuan ($ 22 billion) by 2020.
Baidu should focus on building self – driving cars, Using its deep learning algorithm, Baidu AutoBrain has invested and after several years of testing, plans to launch its fully automatic vehicles in 2018 and mass-produce them by 2021.
A combination of privacy laws, big investment, consistent data collection, and big data analytics by big companies like Baidu, Alibaba, and Tencent means that some analysts believe that China The field of artificial intelligence will be more superior than the United States .
How to start working with artificial intelligence
Although you can try to build a GPU array at home and start learning a machine learning model, it is probably the easiest way to test AI services over the cloud.
All large technology companies offer a variety of artificial intelligence services: from infrastructure to building and training their own machine learning models through web services that allow you to use AI-based tools such as speech, language, vision, and more. Access to emotion recognition if needed.
Artificial intelligence and changing the world
How will AI change the world?
Robots and drones
The desire for robots to be able to act independently and perceive the world around them means that there is a natural overlap between robotics and artificial intelligence.
While artificial intelligence is just one of the technologies used in robotics, its use in this area helps robots into new areas such as drones, as well as helping robots learn new skills. Find their way. GM recently announced that a car without a driver, steering wheel or brake will be built by 2019, and Ford has promised to do so by 2021.
We are on the verge of achieving neural networks that can make realistic images or make a personal voice quite accurate and similar. However, there is a possibility of very destructive social changes, such as making unreliable films and images. There are also concerns about how to use such technologies to exploit people’s images.
Speech and language recognition
Machine learning systems have helped computers recognize what people are saying with almost 95% accuracy. Microsoft Research Group recently reported that it has developed a system that can speak human language. Given that researchers pursue the goal of 99% accuracy, they expect human-machine interaction to become a standard or a norm alongside the usual forms of communication.
Face recognition and monitoring
In recent years, the accuracy of face recognition systems has improved so much that Chinese tech giant Baidu says it can match faces with 99% accuracy (provided the face is clear enough).
In China, authorities are implementing a nationwide program to connect CCTV cameras across the country to identify faces and use artificial intelligence systems to track suspects and suspicious behavior.
Police are also using face recognition glasses. Although privacy regulations vary around the world, it is likely that more use of artificial intelligence technology – including emotion-sensing AI – will gradually spread to other parts of the world.
Artificial intelligence can ultimately have a significant impact on health care, helping radiologists diagnose tumors and help researchers find disease-related genetic sequences and identify molecules that can help find more effective drugs. Experiments on artificial intelligence technology have been conducted in hospitals around the world.
These include IBM Watson Clinical Decision Support Tool, which is used by oncologists at Memorial Sloan Kettering Cancer Center, as well as the use of Google DeepMind systems by the UK National Health Service, where Helps you diagnose eye abnormalities and simplify the screening process for cancer patients. See the Technology News page for more news .