Artificial Intelligence, neural networks, machine learning… These terms are buzzwords not only in IT world. And these words are usually sold very easily. But do these loud and pretty names have a direct relation to artificial intelligence, or neural networks similar to those that function in the human brain? Today we are debunking the myths with experts in Data Science Andy Bosyi and Mykola Kozlenko.
Recently, the head of the Artificial Intelligence laboratory of Facebook, Yann LeCun, has criticized the well-known robot Sophia, which is quite popular for its anthropomorphism and as a result of very active advertising by its creators. Last year Sophia became the first robot in the world to receive a citizenship. Despite all this, Yann LeCun is undoubtedly right, Sophia should be called a good chatbot, but “This is to AI as prestidigitation is to real magic”, as LeCun said. “This is just a good marketing and nothing more,” — agrees our expert Andy Bosyi. But if so, what is a true Artificial Intelligence then? Interestingly, Andy believes that nothing can be called so. So, let us start debunking the most popular myths:
Artificial Intelligence is actively developing in the world and there are already many examples of its implementation. A good example of AI is contemporary robots that are capable of keeping a conversation and even look like people.
Artificial Intelligence does not exist and it is very unlikely that it will exist in the future.
Modern robots which are so popular and attract so much attention in society actually work on the basis of sets of scripts. Yes, they look like a human, and yes they seem to produce some emotions, but they do not have any relation to AI.
“Intelligence cannot be artificial by its nature. If you think carefully about the concept of intelligence, obviously, it contains a lot of components: first of all, it is self-awareness. Can you imagine a computer that is aware of itself? It must also be able to identify various characteristics of the surrounding world. For instance, it must know that one thing is small, and another thing is big, and it must comprehend it by itself not because it has received this instruction in the scripts. True intelligence is capable of love and self-sacrifice, it also has a sense of humour — these things are completely irrational, they cannot be expressed by variables. True or natural intelligence is capable of creating completely new information, not based on some input data. If artificial intelligence reaches the level of human intelligence, it can no longer be called artificial. So, I believe that “Artificial Intelligence” is something that does not make any sense,”– Andy Bosyi.
I understand what you think. If artificial intelligence does not exist, what are all those companies that claim they create AI actually doing?
Basically, mostly they work with Data science. To be brief, Data science is a science that works with data processing and analysing to optimize decision-making processes or to solve certain tasks, whether for a particular company or for the society.
It is closely connected with Data mining, which is a collection of useful data and identification of the value of this data; understanding how data can be analysed and used. Data science is a difficult but very interesting science that requires knowledge of statistics and programming at the same time.
“A Data Science expert knows how to analyse data, to determine if this data carries some value, understands how this data should be used, and decides how to eliminate mistakes in data with the use of special models. He/she is able to map the results of data analysis onto real situations in the world, because not always can we get an accurate interpretation of a global situation even when we have abundant data about some local features. In order to connect data with the reality, we use statistical tools, formulae, probability theory, applied mathematics, and many other tools. Every data scientist should perfectly understand how simple data can be used to influence the client’s business.” — Mykola Kozlenko.
Neural Networks — is a breakthrough trend in the world of IT. These new systems work analogically to neural bonds in the human brain.
Neural networks work on the basis of the approach that appeared in the 1950s-60s. And these systems have nothing to do with what’s happening in the human head. A neural network is a convenient mechanism that receives input data, categorizes it, and can be trained to identify errors and take them into account in future work. This is a very convenient tool which helps to get a more accurate representation of real-world situation through the data collected.
There are also good mechanisms for Deep Learning. They create additional possibilities for the work with neural networks. For example, such systems are able to select classifiers by themselves. Let’s say, we build a system for a housing search, and there is a set of parameters for each apartment: location, area, whether there is a school nearby, and many others. We can set up the system so that some of these features will have an impact on the price. And when Deep Learning is applied, the system itself can choose the required parameters from a large amount of data to optimize the process of work with this data. This works according to an algorithm built in advance.
At the same time, neural connections in the human brain are chaotic and incomprehensible. Consequently, neural networks have no relation to artificial intelligence.
“A neural network is a great and successful classifier and regressor, but it has nothing to do with what’s happening in the human brain,” — says Mykola Kozlenko.
We have only a few years left before robots and machinery replace most human occupations, resulting in global unemployment problems.
Robots and machines will never be able to replace humans in many professions. They are bad at reacting to sudden unpredictable changes in their environment; they work well only in ideal conditions, not to mention their insufficient capability to solve creative tasks. Yes, they are perfect for handling routine work, but they are not yet reliable and safe in spheres where there is a high risk to people’s lives and health. We asked Andy Bosyi, if he had an opportunity to try riding in a self-driven car, whether he would do it. Andy jokes: “If they gave me a Tesla I would take it! But to be serious, I cannot confidently assure you that they are safe, because such cars work well only under a certain set of ideal conditions: perfect lighting, visible markings on the road, clear road signs, adequate drivers on the streets etc. Honestly, I would never ride it in our city in the evening.”
Today we already have many AI and Data Science solutions, they are around us and we use them in our everyday life.
Only a tiny part of tasks in this area are solved now, most of them only partially. Yes, now we frequently use voice recognition technology in search engines. However, voice recognition today is still far from ideal, because every person speaks in his/her own way. The image recognition problem is also only partially resolved because even if we turn a picture upside down, the system will no longer identify it as the one we search for. Self-driven cars still depend on ideal road conditions.
“You know, we now have a kind of dualism in our work: there are crucial tasks that should be solved and these solutions could bring so much benefit to businesses, but we have a lack of data to solve them because clients are not willing to give us their clients’ data. At the same time, there are tasks that are very difficult to solve as of now, but we have so much data to solve them. For example, text recognition: the amount of data is enormous, however, it is not easy to find a perfect solution. Even now there is insufficient technical capacity of computers to solve problems in this sphere, and of course, many solutions are not found because developers simply have not figured out how to do it yet. I can assure you that the prospects and the future of Data Science are incredible.”– Andy Bosyi.
Despite the fact that today we have changed a little bit some commonly held preconceptions about such popular concepts as artificial intelligence and neural networks, we want to tell you about the examples of interesting tasks that are actively and successfully solved by Data science today:
-Recognition of music sounds and human speech. There are efficient projects when a computer can accompany a person playing a musical instrument.
-OCR. Recognition of handwritten text in very old documents. For example, a company has an archive of various types of documents, such as invoices, contracts, regulations and many more. Originally, there were special employees, spending days trying to identify and organize those documents. But then a smart system was created. The system determines a type of each document and collects data from these documents (date, name of the company, sum of money and others) and fills in a special electronic database. How is this done? A system is trained to determine in which parts of the documents we can usually find its type, date and other information, or which symbols typically are contained in certain types of information; for example, if it is a date, then it probably may contain a “/” character. Today it is already possible to reach a 90% accuracy with such systems.
What is also very successfully implemented in OCR area today is the recognition of barcodes, QR-codes, and it is very widely used.
-Planning of the placement of solar panels on the roofs of houses in accordance with the most appropriate angle of the surface to sunlight.
-A visual count of the number of people entering and leaving a store during a day, defining and analysing a set of characteristics of these visitors.
-An analysis of satellite images of rooftops to determine whether or not the buildings have solar panels on them. It was made to generate leads for a company selling solar panels.
-Neural Networks in mobile eCommerce that analyse the duration of a user’s view of each product and instantly decide which product should be shown next to interest that particular user.
-Smart Merchandising: an analysis of customers’ behaviour, trends in sales, their reasons, making predictions based on tracking the sales boosts, depending on the product shelf location.
-Right now in the world, there are attempts to create systems that could recognize whether the news is real or fake.
-“The Young Pope”, which tracks comments on social networks and responds to them with phrases from the Bible.
-A model to determine the exact market value of real estate objects. Due to the developed algorithm, hundreds of static and dynamic parameters of each object of real estate (measurable features: area, location, and non-measurable: colours, availability of solar panels, quality of furniture, etc.) are analysed, as well as the dynamics of the market and the history of sales in the region. The developed model gives an accurate estimate of the value of real estate in 95% of cases, as if it was done by a human real estate expert.
-Diagnosis of cancer on the basis of analysis of a large amount of magnetic resonance tomography results and providing recommendations to patients.
-A profound analysis of social networks for the implementation of effective communication digital strategies.
-Search for the regularities, interrelations in the history of sales (by receipts, time periods, goods or groups of goods) and the construction of models to predict the effect of changes in certain parameters on sales.
-Analysis of users’ search queries and adaptation of the website to the user in a real-time mode.
This is only a very small part of what Data Science is capable of today. Data is a great value, you just need to know how to get this value from data. Do not hesitate to contact experts!
Source: Deep Learning on Medium