Artificial Intelligence vs Human Intelligence? Ft. Hrayr Harutyunyan

In September 2023, artificial intelligence (AI) was first introduced as a field of study in the Armenian public education system. The Foundation for Armenian Science and Technology (FAST) together with the Ministry of Education, Science, Culture and Sport of the Republic of Armenia started piloting the Generation AI program in 16 high schools of 6 regions of Armenia and Yerevan.

It all started in my 3rd year of undergraduate studies.

Being interested in computer science and mathematics, I began to explore artificial intelligence and machine learning in particular. Under the leadership of Hrant Khachatrian, I found other people interested in the same topic, and when YerevaNN was founded, I began to deal more methodically with machine and deep learning, making my first scientific steps in this direction.

Next, I started to collaborate with the Armenian professor Aram Galstyan of the University of Southern California. I did an internship at the USC Information Sciences Institute and was subsequently admitted to the university’s doctoral program in computer science.

I became actively interested in mathematics while studying at High School No. 4 in Goris. After success in the local math Olympiad, Sevada Khojabaghyan, who was recently honored at the "Hero of Our Times" award ceremony, contacted me and offered to join his group. There I developed my skills in mathematics and informatics.

When I was still in graduate school, I applied for membership in FAST’s NextGen Council, seeking to use my knowledge and experience to support science and technology initiatives in Armenia.

As a result, I got acquainted with the various programs being implemented and, in particular, with the idea of Generation AI. After addressing all my observations to the Foundation, I received an offer to deal with the content development of the program. It was clear to me that, if successful, this innovative program would have a very big impact in our country. So, I accepted the offer.

Currently, our working group consisting of Varazdat Avetisyan, Tsolak Ghukasyan, and Vahan Huroyan, and myself, is working on the educational content and materials based on the developed curriculum.

AI is a man-made phenomenon, the carriers of which are not people, but certain programs, robots, devices, or systems. Human intelligence, on the other hand, is a complex set of knowledge, skills, and abilities, consisting of elements such as understanding what is seen and heard, analyzing, learning from data, making decisions, being creative, or adapting to new situations.

Existing artificial intelligence systems are often designed to solve a specific problem and lack the generality specific to humans. Nevertheless, in some cases, they can outperform humans in their domain.

For example, a chess-playing AI is capable of surpassing any grandmaster. This becomes possible thanks to modern computing resources, which allow analyzing a huge amount of data and performing billions of operations in a unit of time.

Another important difference between human and artificial intelligence is the degree of independence. Current AI systems are not fully autonomous and need to receive human instructions from time to time. Sometimes we deal with self-sufficient systems that are limited in their range to solving certain problems.

There are still no AI systems with the generality and independence specific to humans.

At the heart of modern AI technologies are learning algorithms and artificial neural networks. The latter consists of artificial neurons, connected with each other in a specific way. Some of these neurons are input neurons, some are output neurons, while the remaining ones are intermediate neurons. 

Each neuron receives data from the neurons connected to its input, performs a calculation and transmits the obtained result to the neurons connected to its output. Thus, the value of the output neurons is determined from the values given to the input neurons. It’s important to take into account that the calculation performed by each neuron has some parameters, by changing which the behavior of the neuron can be changed, and therefore also the behavior of the entire network.

The goal of learning algorithms is to find such a configuration of parameters that the output data obtained from the input data corresponds to the solution of the problem we are interested in. For example, in the case of a neural network that distinguishes a dog from a cat, we would like that when presenting a picture of a dog or a cat to the input neurons, the only output neuron would receive the value of "0" or "1", depending on the picture in front of it.

In learning algorithms, many examples are used to find the desired configuration of parameters (as in the image-label pairs in the example above). The basis for training chatbots like ChatGPT is the billions of texts available on the Internet—more than one person is capable of absorbing in a lifetime.

Of course, artificial neurons were designed with the structure of natural neurons in mind. Nevertheless, the learning process of modern AI systems is quite different from the human one in its form and the amount of data used.

The learning process of AI is still not fully understood, and we don’t know how the neural network reaches the conclusions it makes or generates the images we receive. One thing is clear, studying this is one of the priorities for scientists.

AI today and AI in its early years are completely different things. What we see now has been formed as a result of the transformative power of education, namely the development of some branches of computer science and mathematics.

There is also the reverse process. Modern AI systems are already having an impact on education and are sure to have an even greater impact in the future. This has, of course, its desirable and not so desirable manifestations. On the one hand, AI chatbots help make good education more accessible and can somewhat complement teachers and mentors. On the other hand, students get a good tool to avoid doing homework on their own. This, however, is more of a problem with our learning culture than with AI systems.

Along with the development of AI technologies and becoming an integral part of everyday life, the issue arises of preparing a society whose members will have sufficient literacy to correctly assess AI technologies and their opportunities and risks and use them in the best way in their field.

We also have a problem with preparing good specialists and scientists in the field of AI. These two priorities entail a serious transformation of the educational system.

By implementing the program in the general education curriculum, students have the opportunity to deepen their mathematical knowledge with an improved methodology, learn the Python programming language, develop algorithmic thinking, and get acquainted with the foundations of modern AI.

Diving into more detail of the artificial intelligence component, I should mention that students will start studying it in the 11th grade.

In the first semester, we plan to get acquainted with the sub-branches and areas of application of AI, study the basics of machine learning and three representative methods (k-nearest neighbors, decision trees, and linear models). Through theoretical and practical lessons, students will learn to work with data, perform model evaluation, and gain basic knowledge of overfitting, regularization, and curse of dimensionality.

In the second semester, it’s planned to get familiar with the fundamentals of deep learning, simple unsupervised learning algorithms, and key components of modern natural language processing technologies. In parallel, through separate lessons and discussions, students will explore the history of AI, the scope of its ethical application, and potential risks.

Considering the growing influence of AI, the inclusion of the Generation AI curriculum in the entire public education system available to all is the equitable and logical thing to do.

Today, leaders and decision-makers around the world are interested in this question. It’s difficult to assess the impact of AI in the long term. One thing is clear. Societies that fall behind the wave of AI technology development will pay a heavy price for missed opportunities, just as was the case with societies that missed the industrial revolution wave.

In this context, Armenia has many gaps, but there are also advantages. Not being a large state and the homogeneous population give us the opportunity to make faster and more effective educational reforms, bypassing many of the complications arising from larger scales and differences.

Another advantage is the Armenian Diaspora, where we have wonderful experts in the field, and we must also get them involved in Armenia’s development.