The Renaissance of Artificial Intelligence

Speaker: Sergey Golitsyn, Head of T1 Artificial Intelligence, T1 Holding.

In terms of its importance and impact on humanity, it is compared to the emergence of the Internet and smartphones. We are talking about artificial intelligence, which has become one of the main topics of 2024. Although we have been experiencing a new wave of popularity of technology for several years, we have seen a breakthrough only recently. While AI is rapidly developing and penetrating all areas, people are increasingly thinking about ethical issues and sharing concerns about its uncontrolled use. How did we get to this point and what will happen next?

I asked ChatGPT to compare 2014, 2024, and 2034 in terms of AI development. Here's what he had to say:

2014: Artificial intelligence is used in search and recommendation systems, machine learning and process automation; most systems are limited in their capabilities and cannot make complex decisions.

2024: Machine learning and neural networks become more widespread and powerful; AI exhibits the ability to self-learn; AI applications in medicine, finance, industry, and other areas expand; safety and ethical issues arise.

2034: AI is an integral part of everyday life; used everywhere, from managing city infrastructure to healthcare; work processes are automated; ethical use of AI will become an international standard; regulation will allow the safe development and use of the technology.

Is this really true?

The Renaissance of Artificial Intelligence

The field of AI has experienced several “winter” periods, with scientists finding themselves at a dead end and beginning to doubt the technology’s prospects.

However, over the last 15 years, we have seen a new wave of interest in AI research. It is beginning to actively develop and benefit people, and examples of commercial application are emerging. Many problems that humanity has been racking its brains over for decades have been solved in a fairly short period of time – thanks to artificial intelligence.

In my opinion, there are several reasons for these changes. First, the development of the Internet and industrial automation opened up access to large amounts of data. At the same time, computing power reached the level necessary to process entire “data lakes”. And finally, advanced machine learning models appeared. The basis for training was already there – thanks to big data.

The mid-2010s marked a turning point, after which deep learning began to develop rapidly. It allows creating new types of applications for working with huge amounts of information. At the same time, variational autoencoders (VAE) began to develop. These are generative models that consist of two parts: one analyzes any data, the other produces a similar unique result based on it. For example, using VAE, you can write music or create an image.

For some time, simple models that handled analytical tasks well prevailed. Since 2015, new solutions have appeared – more advanced, cheaper and more accessible. The race for scaling begins. As a result, new architectures appear that allow language models to work more efficiently with graphic and text data. Large language models (LLM) are developing and natural language processing is improving, including machine translation and language modeling. Experts admit that this type of architecture may become the leading one in the further development of AI.

From here on, we see a significant leap in the scale and capabilities of LLM. New models are constantly emerging, more efficient and functional. They help in a variety of areas – from image generation to code writing. This is where the story of the familiar ChatGPT begins: its developer joins the race and releases several generative models – the predecessors of the GPT chatbot.

Then the potential of multimodal artificial intelligence is revealed, and neural networks appear that are capable of processing and linking different types of data: text, video, audio, images. For example, the popular DALL-E.

And these are just some of the achievements in the field of AI – the most striking and revolutionary. As we can see, the previous 10 years have been very dynamic and eventful for artificial intelligence – the industry has "come out of hibernation". From local research into the potential of deep learning, we have come as close as possible to the widespread use of technology right down to the household level.

Read more materials on this topic in Compass CIO

AI is becoming generative and ethical

By the beginning of 2024, several advanced generative language models had already appeared, including the most advanced GPT-4 to date and its closest competitor, Google's Gemini Ultra. They quickly found application in business and everyday life. Thus, in just a couple of months of 2023, 100 million people became active users of ChatGPT – a historical record. Today, analysts estimate their number at 180 million. More and more solutions are appearing on the market, and the speed of their development is growing exponentially. Perhaps, while you are reading this text, a new, more advanced model has appeared.

Just 10 years ago, technologies could barely cope with text and image recognition. Today, they take on complex analytical and creative functions and are used in a variety of areas. In medicine, AI can be useful for creating drugs, making diagnoses, recording data, and providing consultations. In agriculture, it can be used to accurately calculate fertilizer doses and forecast the weather. In the aerospace industry, it helps manage flights and model ships.

The labor market is also gradually changing: the demand for technical specialists and scientists who come up with new algorithms and models is growing. Among them are Data Scientists, ML engineers, ML/AI Scientists.

However, social and ethical aspects are increasingly coming to the fore, and the world is calling for rapid regulation of this area. There is growing concern in society about the rapid development of technology. The developers themselves share these concerns: in March last year, industry leaders called for a thoughtful approach to further work on AI. At the World Economic Forum in January 2024, Microsoft CEO Satya Nadella called on the world to reach a consensus on the challenges associated with the development of artificial intelligence, and UN Secretary-General Antonio Guterres warned big business against recklessness in the pursuit of growing profits. It is important to ensure the positive consequences of the spread of AI and to develop common security standards.

The Russian legislative framework for regulating AI is still in the process of "maturing". However, today an active community of private and public sector organizations has already formed in the country, which is also focused on developing concepts for legal and self-regulation in this area. Back in 2019, a single body was created for the development of AI in the country – the Alliance in the field of artificial intelligence. And two years later, its participants, leading domestic companies, adopted a Code that establishes the principles of interaction with the technology. In the coming years, issues of special regulation in this area will be given more and more attention both at the local and international levels.

What happens next?

AI is no longer something futuristic and has a chance to become an indispensable assistant to humans – just like smartphones once became. Today, it already solves complex scientific problems, recommends products to you on marketplaces, helps you find vacancies, and conducts credit scoring.

Among the potentially interesting tasks for generative artificial intelligence in the spirit of our time is code generation for domestic products: this will accelerate migration to Russian technologies and the achievement of digital sovereignty. The use of automatic documentation tools based on generative intelligence will allow for faster implementation of projects in IT, construction and science. Consequently, in the coming years, the technology can increase labor productivity and change the global economy in an unprecedented way. The main challenge here will be the formation of an ever-growing demand for new goods and services and the search for sales markets. The companies that are able to discover these opportunities will achieve the best results.

For the last 10 years, humanity has been working on applied AI systems, constantly increasing the number of parameters and creating new models. But the era of “more is better” is coming to an end. I think that now the attention of developers will be focused on increasing the capabilities of models and security issues. It is important not only to move forward, but also to ensure that new tools meet the interests of society.

There is an opinion that today we are on the threshold of creating strong artificial intelligence (AGI), capable of thinking and acting like a normal person. AGI are hypothetical systems that will have the same versatility as our brain and will be able to solve an unlimited number of intellectual problems. Such solutions do not exist yet, but work in this direction is already underway. Most experts agree that AGI will be available by the end of the 21st century. And according to the most optimistic estimates, the technology may appear as early as 2032. If we do not slow down, then, in my opinion, the forecast may turn out to be quite realistic.