How artificial intelligence becomes an effective business tool

By 2030, artificial intelligence will increase the volume of the global economy by $15,7 trillion, PWC expects. How AI became available for business thanks to machine learning and cloud services – in the material of Trends

Intelligence Laboratories

In the summer of 1956, a two-month scientific seminar on Artificial Intelligence was held at the American Dartmouth College. It brought together leading American scientists in the field of game theory, neural networks and AI. Participants did not set themselves global goals. They were just trying to see if it was possible to teach a machine natural languages, train it to formulate concepts and create abstractions.

It was at this seminar that computer scientist and cognitive scientist John McCarthy proposed using the term Artificial Intelligence. Just a year later, AI labs appeared at Carnegie Mellon University, Stanford, and MIT. So the study of artificial intelligence has become an official academic discipline.

The first AI projects looked like little more than toys. For example, in 1966, ten years after the Dartmouth seminar, the ELIZA program was created by the American scientist Joseph Weizenbaum. She imitated, or rather parodied, a conversation with a psychotherapist. ELIZA was able to highlight key words in the user’s statements and build template responses.

Until the 1990s, there were no noticeable breakthroughs in the field of artificial intelligence: it seemed that the technology would never take on a distinct form and would never come close to the all-powerful AI that science fiction writers wrote about. But when IBM’s Deep Blue supercomputer beat world champion Garry Kasparov at chess in 1997, artificial intelligence was taken seriously again. In the same year, NaturallySpeaking speech recognition technology appeared, which allowed machines to understand a person “by ear”. Many other projects followed in the field of machine translation, image recognition and classification, and object detection.

In the early 2010s, such a type of artificial intelligence as neural networks gained popularity again, as well as ways to train it – machine and deep. If earlier algorithms were trained mainly to perform specific tasks, now they began to master the so-called representations (features / representations) and learn to recognize images. This was possible due to the sharp increase in the power of computers. Calculations began to be performed using GPUs, which were able to speed up the process of training models by orders of magnitude.

According to one of the godfathers of AI, British computer scientist Geoffrey Hinton, thanks to deep learning, in the near future, a machine will be able to reproduce human intelligence. But so far, artificial intelligence lacks scale. The human brain has about 100 trillion synapses (points of contact between two neurons. — Trends). For comparison, GPT-3, the most advanced language model to date, uses 175 billion parameters.

Closer to business

One of the main obstacles to the widespread use of AI until recently has been the weak distribution of machine learning (ML) models. Businesses still have little understanding of how to implement them in business processes and products, or are poorly aware of their capabilities. The development and application of such models seem to entrepreneurs to be a costly and time-consuming process. But the situation is rapidly changing.

Just a few years ago, they really needed high-performance hardware to work with them. There were few powerful machines, their rent was expensive.

Cloud computing has changed the situation. They have radically democratized access to powerful computing infrastructure and provided convenient and understandable tools for working with AI.

So, already now AI Cloud cloud services from SberCloud can provide any business, from a multinational corporation to a startup, with the infrastructure and tools for solving AI problems. Moreover, AI Cloud services operate on the basis of the Christofari supercomputer, the most powerful in our country and the CIS, specially designed to work with artificial intelligence.

AI Cloud includes the ML Space cloud platform, a set of products for full-cycle ML development. Sberbank previously noted that when creating AI solutions, specialists spend only a third of their time training models. The rest goes to preparation and other routine. With AI Cloud products, data scientist teams can devote 99% of their time to model training, working from anywhere in the world via the cloud.

ML Space allows you to organize distributed training on more than 1 thousand GPUs (Graphics processing units – GPUs for high-performance computing). According to its creators, now it is the only cloud-based ML platform in the world with such functionality. With its help, resource-intensive models can be trained in a few hours. The service also includes the AutoML module, which, in fact, is a factory for the production of ML models for those companies and organizations that do not have their own data scientists

The creators of ML Space compare the combination of cloud technologies and new machine learning tools with the transition from manual labor to industrial production. Together, they significantly accelerate the creation and entry into the market of ready-made AI solutions and make artificial intelligence as accessible as possible for businesses.

To unlock the potential

According to PwC forecasts, thanks to artificial intelligence, the global economy could grow by an additional $2030 trillion by 15,7. The global AI technology market will add approximately 31% annually, analysts at Frost & Sullivan predict. The company is sure that in 2022 it will reach $52,5 billion. This is four times more than the volume that analysts recorded in 2017.

Among the main vectors for the use of artificial intelligence by companies are risk management and cybersecurity, routine automation, and assistance in making optimal decisions. In addition, businesses are successfully using AI to better gather information for forecasts and automate customer operations.

It is expected that by 2025 humanity will store about 175 zettabytes (175 billion GB) of data. Already today, most of them are generated not by people, but by machines – various information systems, sensors, the Internet of things. Obviously, processing all this information and extracting value from it for business is simply impossible without artificial intelligence and machine learning. Moreover, its number continues to grow.

According to IDC, the amount of data created over the next three years will exceed the amount of information that has emerged over the past three decades. And over the next five years, the world will generate three times more data than in the previous one. And this will encourage the active use of AI to collect and process information.

According to SberCloud CEO Evgeny Kolbin, it is the clouds that will become the main driver for the development of AI in the coming years, since only with the help of cloud technologies can one overcome the main barriers to the development of AI — insufficient availability of high-performance computing resources for working with AI and an acute shortage of specialists — data scientists, data analysts and data engineers. Now in almost all industries there is an acute shortage of highly qualified specialists for working with data. QuantHub, a platform that specializes in recruiting data scientists, has calculated that there is only one potential applicant for three job postings. According to Kolbin, it is the development of cloud ML services and AIaaS (Artificial Intelligence as a Service) that will allow artificial intelligence to fully reveal its business potential.

Subscribe also to the Trends Telegram channel and stay up to date with current trends and forecasts about the future of technology, economics, education and innovation.