background Layer 1

AI and the Hierarchy of Needs: Do All Organizations Need to Assert Themselves? Part 1

Dmitry Basisty, Director of Strategy and Consulting Department at Rubytech, talks about what should be done to ensure that the new fashions of artificial intelligence technologies do not negatively affect the fair requirements of steady automation and digitalization.

In an ideal world, the level of process automation should strive to reach the maximum 100%, but in real life this is almost unattainable. As the degree of automation increases, its cost increases and, as a result, the expected economic effect decreases. As an indicator of the level of business process automation, 70-75% is a very decent result, and for a number of production processes, the automation level of 50% can be considered optimal. Before automation had time to reach its apotheosis – achieving maximum process efficiency, a new topic appeared – digitalization. One of its key goals is to increase the efficiency of both individual processes and entire enterprises and, as a result, reduce costs. A few years ago, a new trend emerged – the use of artificial intelligence (AI) technologies.

Human and organizational needs through the prism of Maslow's theory

Levels of needs. One of the popular and well-known theories of motivation, the theory of the hierarchy of needs, logically fits into the pyramid of human needs (Maslow's pyramid 1 ): from simple physiological needs through social needs to self-expression.

The human essence of these needs (love, knowledge, self-expression, etc.) can to a certain extent be transferred to an organization that owns its own information systems.

Organizational needs. Guided by the logic of Maslow's pyramid, any organizational need cannot be considered in isolation from the level of automation of its processes: the initial level of automation indicates the initial needs from the hierarchical model, a high level of automation opens the door to its advanced levels. It is natural that without satisfying the initial needs, a sudden transition to higher needs cannot but be accompanied by negative phenomena: first of all, this is the misuse of resources to achieve higher needs that are not in demand – for example, automation and digitalization for the sake of the process and following fashion, and not for the sake of the result.

Read more materials on this topic in Compass CIO

We will discuss the advantages and disadvantages of various approaches, as well as the factors for choosing the optimal model. We will consider possible intermediate solutions and key trends in IT development in the financial sector.

Social formations in the mirror of IT

In information technology, everything is approximately the same as in the development of society: it is necessary to strictly maintain the sequence of individual stages of development or socio-economic formations. If in society this is a primitive communal, slave-owning, feudal; capitalist, socialist and communist system; then in the development of an organization these are the stages of improvement, automation, digitalization, etc.

If you don't follow the sequence, the outcome will not be the most pleasant. And here's why:

  • automation of unclear, chaotic, incomplete in their totality business processes leads to the emergence of a phenomenon called “automation of chaos”;
  • starting digitalization from a low level of automation, without reaching the limit of automation and without using all its possibilities, is a direct path to the “digitalization of chaos”;
  • "under-digitized" organization – an organization that has not received (failed to receive) all the benefits of the transition to a digital model of activity, but boldly steps on the path of implementing AI technologies – a threat to itself. Such "intellectualization of chaos" gives rise to "thinking chaos", which carries within itself "chaos cubed".

It is worth noting separately that the mandatory sequence of movement from one formation to another does not entail a standardized duration of the stages of development, since they are very dependent on the cultural and economic characteristics of a particular country.

Deviation and “jumping” from one state to another (without proper automation to digitalization and beyond), neglecting the rule of consistent, evolutionary development of IT in the organization leads to unnecessary expenses and disappointment from results not fully achieved.

Assembling AI tools into a classification pyramid

There is a common mistake when, when saying the words "artificial intelligence", something like the legendary Skynet network is pictured in the minds of smartphone users, completely autonomous, complex and extremely innovative-capacious and necessarily – smart. Such a false understanding (or rather, misunderstanding) of the applied essence of AI leads to temptation, causing a desire to aim at implementing the most complex and, most likely, overwhelming, and most importantly, economically inexpedient tools and AI technologies.

To prevent this from happening, let's try to formulate a simple classification of AI 2 . All tools (instruments, technologies) can be divided into two segments: discriminative and generative AI. Tools from the first segment analyze the differences between data sets and classify them. The second segment contains tools that use data sets for training and generating (combination) new data based on them – for example, responses to queries.

Discriminative AI includes technologies such as image recognition and video analytics, text-to-speech and text-to-speech systems, preference-based advice systems, etc. (the so-called "classifiers" ). These technologies are the entry level in the pyramid of needs for AI technologies (level 1), which the vast majority of companies will be able to master. It is worth answering that such technologies have been developing for quite a long time, and there are many proven products and solutions on the domestic IT market – there is plenty to choose from.

Generative AI contains 4 types of technologies and tools:

  • chatbots (level 2);
  • robotic process automation (RPA) (level 3);
  • decision support systems (level 4);
  • autonomous intelligent agents (level 5, highest).

They are united not only by a related purpose (generation), but also by common technologies. Increasingly, although not yet necessarily, they rely on a special class of machine learning systems – large language models (LLM) or multimodal models (LMM) 3 .

Chatbots (Level 2 AI tools) have the following capabilities: they communicate only with humans, provide mainly reference information, and in principle may or may not use various AI methods, including decision trees, simple large language model (LLM) capabilities, etc.

Robotic process automation (RPA) (level 3 AI tools) is a technology for closing gaps in full-fledged automation of business processes, based on metaphorical software of robots (bots, artificial intelligence workers), which has developed into an independent automation tool. RPA technology can be implemented without generative AI. Evidence of this is the variety of products in the RPA class both in the global and domestic IT markets. This is no longer a new technology, widely used, in particular, in dedicated service companies – shared service centers. However, with the use of LLM, it reaches a fundamentally new level.

The listed AI technologies of levels 1–3 (classifiers, chatbots, robots) have a common property. These tools operate on the basis of instructions, i.e. they are not capable of “independent” decision-making “by design”.

The next two levels in the AI ​​technology needs pyramid (levels 4 and 5) are represented by tools that operate on the basis of authority, i.e. are capable of making “independent” decisions. Naturally, the tools of these levels, as well as the tools of the previous two levels (2 and 3), belong to the generative AI segment.

Decision support systems (level 4)

Decision support systems:

  • Generate forecasts, tactical scenarios and/or proposals for the organization’s strategy, including in key areas of its activity, based on all available (internal and external) heterogeneous information: data, facts, judgments – and using LLM/LMM and vector DBMS for its processing.
  • In many cases, they must be accompanied by auxiliary AI systems whose main purpose is to provide a human-readable explanation of the proposed measures.
  • The final decision in an automated decision-making process usually rests with an authorized person.

General-purpose AI systems (level 5). The original English name for this class of AI systems is GAI (General-purpose Artificial Intelligence). They are designed to solve a fairly wide range of problems. Including those that were not detailed at the time of their design/instruction, similar to how people can navigate in a changing environment. GAI does not necessarily have a level of intelligence similar to that of a human. This class of AI solutions includes machine learning (ML) systems that can perform various tasks: image recognition, language translation and data analysis, etc., but not necessarily at the level of human intelligence. GAI is a practical implementation of the ideal AGI (Artificial General Intelligence) system, which is currently perceived more as a theoretical concept than a working sample.

1 Maslow A. H. Motivation and Personality. – NY: Harpaer and Row, 1954.

2 The issue of classification of AI systems has received quite a lot of attention in the modern technological community. Among the variety of methodologies, frameworks and documents, as a fundamental scientific, but somewhat complex for a wide range of users approach to defining the main AI, one can note GOST R 59277-2020 "Artificial Intelligence Systems. Classification of Artificial Intelligence Systems".

3 LLM – large language model, LMM – large multimodal model.


We use cookies for analytical purposes and to deliver you the best experience with our website. Continuing to the site, you agree to the Cookie Policy.