Five Stages of AI Development: From Chat to Superintelligence

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

The development of artificial intelligence is often described in terms of increasing model size, training data volume, context length, benchmark performance, and computational resources. These parameters are important; however, they do not necessarily reflect the most fundamental changes in the behavior of an AI system. A system can become orders of magnitude more powerful while continuing to operate according to the same principle: a human asks — the AI answers; a human gives a task — the AI performs it. This paper proposes shifting the question. Instead of asking, “How much more capable has the system become?”, we ask: “Who determines what the AI should do next?”, “On what basis does the system choose?”, and “What exactly is the system capable of creating?” These questions form the basis of a conceptual model consisting of five stages: Chat Agent Autonomy Awareness Superintelligence The stages are not treated as strictly separated states. Development is understood as a gradual transition in which neighboring stages may coexist for extended periods. A system may simultaneously demonstrate characteristics of several stages while one of them gradually becomes predominant. The central hypothesis of the model is that the transition to a new stage is determined not by the number of functions available to the system, but by the emergence of a new fundamental mechanism that changes the very principle according to which the system operates. The five stages are described using the analogies of a library, pupil, student, colleague, and professor. These analogies illustrate the gradual transformation of the relationship between humans and AI — from obtaining information to assigning tasks, then to independently choosing a direction, the emergence of an own value system, and finally interaction with an intelligence capable of creating new principles rather than merely creating new objects within existing principles. Particular attention is given to three methodological propositions. First, the stages are cumulative but do not absorb one another: a system at a higher stage retains access to the modes of previous stages, while each mode preserves its own nature. Second, the key distinction between awareness and autonomy is the presence of an own value system: an autonomous system independently chooses a direction, but the criterion for that choice may remain external; an aware system chooses on the basis of internal values. Third, awareness and superintelligence are qualitatively different: an aware AI creates new things within existing principles, whereas a superintelligence creates the principles themselves. The fifth stage requires particular methodological caution. Its defining characteristic can be formulated conceptually, but the fundamental mechanisms capable of producing genuine superintelligence remain unknown. Therefore, the fifth stage is described not as an ordinary stage of transition, but as an epistemic horizon — a conceptually definable boundary whose internal nature remains open to further investigation. Keywords: artificial intelligence, AI development, AI agents, autonomy, autonomous AI, awareness, value system, superintelligence, self-developing AI, human-AI interaction, artificial intelligence systems