AI models like GPT-4 demonstrate advanced language processing, but true consciousness and self-awareness in AI remain unresolved, necessitating further research.
Authors
Mykhailo Zhylin, Department of Practical Psychology, Educational and Scientific Maritime Institute of Humanities, Odessa National Maritime University, Odessa, Ukraine
Tamara Hovorun, Professor,Jindal Institute of Behavioral Sciences of (JIBS), O.P. Jindal Global University, Sonipat, Haryana, India
Bilal Alizade, School of political and Social Sciences, Western Caspian University, Baku, Azerbaijan
Maksym Kovalenko, Interregional Academy of Personnel Management, Kyiv, Ukraine
Alla Lytvynchuk, Department of Psychology, Faculty of Law, Public Administration and National Security, Polissia National University, Zhytomyr, Ukraine
Summary
The relevance of the research is determined by the need to study the consciousness of neural networks and the possibility of developing artificial self-awareness. The aim of the article is to investigate the main functional elements of models of consciousness in artificial intelligence (AI). The study employed such methods as the Turing test, Context-driven Testing and analysis of generation models. F1-score, Accuracy, t-test were used for statistical analysis. The reliability of the selected methods was checked by Test-Retest Reliability. The results were obtained that demonstrate the key aspects of the functioning of artificial consciousness models. GPT-4 shows the highest accuracy (92%) and F1-score (0.91), but has difficulties with complex logic problems. AlphaZero has the lowest accuracy (85%) and has trouble understanding abstract concepts. IBM Watson shows medium performance, but does not recognize irony well. DeepMind’s Gato is 90% accurate and wrong on coreference problems.
The resulting analysis showed that modern models, such as GPT-4, have a high level of development of perception and attention, which contributes to the effective processing of natural language. However, the question of true consciousness and self-awareness of AI remains open, requiring further research. Understanding the functional components of consciousness is important for the development of ethical norms in the field of AI. Therefore, it is necessary to improve the algorithms to grade up the cognitive functions of the models. Prospects for future research in neural network consciousness include an in-depth study of the mechanisms that provide true consciousness and self-awareness in artificial systems.
Published in: Journal of Theoretical and Applied Information Technology
To read the full article, please click here.
