A deep understanding of the principles underlying living organisms remains elusive, and novel approaches are needed. The rise of artificial intelligence (AI) has given researchers hope of applying these technologies to various fields, including tissue engineering and regenerative medicine 1,2. To date, AI-based solutions, such as ChatGPT and DeepSeek, have already been applied to natural language processing, computer vision, fraud detection, virtual assistants, autonomous vehicles, and robotics. Studies on AI applications in tissue engineering currently focus on solutions for structure-function relationships, material design, scaffold and device fabrication, and cell patterning 3,4,5. Most studies have concluded that machine learning models are well-suited for analyzing large, high-dimensional data, and that certain biological relationships can be readily analyzed using traditional statistical methods 1,6.

Therefore, AI holds promise as a unique tool for advancing the development of artificial organs and tissues, which are currently facing considerable challenges. Initially, it was expected that neural network models would contribute significantly to the development of bioengineering techniques7, but relevant principles have not yet been developed. Only a few such models based on engineered principles can be found in tissue engineering applications, such as finite element models 8,9,10, the phase space method 11, the minimax method 12, failure mode and effects analysis (FMEA) 13, reliability engineering 14, and design for manufacturability 15,16. The absence of AI-based generative models for tissue engineering and the design of human organs has led to substantial restrictions in this field. On the other hand, synthetic morphogenesis is a promising concept for overcoming limitations in tissue engineering approaches 17, but it has encountered challenges in genome organization and has yet to lead to significant breakthroughs in the field.

There are some interesting studies on the applications of AI in tissue engineering and regenerative medicine (TERM). For instance, computational methods are used to model scaffolds for neural tissue engineering. The finite element method (FEM) has been identified as an essential tool for modeling the boundaries and initial value problems that are characteristic of many physical processes related to the functioning of tissue-engineered scaffolds 18. However, FEM is a standard mathematical modeling procedure, so the benefits of AI integration are unclear.

A created convolutional neural network architecture is efficient at detecting and predicting continuous values for DNA and protein concentrations 19; however, this method is related to regression analysis because it is questionable and depends on initial conditions. A deep learning–based inverse design framework for disordered cellular materials (Voronoi lattices) with the desired target relative density and Young's modulus was developed for three-dimensional architected cellular materials 20. The applications and implications of artificial intelligence in regenerative medicine state that ‘AI has emerged as a powerful tool that analyzes the physicochemical and biological properties of a wide range of materials to predict the most successful outcomes’ and that ‘AI algorithms can identify patterns and associations in cellular behavior and interactions, thereby enabling the prediction of cell behavior in different environments’; however, these claims are not presented with evidence to support them 21. For example, studies on resorbable scaffolds for oral and maxillofacial tissue engineering have not offered examples or approaches for scaffold design 22.

The limited results obtained in these studies are unsurprising. A little-known property of complex systems is that “several simulation studies have found that using logistic regression for propensity score modeling can result in biased treatment effect estimates when the model is misspecified, especially in complex data where non-linear and non-additive terms are not included in the propensity score model. This situational bias occurs because parametric models, such as logistic regression, require assumptions about covariates’ functional form and distribution. When estimating propensity scores with many covariates, this problem can be particularly pronounced” 23. This means that the use of AI-based models for analyzing complex systems is preconfigured and thus subject to configuration-related limitations. For instance, Catalanotti, G. (2024) demonstrated success in contexts where the functions of relations between parameters have clear definitions 24. Unfortunately, this principle, rooted in information theory, is often overlooked by some researchers. Essentially, the critical question is whether physiological responses can be categorized as definite or indeterminate.

Moreover, the problem may be more profound because some researchers believe that modern mathematics is ineffective in biology 25, and that applying AI to biological problems is unsuccessful by definition, primarily due to the absence of a new kind of AI that uses methods of proof theory to provide justifications along with answers. This approach differs from calculation-based AI, machine learning, and data science methods.

Current AI models use correlation matrices to analyze large datasets, which are sensitive to initial values and boundary conditions 26. This highlights the intricate relationship between data structure and predictive accuracy. This sensitivity underscores the importance of carefully selecting and processing initial data because minor variations can lead to significantly different outcomes. Previous studies have shown the ineffectiveness of applying mathematical approaches in biology 25,27, indicating that understanding biological processes requires developing a new mathematical language 27. These aspects are likely the main cause of mathematics' unreasonable ineffectiveness in biology 25,28. The challenge lies in capturing the dynamic and often nonlinear nature of biological processes and translating these complex phenomena into a precise and flexible mathematical language that can capture the nuances of living systems.

Currently, approaches are being developed for designing and producing biological networks from high-level program specifications 29. However, the absence of engineering approaches in tissue engineering makes it seemingly impossible to adequately apply these methods 30. This review describes essential steps for applying machine learning and AI to achieve breakthroughs in tissue engineering and regenerative medicine.

Standard engineering approaches have utilized variational models, coordinate fields, the minimax principle, and other abstract mathematical concepts. However, these methods have not been adequately characterized in recent years. Moreover, manual manipulations are labor-intensive, complex, impractical, and exceed simple abstractible rules that can be formalized.

Despite years of attempts to establish tissue engineering as a rigorous engineering discipline, it remains largely empirical. The absence of regularities and valid limiting laws prevents the formulation of problems for which variational approaches could be used to find solutions. The potential of AI-based solutions is also limited by the need for biologically interpretable explanations of a model's output. Additionally, the effectiveness of such models in understanding underlying biological mechanisms and clinical applications is restricted 41,60,61. These issues can be summarized as the five challenges of artificial intelligence in tissue engineering (Figure 1).

Attempts to quantitatively describe biological self-organization mechanisms have thus far failed to formulate consistent systems of equations. These attempts have focused on synthetic morphogenesis 62, the standard morphological element approach 63, group theory 64, non-cooperative game theory 65, and parametric symmetries and their implications 66. These limitations highlight the opportunity to apply deep learning-based correlation models to transition toward formalized engineering methods. However, genetic technologies can extend the framework of existence beyond genomics. Throughout history, science has shown that prediction requires understanding. Therefore, without understanding the mechanisms underlying morphogenesis and the development of living organisms, designing them will be impossible.

Contemporary approaches lack proven techniques for describing the behavior of biological systems. However, a number of studies on quantitative and logical relationships in living systems offer hope, including those on cell fate 67,68, division of labor in tissues 69, and principles of intertissue connections 70. Undoubtedly, the crucial breakthrough will be in mathematics and mathematical logic, which will explain the identified patterns and effects that have previously escaped researchers' attention. Indeed, collaborations between biologists and physicists are already underway to understand evolution as multi-level learning 71,72,73.

All of this suggests that insights from the field of self-organization of chemical systems could be pivotal. The optimal paths remain unclear, as are non-cooperative games in living tissues or reinvestigating functional anatomy. Variant anatomy introduces uncertainty regarding the robustness of anatomical patterns 74,75. Organs and tissues may not develop entirely according to a single morphogenesis program, but rather undergo natural, nonpathological changes. This suggests that such changes are subject to mechanisms governed by human genetics and epigenetics.

Developmental biology is the study of the natural growth of tissues and the interactions between progenitor tissues. During embryogenesis, cells are capable of proliferation and interaction, including symbiotic interaction 76. As expected, natural interactions in the early stages of development can maintain neonatal growth and development, and present an original, non-conventional approach to understanding human anatomy by examining interaction variances.

The origins of competition among human tissues during morphogenesis follow complex underlying mechanisms. For instance, it has been discovered that tissues compete for amino acids 77. Game theory is a tool for understanding cell behavior in multicellular organisms 78, cellular sociology 79, and division of labor in organisms 80. In brief, this approach can predict the outcomes of complex organ transplants and even symbiotic interactions between two or more tissues within one serous membrane.

Nobel laureate Peter Medawar described immunological patterns associated with cell-to-cell communication and immunological compatibility during pregnancy 81. The issue is that pregnancy resembles organ transplantation because the fetus possesses paternal antigens and is a semi-allogeneic graft that can survive without immunosuppression. Hypotheses have been proposed to answer the question of how a mother supports her fetus in utero, now known as "Medawar's Paradox." The mechanisms governing fetal-maternal tolerance are incompletely understood but may provide critical insight into achieving immune tolerance in organ transplantation 81.

Another crucial category of physiological responses to both external (exogenous) and internal (intrinsic) factors is represented by tissue competitions, and these highlight the dynamic interactions within multicellular organisms 82. These competitions are not merely localized events, but are significantly influenced by, and contribute to, the broader physiological context. This context encompasses various types of tissues and their interactions. Cross-tissue connections play a pivotal role in macrospatial responses, showcasing how tissues communicate and influence one another across different regions of the body 83.

Initially studied in the context of developmental biology and cancer research, cell competition has revealed the mechanisms by which cells compete for survival, dominance, and space 84,85. When extrapolated to the level of tissues and organs, these mechanisms suggest the broader applicability of competition and cooperation principles in shaping the architecture and function of multicellular organisms 86. However, the understanding and formalization of these principles are primarily limited by the lack of physiologically relevant engineering approaches. The absence of underlying principles and models limits the support machine learning systems can provide to researchers and impedes the physical creation of tissue engineered constructs using irrelevant methods, such as 3D bioprinting 87.

The perspective of tissue morphology formalization shifts the traditional view of organ and tissue interactions from a harmonious, cooperative framework to one that recognizes competition and cooperative dynamics as fundamental aspects of biological organization and function 88,89,90. This new understanding of conservation laws in living systems opens avenues for exploring how tissues and organs negotiate space, resources, and functional roles within multicellular organisms. This framework challenges the conventional notion of cell competition as solely a microscopic phenomenon, suggesting that it is also mirrored and amplified at the tissue and organ levels.

Despite the dramatic rise of artificial intelligence technologies and deep learning approaches, there is currently no place for their application in tissue engineering, partly due to the lack of solutions for general requirement issues. Before artificial intelligence models can be applied to novel and groundbreaking ideas in this field, they need to be thoroughly described through logical constructs, and multitested on large datasets.

AI: artificial intelligence; FEA: failure mode and effects analysis; FEM: finite element method; ML: machine learning; TERM: tissue engineering and regenerative medicine.

Figure 1 was created using Biorender.com service.

All authors read and approved the final manuscript.

The present study was supported by the Russian Science Foundation, grant No. 25-24-20176.

Not applicable.

Not applicable.

Not applicable.

The authors declare that they have not used generative AI (a type of artificial intelligence technology that can produce various types of content including text, imagery, audio and synthetic data.

The authors declare that they have no competing interests.