Artificial Intelligence is dominating the world. Researchers are studying its working mechanism, its effect on human beings, and how it can be improved by using concepts from psychology and human behaviour. A recent study shows that AI systems learn better when they start with simple tasks. This learning process is compared to the similar learning and development of humans and animals (Devitt, 2025). The study emphasises the concept of Kindergarten Curriculum Learning, which focuses on a curriculum design based on child development and cognitive theory. It involves learning basic knowledge first to build a strong learning foundation and then using cognitive skills to understand more complex topics. For example, in mathematics, children learn numbers first before moving on to complex topics like addition and division for problem-solving (Devitt, 2025). Researchers from New York University demonstrated this principle using Recurrent Neural Networks (RNNs).
They observed that when RNNs were first trained on simple tasks, they could process more difficult and complex concepts quickly and effectively. The study further compared this to traditional learning methods, which consist of AI learning multiple tasks with various difficulty ranges simultaneously. This was said to be inefficient and much harder for learning. These methods hinder reaching the goal of Artificial Intelligence (AI), which is to program and respond in ways similar to human cognition and behaviour (Devitt, 2025).
Read more: Exploring the Many Facets of Intelligence in Children
Study To Train AI
The study used Recurrent Neural Networks, which were applied for language development, speech identification, and generating images, videos, and text (Schmidt, 2019). RNNs were trained using the Kindergarten Curriculum Learning, and to observe this learning, the researchers experimented using laboratory rats. The aim was to get the rats to seek water in a box with multiple ports. However, the rats initially needed to learn to associate the delivery of water with certain stimuli, like sounds or lights, along with the delayed delivery of water after the stimuli. The rats had to develop the basic knowledge of multiple stimuli, like sound or light, before water, and they also learned the duration for the cues and then combined these tasks to successfully retrieve water (Devitt, 2025).
This similar observation was to be expected when applied to RNNs. The researchers found that the networks trained on the principle of increasing difficulty levels learned faster and performed better than traditional methods (Devitt, 2025). This highlights the evolution of the learning process of AI.
Past Research
The idea of training machine neural networks with a curriculum originates from the research done by Elman (1993). This concept was further broadened by Bengio et al. (2009) The initial experiment conducted by Elman (1993) was based on developing a simple grammar structure with a recurrent network, and the experiment observed that success lay in starting the earning with basic foundational knowledge that is limited in complexity and then gradually enhancing its resources. These ideas were not just explored in AI but also in fields like robotics (Sanger, 1994).
Theoretical Context
Biological Theory
The Recurrent Neural Networks (RNNs) are very similar to the functioning of neural networks found in the human brain (bRNNs). bRNNs are used for several functions, such as basic decision-making, storing spatial patterns in short-term memory, and retaining sequences of events in working memory (Grossberg, 2013). After learning and understanding these basic tasks, the neural connections were introduced to complex models that required prior knowledge and planning.
Cognitive Psychology
Cognitive psychology consists of several theories explaining knowledge acquisition, processing, storage, retrieval, learning, etc., to improve the performance of cognition and behaviour (Davies & Nisbet, 1981). Krueger & Dayan (2009) explored the use of cognitive psychology in curriculum learning for AI. Their research stated that the learning process is similar to that of a human infant, which starts with simpler tasks and then develops and understands more complex schemes.
Developmental psychology
The concept aligns with developmental psychology, particularly through the method of scaffolding described by Lev Vygotsky in his sociocultural theory. It outlines how teaching is conducted using simple concepts, followed by the gradual removal of support as the learner improves in each skill. (Van Der Stuyf, 2002).
Criticism and Conclusion
All methods have their benefits and disadvantages. While Kindergarten Curriculum Learning has its strengths, it may not be effective across all types of neural networks. Its success depends on the complexity of the task material, whether it can be broken down into simpler stages or if the task is inherently complex from the start. Additionally, while it aids AI in the long run, this process can lead to complications and may be time-consuming. Furthermore, it oversimplifies the differences in how AI systems and the human brain process information.
Another disadvantage is the biological comparison. While this type of learning can help AI become more human-like and efficient, it could create issues such as large-scale unemployment. In the event of a fault or malfunction, it may process incorrect information. In conclusion, although this method shows great potential, its limitations and implications must be considered.
FAQs
1. What is Kindergarten Curriculum Learning in the context of RNNs?
This refers to training RNNs by first exposing them to simple tasks, similar to how children learn basic skills, and then gradually increasing the complexity of the tasks.
2. Why is Kindergarten Curriculum Learning beneficial for RNNs?
It allows RNNs to build upon a foundation of basic knowledge, making it easier for them to learn and perform more complex tasks later on.
3. How is it different from traditional RNN training methods?
Traditional RNN training often throws all the data at the network at once, without a structured progression of difficulty. Kindergarten curriculum learning provides a structured approach to learning, starting with easy tasks and gradually increasing complexity.
4. How does it relate to human learning?
Just as children learn basic skills (like letters and numbers) before tackling more complex concepts, AI systems can benefit from a similar structured approach to learning.
5. Is this approach applicable to all RNN tasks?
While it can be beneficial for many tasks, it might not be necessary for all types of RNNs or data. The effectiveness of this approach depends on the nature of the task and the data.
REFERENCES +
Bengio, Y., Louradour, J., Collobert, R., & Weston, J. (2009). Curriculum learning. Proceedings of the 26th Annual International Conference on Machine Learning. https://doi.org/10.1145/1553374.1553380
Davies, G. M., & Nisbet, J. (1981). Cognitive Psychology and Curriculum Development. Studies in Science Education, 8(1), 127–134. https://doi.org/10.1080/03057268108559891
Devitt, J. (2025, May 20). Kindergarten for AI: Basic Skills Boost Complex Learning in RNNs – Neuroscience News. Neuroscience News. https://neurosciencenews.com/ai-rnn-learning 28986/
Elman, J. L. (1993). Learning and development in neural networks: the importance of starting small. Cognition, 48(1), 71–99. https://doi.org/10.1016/0010-0277(93)90058-4
Grossberg, S. (2013). Recurrent neural networks. Scholarpedia, 8(2), 1888–1888. https://doi.org/10.4249/scholarpedia.1888
Krueger, K. A., & Dayan, P. (2009). Flexible shaping: How learning in small steps helps. Cognition, 110(3), 380–394. https://doi.org/10.1016/j.cognition.2008.11.014
Sanger, T. D. (1994). Neural network learning control of robot manipulators using gradually increasing task difficulty. IEEE Transactions on Robotics and Automation, 10(3), 323– 333. https://doi.org/10.1109/70.294207
Schmidt, R. M. (2019). Recurrent Neural Networks (RNNs): A gentle Introduction and Overview. ArXiv.org. https://arxiv.org/abs/1912.05911
