Abstract
The prevalence of artificial intelligence (AI) is witnessed by the emergence of connectionist models that mimic the relationship between the neurons and the synaptic networks in the human brain. These models that use neural networks are a radical shift from classical symbolic AI: rather than explicit rules; they code knowledge of strength and connectivities of connections. However, using connectionist models in AI is very difficult to interpret and vulnerable to adversarial attacks; machine intelligence is also an ethical concern. However, these difficulties still maintain the attractiveness of connectivist models, which are more convenient and flexible in representing knowledge and performing complex tasks than other approaches.
The emergence of connectionist models, or neural networks as they are also known, that imitate the highly complex interconnection of neurons in the human brain has greatly transformed the field of artificial intelligence. Unlike the symbolic AI that uses explicit rules and logical reasoning, the connectionist models represent knowledge in a distributed form, where the information is spread out over a network of interconnected nodes (Zhang et al., 2021). These connectionist models are different in that they are not rule-based. Instead, they learn from data using a mechanism that mimics human cognition, and as a result, they adapt and generalize across many tasks. As such, connectionist models have gained much popularity since the development of computational power, algorithms, and databases boosts them.
Nevertheless, the appearance of connectionist models in the artificial intelligence domain brings both opportunities and challenges. However, these models offer a more biologically plausible way of understanding cognitive functions and linking perception and cognition. Though the connectionist models lack interpretability, resistance to adversarial attacks, and ethical issues, they are still the main obstacle to their application (Taylor & Taylor, 2021). However, the dominance of connectionist models in AI can transform the perception of intelligence as we move towards almost human-like AI in cognitive processes.
New Issues in AI
Connectionist models are very promising; nevertheless, they have some problems that prevent their widespread usage and implementation in practice. Transparency is another issue, where neural networks work as black-box models; no method exists to explain the decision-making process. Neural networks differ from symbolic AI because they are difficult to interpret and do not allow us to see how the conclusion was reached (Zhao et al., 2022). This lack of transparency is especially critical in areas where AI decision-making can create alienation, as in healthcare and criminal justice.
Also, the connectionist models are vulnerable to adversarial attacks, in which the model gives faulty answers by only employing slight changes to the given data. These attacks take advantage of weaknesses in the network systems they are based on, such as the system’s sensitivity changes in the input patterns (Hu et al., 2021). The use of adversarial examples created specifically to trick AI systems without prior knowledge, giving out the human grain of intelligence, adversely affects the current machine learning algorithms and, therefore, raises questions of reliability regarding AI in life-threatening issues. This flaw requires constructing resilient and dependable AI systems that can work well under malicious attacks even when they lose their performance.
The issue of the ethics dimension in the argument for AI is another main issue because it causes problems such as biases, fairness, and accountability. The notion of a society where artificial intelligence technologies are widely applied is also being raised about their ability to enhance and aggravate the existing inequality problems. Algorithmic bias refers to a situation where AI systems may contain unconscious biases in the training data, leading to discriminatory outcomes such as employment, credit, and criminal justice (Zhang et al., 2021). In addition, the lack of a monitoring and correcting system for biases in neural networks leads to uncertainties about whether the algorithms are accountable and transparent. There is a need for interdisciplinary approaches to develop AI systems that strive for fairness, equity, and social justice.
Impact on Cognitive Psychology
The addition of connectionist models into the scope of cognitive psychology substantiates a novel approach to studying human cognition. Traditional studies on cognitive science have used abstract symbols and concepts of information-processing theories and AI. On the contrary, the connectionist models are different in that they resemble cognitive processes by interacting with simple neurons to form networks of linked units (Taylor & Taylor, 2021). This transition to distributed representation resembles the decentralized structure of the brain, which distributes the information across the entire network of neurons, thus encoding information.
Additionally, the connectionist models provide a neurobiologically more aligned explanation of the learning and memory processes. On the other hand, symbolic AI relies entirely on explicit rules and logical inference, unlike connectionist models that learn implicitly from the exposure to patterns in data via a learning mechanism known as “connectionist learning”(Zhao et al., 2022). During the process described above, the strength of connections between neurons is adjusted based on the input received and the desired output, a mechanism akin to synaptic plasticity in biological neural networks. Hence, connectionist models can cover sophisticated processes such as pattern recognition, generalization, and error correction, which are basic human cognitive aspects.
Consequently, the study of connectionist models presents an understanding of the behavior between bottom-up sensory input and top-down cognitive processes. Traditional cognitive psychology has paid prominent attention to the top-down processes (attention, perception, and decision-making) without considering their interaction with fundamental sensory mechanisms (Hu et al., 2021). Connectionist models provide an integrative framework that links sensory processing with higher-level functions, permitting researchers to investigate how sensory information helps form cognitive representations and vice versa. This integrated perspective conforms to Embodied cognition theories, which posit that cognition occurs due to the interaction between the body, the environment, and the mind.
Parallel to developing connectionist models, cognitive neuroscientists begin to identify neural substrates involved in cognitive processes. Cognitive neuroscientists simulate neural network dynamics, making predictions by comparing model outputs with neuroimaging data. The neural mechanisms behind cognitive phenomena are discovered with this approach (Hu et al., 2021). Take, for example, research using connectionist models that connected neuronal activity with language processing, decision-making, and social cognition, which found different brain areas responsible for these complex processes.
Also, connectionist models find their application in conventional cognitive tasks and traditional perception, motor control, and affective computing. These simulations then investigate the mechanisms of interaction between human beings and their surroundings in the present and activities, like sensorimotor integration, motor planning, and emotion regulation (Zhang et al., 2021). Researchers can create more reliable and ecologically valid cognition simulations in the complex and realistic interactive by connecting modeling with the connectionist model in the computational model of human performance.
Besides, connectionist models provide an efficient tool that may help us discover the issue of individual differences in cognitive activity and brain structure. The studies utilize network connectivity methods, learning rate, and topology variations to examine how the differences in cognitive abilities and brain structural features affect behavior and performance (Taylor & Taylor, 2021). The rationale of such a strategy belongs to the realm of personalized medicine and cognitive enhancement since cognitive-directed interventions are regarded as tools to improve learning, memory, and cognitive abilities.
Conclusion
The central thesis is that incorporating the system of the connectionist models into the AI domain is a revolutionary impasse with deep implications for cognitive psychology. However, these models present problems such as interpretability and vulnerability to adversarial attacks. However, they provide a more biologically plausible account of human cognition and offer key insights into the underlying relationship between bottom-up sensory input and top-down cognitive processes. Beyond this, connectionist models can be applied to cognitive tasks, artificial perception, motor control, and affective computing. That is why we can comprehensively view human behaviors in a dynamic real-world environment. In the context of the expansion of AI-related technologies, it becomes vital for AI researchers and cognitive psychologists to work intricately so that what connectionist models seek, taking our studies of human cognition to a new level, can be achieved.
References
Hu, Q., Lu, Y., Pan, Z., Gong, Y., & Yang, Z. (2021). Can AI artifacts influence human cognition? The effects of artificial autonomy in intelligent personal assistants. International Journal of Information Management, 56, 102250. https://www.sciencedirect.com/science/article/abs/pii/S0268401220314493
Taylor, J. E. T., & Taylor, G. W. (2021). Artificial cognition: How experimental psychology can help generate explainable artificial intelligence. Psychonomic Bulletin & Review, 28(2), 454–475.https://link.springer.com/article/10.3758/s13423-020-01825-5
Zhang, X., Wang, R., Sharma, A., & Deverajan, G. G. (2021). Artificial intelligence in cognitive psychology—Influence of literature based on artificial intelligence on children’s mental disorders. Aggression and Violent Behavior, 101590.https://www.sciencedirect.com/science/article/pii/S1359178921000446
Zhao, G., Li, Y., & Xu, Q. (2022). From emotion AI to cognitive AI.https://oulurepo.oulu.fi/handle/10024/45666
write