Can a machine mimic the functions of the human brain?
Hello, dear colleagues,
Yes, machines can perform tasks resembling some functions of the human brain. Examples include recognising images, processing speech, learning patterns and solving problems. Success at a particular task does not, however, amount to reproducing the brain's entire operation or sharing a person's subjective experience. A sound assessment of artificial intelligence begins by asking which function is being reproduced, under what conditions and against which criteria.
What does “mimicking the brain” mean?
Functional imitation: A machine can produce a task outcome similar to a human's, such as classifying a component in an image or converting speech to text. It does not need to use the same processes as a human brain to achieve that outcome.
Biological modelling: Researchers can study the behaviour of nerve cells or particular brain circuits using computational models. The model's correspondence with biology is tested against experimental evidence. Modelling one brain region is a different objective from copying an entire human brain.
Subjective experience: Whether a system actually feels anything is a separate question. Functional performance, biological similarity and consciousness cannot be reduced to a single measure of success.
What do artificial neural networks reproduce?
An artificial neural network uses connected computational units and adjustable weights. Its “neuron” is a mathematical unit that does not embody every property of a biological cell. In artificial intelligence and machine learning, training can adjust a model's settings using examples. Deep learning uses networks with multiple layers to learn different features of the data.
Big data and data analytics help examine patterns that may be useful for learning. But data suitability and representativeness matter too. More data or a larger model does not guarantee more reliable results on every task.
The brain's scale should not be reduced to one exact number either. A cell-counting study published in 2009 reported an average of approximately 86 billion neurons in adult male brain samples. This is not a fixed count for every person. A model parameter, an artificial neuron and a biological neuron are also different concepts; comparing their counts directly can be misleading.
How can machines reproduce perception, language and memory tasks?
Image processing can help identify an object or defect in a photograph; speech processing can interpret spoken input; and natural language processing can support meaningful operations on text. Task performance depends on the inputs available and the evaluation method. Research into reading human thoughts likewise needs to be considered in terms of specific inputs and experimental conditions. It does not amount to unrestricted access to everything in a person's brain.
A machine can also store or use information in different ways: retaining conversational context, retrieving information from a database or updating parameters during training. All may be loosely described as “memory,” but their mechanisms differ. Having a conversation does not necessarily mean the model is being permanently retrained.
Where do neuromorphic computing and brain simulation fit?
Neuromorphic computing investigates computing and hardware approaches inspired by nervous systems. For example, Intel announced Hala Point in 2024, a system using Loihi 2 processors developed for research into brain-inspired computing. The existence of such a system does not demonstrate that every function of the human brain has been reproduced.
Brain simulation can aim to investigate selected biological processes within a model. The level of detail and the validation method matter. Similarly, discussions about transferring consciousness to another machine should distinguish copying a file, model settings or records from transferring a person's subjective experience.
A practical example from a production line
Imagine a robot identifying components on a conveyor and placing them in the appropriate bins. Robotics covers the mechanical action and the wider application, while the relationship between AI and robotics appears in tasks such as interpreting images. Applying machine learning to robotics can help improve classification using examples of different components.
Automation and AI work together in this example: a camera supplies an image, the model predicts a class and the control system executes the movement. From the perspective of cybernetics and system control, feedback allows the intended outcome to be compared with the actual result. This connects perception, decision and action.
Recognising a component does not establish that the system thought like a person. Engineers should measure the misclassification rate, response time, performance under different lighting conditions and behaviour when the result is uncertain. The engineering value lies in verified task performance rather than the scale of the claim.
Where can performance fall short?
Generalisation means applying what has been learned to different situations. A model that performs well on images resembling its training examples may produce different results with a new product, a different camera angle or changed lighting. Applications such as supporting human decision-making or predicting human behaviour should therefore explain the conditions under which performance was measured.
Data quality, error analysis and independent validation provide more concrete information than a statement that a system “works like a brain.” Cybersecurity also matters in connected systems: changes to data or the model can affect outputs. Success in one area should not be extended to every untested area.
What about emotion, consciousness and creativity?
Responding appropriately to human emotions does not demonstrate subjectively experiencing the same emotion. Statements from a machine questioning its own existence are not, by themselves, evidence of consciousness either. There is uncertainty about subjective experience in AI; researchers assess indicators derived from different theories.
A machine producing a work of art demonstrates the possibility of a creative output. The feelings it evokes in a person do not, by themselves, answer what the producing system might feel. Measuring capabilities and judging inner experience are separate assessments.
Conclusion: particular functions can be reproduced
Machines can produce outcomes resembling some tasks performed by the human brain and show strong performance in particular applications. Completely reproducing the brain's biological operation and human experience has not been demonstrated. Rather than declaring that this will certainly happen or can never happen, we should define the function, explain the limits and assess the claim against evidence.
Sources: Deep learning · Neuron-counting research · Neuromorphic computing · Indicators of consciousness in AI
You may share this article with attribution to Otomasyon AVM.