Computational neuroscience attempts to understand the brain by making mathematical or software models of neural systems.
For the most part, the models are not yet sophisticated enough to be as complex as the neural systems they replicate.
In most cases, models look at both the brain's local and global structures and relations. Connectionist models in particular
build more complex models of cognition or brain function on these simpler parts. The end point of this pursuit would be models that encompass a full understanding of the function of all brain systems. (9)
This "qualitative" approach offers a formalized account of the way that human behavior results from the functioning of the simple parts, but they might not succeed in exhibiting the intelligence or complexity of human behavior.
The "quantitative" approach would instead examine the biological details of neuronal composition, structure, behavior, and interaction. The aim would be to provide a complete list of the brain's biological parts and to accurately model their interaction.
Given this information increasingly large and complex simulations of neural systems can be created. WBE represents the logical conclusion of this kind of quantitative model: a 1‐to‐1 model of brain function. (9)
[Here then is something incredibly interesting and important about Bostrom's & Sandberg's approach: it is based on emergentist principles. According to this theory, such higher-level properties as consciousness "emerge" from the lower level properties of the neuronal activity. This is why Bostrom & Sandberg do not think we need to obtain "whole system understanding." For, so long as we replicate the lower-level properties of brain functioning, we can expect such higher ones as consciousness to emerge. Thus if we can replicate these neuronal properties accurately enough in computer software, Bostrom & Sandberg would conclude that thereby we have created consciousness. A mind will "emerge" from the computer's operations just as it emerges from the totality of all our brain's neuronal activity.]
As Bostrom & Sandberg write:
Note that the amount of functional understanding needed to achieve a 1‐to‐1 model is small. Its behaviour is emergent from the low‐level properties, and may or may not be understood by the experimenters. (my emphasis, 9)
Computational neuroscience often uses hybrid models that are dually quantitative and qualitative in nature, because both approaches have their advantages and disadvantages.
There are interacting factors that determine the sort of model used to create a neural emulation.
1) the need to make the model faithful to the biological properties of the original neural system: what Bostrom & Sandberg call "biological realism."
2) the extent to which conditions allow-for either quantitative or qualitative simulations: "tractability."
3) the manner by which the experimenter mentally represents the main components of the system: "understanding."
If the mind will emerge from the workings of the lower-level parts, brain emulation scientists then do not need a highly sophisticated mental representation of how the emergence happens. They only need to know the conditions which bring that about. But then they also need a firm grasp on the dynamics of the lower-level processes. This then requires a high degree of biological realism, so that they can be accurately replicated.
Yet, what we want to obtain are such higher-level phenomena as consciousness and intelligence. So we have to have some sense of the dynamics of higher-level functioning to test the emulations and to determine the data that will confirm their success. Nonetheless, the focus will still remain almost exclusively the lower-level activities, which are the most essential for creating the conditions that allow for minds to emerge. (9-10)
No comments:
Post a Comment