Schedule of Topics
Wiki Stuff |
Topics /
MindsAndBrainsAssigned Reading
Additional Resources and Readings
Then It's A Duck! - Mark Schar Attempting to Understand Understanding -- Matt GarrI found Searle’s Chinese Room situation to be initially a highly alluring example, and when reading it, the intuitions somehow just feel right. As a conscious human, I feel that I can make a good judgment on my own level of understanding. My hunch is that my ability to evaluate my level of understanding is beyond what a computer can do. Therefore, Searle’s appeal to commonsense argument resonates with readers who have an intuitive understanding of what it means to understand. However, when analyzing the arguments more critically, especially after reading Professor Winograd’s piece, to me the fundamental problem with Searle’s argument is an appeal to a widely-shared, human-centric agreement that understanding is linked with consciousness. The ability to be self-aware of a perceived level of understanding seems to form the crux of the argument over understanding. By extension, the fact that we feel that computers are not self-aware excludes them from the class of understanding machines. If self-awareness of understanding is required for true understanding, it seems that the question of whether any human, system, or subsystem inside the Chinese Room understands is unanswerable. We do not know what or who is inside the Chinese Room; it could be a fluent Chinese speaker, Searle’s English speaker, a computer, or a set of water pipes. To find out if the room understands, the only way to find out would be to ask the system. If it has the full richness of expressions that the native Chinese speaker has, it would surely interpret the question “do you understand Chinese” as the native Chinese would, and provide the same answer. Searle would still argue that the system does not understand Chinese, even though it says it does, because it does not have the same semantic knowledge about the symbols passing through as does a native Chinese speaker. I am tempted to agree with Professor Winograd, at this point, that the question “does the Chinese-Room system understand Chinese?” is not meaningful. If we are machines processing perturbations and changing state on account of those perturbations, and if the Chinese Room does the same thing in reaction to the same stimuli (assuming they are all available to the system), I see no way we can properly distinguish understanding. For practical computing purposes, I see no reason to make the distinction. A system that fully simulated a native Chinese speaker’s interactions with his or her environment would have achieved the pinnacle of what the AI field hopes to achieve. Furthermore, Searle seems to be reckless with the concept of understanding. He claims, on page 288, that he can understand stories in English, less so in French and German, and not at all in Chinese. At first, it appears that he is setting up a continuous scale of levels of understanding where complete inability to work with a set of symbols establishes zero understanding. His subsequent exclusion of the system of a photoelectric cell linked to a door opener from the ability to understand seems questionable in this light. It is functionally indistinguishable from a person reliably reacting to a photoelectric cell—the same perturbation produces the same outcome in either case. The specific example of the photoelectric cell seems quite similar to the frog example we discussed in class, where the frog must attempt to eat a small black dot. The question, “does the frog know about flies,” or dots, we found was unknowable. Is the frog, “not in the business” of understanding, as Searle claims the photoelectric cell is not? Searle claims, commonsensically, it is not the same “sense” of understanding, but I find this concept of sense, and his related concept of intentionality never fully substantiated, beyond the fact that in the case of the photoelectric cell, it just doesn’t feel like “good old fashioned understanding.” On the meaningfulness of the question about understanding: it's worth noting that this is exactly how Turing begins the paper in which he proposes (what is now known as) the Turing test.
Thoughts on Understand - Robert GraebertUnfortunately I missed the talk today but some of the participants relayed the gist of the presentation and following conversation. From the account it appears that it is another case where AI is suffering from offering too grandiose statements about its (potential) capabilities. In that context Searle’s disputing that strong AI can exist is worthwhile. He constructs compelling arguments to support his claim. Having read Professor Winograd’s response afterwards though was however eye-opening; Searle uses terminology for his cause carefully and refuses argument in a number of places by declaring them common sense. I can relate much better to the argument in Professor Winograd’s response that words such as understanding need to be seen in the context of the conversation/discourse. Especially for Searle to base his arguments on a very strict definition of understanding (“Nowell and Simon write that the sense of ‘understanding’ they claim for computers is exactly the same as for human beings”) leaves out a lot possible “kinds” of understanding that exist in our world and that we can strive towards in AI. Other mammals, especially domesticated animals, show at least a willingness to learn and understand what their owners mean when calling commands and act accordingly. Humans will also attribute intelligence to the animals although the interaction can be quite limiting. So, do animals have a mind? I think it is again driven by are perception and not a universal truth. An aside from Wikipedia: “In 1997 the concept of animal sentience was written into the basic law of the European Union. The legally-binding Protocol annexed to the Treaty of Amsterdam recognises that animals are ‘sentient beings’, and requires the EU and its Member States to ‘pay full regard to the welfare requirements of animals.’” I think the underlying problem is that we currently do not have an understanding of how the brain and body work to create the mind, awareness and our higher level concepts. Lakoff reasoned how our metaphors stem from our cultural context but also our sensory input and our bodily sensors and neural connection. However, I have not yet seen why that should result in human beings doing so well at reaching understanding and cognition. Is it our embodiment in the world? Is it our specific neuron count and arrangement? Although I am not sure we can achieve or even aim for human-type understanding the way Searle describes in AI, some level of understanding is definitely in the realm of the achievable. For me and Searle, at least how I interpret the text, mimicking behavior or purely operating in the given domain as not sufficient to call that process understanding. I would rather see a test where abstraction and frame/context interpolations are part of the distinction. If an AI can apply concepts learned (or acquired) in one context and apply to another that has not been “programmed” my sense of understanding is reached. In theory such a system should be buildable on top of existing computing architectures. I pre-suppose that we need to reach a deeper understanding on how mind and brain come together though. We might just be the wrong species (with our embodied blindness and lack of complete detachment) to achieve that. Understanding, Orientations, and my roommate - Chris AndersonWhile I was reading “Understanding, Orientations, and Objectivity”, my roommate and a friend of ours got into a discussion of an experiment where monkeys choose between juice and sexual reward. One held that the brain was doing some sort of decision-making computation to figure out which alternative would provide the greatest dopamine reward, while the other said that that was too simplistic. So, in the spirit of inquiry, I wanted to find out what he meant by “computation” and “decision”. It turned into a long discussion of whether his notion of computation stood for anything, and whether the monkey was cognizing at all. It reminded me a lot of the argument against Searle’s notions of objectivity. My roommate eventually conceded that he was perhaps ascribing notions of language (symbols, etc.) to bits of matter that had no such notions. Which lead to my next question – what was Searle’s response to these arguments? The objections in the discourse at the end levy several accusations against Searle that I would like to see him defend – his penchant for property dualism, and the idea that those who are confused simply don’t understand his argument. I looked up the book the article was actually published in, and read Searle’s response (thanks be to Google Reader previews). He seems to skirt around some of the issues, and towards the end comes dangerously close to actually addressing questions of relativism in language. He even discusses relativism in science, saying “The error [in computer science] is to suppose that … success and failure are natural categories. … Nature knows nothing of success and failure. … What we think of as succeeding and failing is relative to our consciousness, relative to our interests.” Of course, he fails to see the parallel to his use of the concept of objective on the previous page: “observer-independent intentionality”, the idea that we can know absolutely if we are happy or tired or cold. There is an entire section in which he claims that “the fact that something is observer-dependent, and hence to that extent ontologically subjective, does not render it epistemically subjective”. This leads to the conclusion that even though we actively interpret the symbols that computers output, we can actually know what they say objectively. It is as if he is presupposing his conclusion – that men and machines are different – and can therefore conclude that we can really know things while a machine cannot. Relativism, we can argue, does not just apply to science and the study of artificial intelligence. It applies to language, and our treatment of other human beings and what we can say we know. I don’t know Searle’s position since 2002, but I wonder if he ever dealt seriously with that criticism. |