Minds, Brains and Programs
What Chinese language can tell us about computer’s intentionality
Artificial intelligence is the science concerned with constructing machines as intelligent as humans. Their behaviour, feelings and intentions should in no way differ from ours. They would be able to understand natural language and have mental capabilities similar to ours. But at which point of similarity can we assume that the machine has artificial intelligence or intentionality? Above all, is it really possible or even thinkable at all to create such an artificial system? Alan Turing proposed an answer to the first question by developing the so-called Turing test. That test adjudges a computer intelligent, if it is not feasible for us to detect in an anonymous chat with a human on the one hand and that computer on the other hand who is human and who is machine. According to the second question John Searle claims that this kind of AI which he coined as “strong” AI is not practicable at present. In the following essay I want to introduce Searle’s theory and his famous Chinese Room Gedankenexperiment to show that computers do not have any intentionality.
Imagine you are sitting in a locked room with books around you which are written in Chinese. The only thing you can understand is an English manual that helps you to do your required task. From outside the room you get certain sheets of paper with Chinese symbols and your job is it to give a response sheet back which is also written in Chinese. You can think of something like a library where people want to know special things like the birthday of Columbus. The library (or a librarian) looks up the topics in its books and answers the people’s question. Since you are not able to speak Chinese you open your manual that provides you with the correct instructions how to deal with incoming symbols, e.g. in which book the librarian has to look for and how the answer should be given, and in the end you write down the symbols although you do not know what they mean. Eventually, from outside the room it seems to be that there is a librarian who understands Chinese.
Maybe you have already recognized that the incident in this story is comparable with a computer. It gets certain input symbols (e.g. a natural language) and transforms them into a defined output (signs on your monitor). During that process your computer only knows how to go from one state into another (how to react). It does not care what for example an “a” means that you have typed in.
By drawing this analogy Searle comes to the conclusion that though the librarian seems to behave intelligent he has no understanding of the things he deals with, so they have no semantic content for him. But at the level of syntax he knows very well the rules what to do when. The computer behaves similarly: its program is completely defined through a formal structure but there is no semantic understanding.
What he wants to say with this is that semantics does not arise from pure syntax. A fistful of instructions how to handle a language does not automatically lead to an understanding of that language. Therefore although the system seems to behave intelligent, because it fulfils our expectations of what it has to do and appears to talk in natural language and may thereby assumed to have intentionality, it always remains a formal evaluation of symbols and this process cannot produce thoughts and thus intentionality.
Now we come to the point where it is important for Searle to make a difference between “strong” and “weak” AI. According to Searle the former which I have already mentioned in the introduction is not available for any computer at present. They can only mimic our behaviour or intelligence. So they simulate our mental faculties and that is what he calls “weak” AI. This claim relies on the distinction he draws between a computer’s program and our brain. The latter has certain causal properties that enable us to come to a real understanding of the things around us, whereas the computer has no possibility with its formal construction to have a semantic reference to the real world. As a result a computer cannot have intentionality.
To sum up briefly, no artificial system like a computer has intentional states only because of the ability to manipulate formal inputs like symbols with the help of a formal program or algorithm. They are in need of other prerequisites to become intentional. As another result of their capability to mimic our intelligence Searle affirms that they are able to pass the Turing test but that thus the Turing test seems not to be the adequate test for artificial intelligence and least of all for intentionality. Although the computer passes the test it is not sufficient to presume human comprehension in this system.
Naturally, there are a lot of objections against Searle’s thesis and especially his Chinese room. I am now going to explain briefly a few of them and give Searle’s response in order to show that they fail to demonstrate that computers can still have intentionality.
We can subdivide the objections into three main approaches. The first one claims that the whole or a different system might be able to understand Chinese. That is the opinion of “The System Reply” and “The Virtual Mind Reply”. For example, we must not only see the man inside the room who does not understand Chinese but all the involved components. In comparison with a computer the man can be identified with the CPU which for itself does not represent the whole capability of the computer.
Searle argues that there is nothing in the room which the man does not include. He is the one that could memorize all the rules from the English manual, has all the important data he needs and makes calculations in his head. You can just as well think of a situation where he then works somewhere outside in the nature and does still answer Chinese questions without problems. Thus if the man in the Chinese room does not understand Chinese the entire room does not either. Therefore the man should be identified with the computer and not only with the CPU so that you can once again say that the computer has no intentionality.
The second approach deals with certain variations of the system which should allow the computer to understand Chinese. There is “The Robot Reply” which assumes that the system will be able to understand if one simply puts the computer in the head of a robot that is able to freely walk along and to perceive human sensory inputs like vision through a camera. The robot can interact with its environment like we do and should thereby be capable to attach meanings to symbols and understand natural language. Searle replies that all these sensory information are only additional syntactic inputs. Therefore the robot would not gain any new feature that neither enables it to get understanding nor intentionality.
Another response in this category is “The Brain Simulator Reply”. It says that we only need a system that simulates in detail the operation of a human brain. If this computer does not understand Chinese why should we do? Well, it neglects the essential thing while simulating only formally our neurons and their connections namely the ability of producing intentional states. Intentionality is nothing that we can reach solely with the help of formal structures. Therefore you have to distinguish between a pure simulation that is not sufficient for intentionality and a real duplication, for instance a clone of you. For a better understanding you can think about a simple experiment in a biology laboratory where you only simulate the brain. No-one would ever expect to create a mind (or consciousness and understanding) in this experiment.
The third approach gets somehow a little bit side-tracked. “The Other-Minds Reply” draws upon the assumption that the man in the Chinese room might understand Chinese regardless of Searle’s denial. The question is how you can know that others understand something. It is rather simple to explain: We attribute understanding to other humans on the basis of their behaviour we have encountered. As a conclusion it is as well reasonable to attribute understanding to the man in the Chinese room because he behaves as if understanding Chinese. That should be the conclusion at which Searle has to come if he presupposes that we can attribute intentionality to humans.
For Searle the question of how we know that other people have cognitive capabilities is quite another philosophical topic. That is not the issue to discuss because humans and computers are two different things. In his eyes humans have some causal properties in their brain that computers do not have and cannot have through their programs and algorithms. So this reply does not show if we can ascribe intentionality to a computer.
As you can see there are a lot of objections but Searle can argue against all of them. What I find the most convincing fact is that Searle does not deny the possibility that computers one day have intentionality or understand natural language. Merely at present with the opinion of strong AI that intentionality and understanding are computational processes that can be derived from formal structures and defined symbols the goal of intentionality cannot be reached. Formal properties are not sufficient for that.
