This is the claim: the concept of the symbol is one of the biggest problems AI has. So what does it mean to abandon the symbol? First, what is a symbol?
A linguistic symbol (A, B, CAT, dog) is a shape that has a meaning. This is the conception Searle uses in his famous Chinese room argument. The input to the room (computer) comprises cards inscribed with Chinese ideograms.
Almost universally, computers are said to process symbols, meaning linguistic symbols. But let’s be clear: they don’t. Stored program electronic digital computers do not, and physically cannot by virtue of their design, process linguistic symbols.
Since talking about computers as if they do process symbols causes such trouble when trying to understand the semantics (or lack of semantics) of a computer, it would be much better to stop saying that computers process symbols. At least if we do talk about computers processing symbols (because it’s a lot easier than being accurate) we should also at the same time explain that the idea that computers process symbols is a myth.
Computers don’t process symbols
The myth says that computers process, or manipulate, symbols of language. This is how one expression of the myth goes. Keyboards are rectangles containing plastic keys inscribed with shapes (symbols). These shapes, and combinations of them, have meanings. When a key of a keyboard is pressed, the thing printed on the key cap – a symbol – leaps off the plastic cap and careens down the wire coming out the back of the keyboard, then it races into the interior of the electronic computer. Now inside the machine, it is manipulated and/or stored.
This is drivel.
In response, a computer scientist might say that of course this does not happen and only an idiot would think that it does. However, the scientist might continue, computers are based on bits, and even though minimalist compared to letters of an alphabet, bits are still symbols, namely 0 and 1.
This is also drivel.
Naming the things computers process
The term “bit” is inappropriate. “Bit” is short for binary digit, and digits are the numerals, 0-9. And numerals are symbols that have meanings. The reason why it is inappropriate to call what computers process “bits” is that what electronic computers process have no meanings, and never could have meanings.
What happens is that today’s electronic computers process tokenised binary values (which for lack of imagination I’d like to call TBVs). A token is an instance of a value, it’s a particular object that has a value of a relevant property.
Why don’t these TBVs have meanings?
Why can’t an electronic computer’s TBVs have a meaning? These are the reasons:
(1) Humans can’t assign meanings to TBVs. This is because they can’t sense TBVs. Humans lack the sense organs that would enable them to perceive TBVs. Humans do have the sense organs needed to sense values of shape (eyes, touch), to sense values of the property of shape. But they lack the sensory apparatus to detect the values of the properties of the things computers process.
Humans can’t detect semiconductor switch states, magnetic domain orientation or clocked voltage level. Humans can’t detect with their senses what computers process. If humans can’t detect with their senses these things, they can’t know the meanings of these things, or assign meanings to them. Humans can assign meanings to shapes, and they do, and that’s written language. And humans can perceive and understand the shapes. But they can’t assign meanings to TBVs.
Now the astute computer scientist might say that this claim is obviously false, since meanings are assigned to TBVs, namely the meanings of 0 and 1. And sure, the computer scientist might continue, humans can’t sense magnetic orientations etc., but they have given these orientations meanings, namely 0 and 1.
Well, what is wrong with this response? This is what is wrong: Certainly, something has been given a meaning, but this recipient of meaning is not the magnetic states. The magnetic states don’t mean 0 and 1. “0” and “1” mean the magnetic states. The shapes, the sensible symbols “0” and “1” have been given a new meaning, namely the magnetic domains.
(2) So to humans, the things computers process have no meanings, but what about to the machine itself? This, after all, is the key issue of the Chinese room argument. An answer can be clarified by reference to the human brain. Can the human sense the neural pulses pulsing around inside its brain? Does the human, via its sensory apparatus, understand the meanings of these pulses? No? So why are we asking whether the alleged human-like computer brain can, by virtue of the operations of its sensory apparatus, know the meanings of its inner magnetic states?