A computer is a device that can store and process data. Data can be entered into a computer using input devices like a keyboard and the results of processing can be sent to output devices such as a screen or printer.
Data is nothing other than numbers, letters and symbols. When we are dealing with computers all data is composed of binary values, so computer data is made of lots of collections of lots of ones and zeroes.
Computers do not process information or store information, only data.
Data can become information when we assign meaning to it and this process can only * take place inside our brains. When you look at a meme of a humorous cat on your phone screen, your phone’s processor is sending binary values to specific addresses that activate particular pixels. There is no concept within the computer of a “funny cat”.
You might be thinking about contesting this point, there are numerous chat-bots that that make use of “large language models” (LLM). You could upload your meme image into one of these chat bots, ask it what it is a picture of and it might appear to identify the picture, giving the impression that it has taken some data and assigned meaning to it.
Even these LLM chat bots do not “understand” information. In a LLM, every word in the English language has been stored as a binary pattern. An algorithm has been used to iterate through every word in every source of text possible, every available book, every web page. This algorithm has been designed to try to create a probability score between every word and every other word that exists. For each word, what is the probability that some other word would be the next best word to use in the response. This probability is also stored as a binary value.
When you type a prompt into a LLM chat bot, the text that you enter is used to determine the most probable best first word to use in a sentence to respond to your prompt. This process repeats until it hasa written response that appears to be the result of some sort of intelligence with awareness of the meaning of information (a similar method has been applied to collections of pixels, binary image data, which appears to give these chat bots the ability to interpret and create image information).
The chat bot has not applied meaning to any data, it has simply processed binary values that represent words and probability scores. It becomes quite apparent that there is no understanding of information in the real sense when we see how frequently these chat bots produce incorrect results(often referred to as “hallucinations”).
Colossal amounts time and money are currently being spent on developing these LLMs, at some point the model may be improved to such an extent that these hallucinations occur very infrequently. At this point, can we argue that the LLM system has a concept of “information”?
The Turing Test
Alan Turing famously proposed that a computer could be said to be intelligent, if a human who was communicating with it, was unable to tell if they were communicating with a computer or another human.
Imagine a person in a room with a keyboard and screen. They type messages into the keyboard and see responses on the screen.
In another room is another person who reads what the first has typed and responds accordingly and in a third room a computer is doing the responding. Is the person able to tell the difference? If they can then that computer has failed the test.
It’s conceivable that an LLM could fool a person into thinking they are communicating with another human and thus pass the Turing test which means it is intelligent and also implies that the computer understands information (rather than just processing data).
The Chinese Room Experiment
The philosopher, John Searle proposed a counter argument to the Turing Test. In this argument he suggests that a native speaker of a language (in his example, this language is Chinese) writes messages and sends the messages to another room.
In the other room is a person with no knowledge of that language. What they have is a very comprehensive set of instructions. These instructions tell them what to write in response if they receive a particular set of characters. Using these comprehensive instructions, the responder is able to post back very convincing responses to the original message sender.
If you were to ask the responder what they had been writing about in their responses, they would have no clue whatsoever. All they did was to do what the instructions told them to, therefore just following rules to produce intelligent sounding results does not imply actual intelligence.
*Currently, brains are the only things that can understand “meaning” and thus, information. Work is underway to create something called “General AI”, while it is not clearly defined what this will be, the consensus is that if and when it is finally achieved, this will have “true” intelligence (in some way) and as such will have the ability to apply meaning and use information.