Header 1

Our future, our universe, and other weighty topics


Monday, January 18, 2016

Why You Won't See an Age of Spiritual Machines

According to the computationalism theory of the human mind, the mind is like a computer, and one day we will be able to develop computers that produce outputs just like human consciousness. Such a theory is assumed by most proponents of the Singularity, the idea that there will before long be an “intelligence explosion” which results in machines with intelligence far beyond our own. Such proponents write books with titles such as The Age of Spiritual Machines. Such theorists usually don't tell us that they are advancing the computationalism theory of the human mind. They usually just pronounce the dubious ideas of such a theory as if such ideas were self-evident.

But the computationalism theory of the human mind is not valid. To explain why, let us look into what happens when computers compute. The following equation covers most of the types of computation that occur.

digital inputs + processing = digital outputs or modification of digital data

There are various types of variations of this equation. One is simply:

no inputs + processing = digital outputs or modification of digital data

Another variation is the following:

digital inputs + processing + retrieval of other digital inputs = digital outputs or modification of digital data

By digital inputs or digital outputs I mean anything at all that can be represented digitally, by a sequence of binary numbers. Here are some of the things that we know can be represented digitally, and which modern computers do use as digital inputs or digital outputs:

Any number
Any set of characters or words
Images
Videos
Databases



To the computer, that pretty gal is just a series of 1's and 0's

Any text can be digitally represented by means of things such as the ASCII system that allows you to represent particular characters as particular numbers. While we don't normally think of an image as digital, it can be represented digitally as a series of pixels or picture elements. For example, a photograph might consist of 1 million pixels, which each can be represented by a number representing a particular shade of color. So the image can be digitally represented by a million such numbers. A video can also be representing digitally, since the video can be represented as a series of images, each of which can be digitally represented.

But there are two things that we can never hope to produce as digital outputs. The first is real conceptual understanding, and the second is experience. By understanding I don't mean “how-to” type understanding, but the high-level conscious understanding of some abstract truth or concept. By experience I mean an actual human experience, such as the life-flow you experience during an hour of your life.

We can imagine no possible way to produce a digital output that would equal a real conceptual understanding of something. Nor can we imagine any possible way to produce a digital output that would equal something like a human experience.

Imagine a conversation like this 200 years from now between a programming supervisor and a programmer who has been doing his job for over a century (thanks to the marvel of life-extension pills).

Boss: Well, I've got an interesting new assignment for you. I want you to compute an interesting new output.
Programmer: This should be a breeze. I've already done functions that compute 12,000 different text outputs, 15,000 different numerical outputs, 25,000 different image outputs, and 4000 different video outputs.
Boss: This time I want the computer to produce waterfalls and Swiss cheese. Not just pictures, but the real things.
Programmer: Are you crazy?

There might be a similar conversation if the boss asked the programmer to produce understanding or experience as the outputs. Just as waterfalls and Swiss cheese are not digital outputs, real conceptual understanding and experience (a slice of life-flow) are not digital outputs. You can make outputs that might mimic some understanding someone might have, but you cannot produce real conceptual understanding as a digital output. You can make outputs that might mimic some sight someone might see while having an experience, but you cannot produce actual experience (a slice of life-flow) as a digital output. 

But, you may ask, doesn't that smart computer Watson already understand something – the game of chess? No, it doesn't. Watson merely can produce a digital output corresponding to a good move to make as the next move in a chess game. Watson has zero conceptual understanding of the game of chess itself, and has zero understanding of the abstract concept of a game. The only way you can understand the abstract concept of a game (or the abstract concept of leisure) is if you have been a human being (or something like a human), and played a game yourself. 

A digital output must always boil down to a series of 1's and 0's. Can we imagine a series of 1's and 0's that would equal a real understanding of an abstract concept such as health, matter, life or world peace? No, we cannot. 

Because conceptual understanding is not a digital output, we should not think that predictions such as the following recent prediction are correct:

The third thing you can expect before the year 2100 is the development of generalized artificial intelligence (GAI). In other words, machines that don't just play games like chess or Jeopardy, but can do the thinking required for any white-collar job, including all the ones at the top.

Such a prediction is based on the idea that future computers will be able to produce conceptual understanding as an output. They won't, because real understanding of abstract concepts is not a possible digital output, and digital computers will only be able to produce digital outputs. 

PostscriptDigital computers translate inputs into digital content, and translate outputs into content that may not seem digital. For example, a compiler translates English-like computer code into digital inputs a computer can understand, while a purely digital output may be translated into something that doesn't look digital. But at the lowest level inside the computer, it's all digital.