Friday, July 20, 2007

Emulation versus Simulation (Muddled follow-on)

Thanks to Stu for starting an interesting topic about artificial life (Loose Words-Muddled Thinking). I'm starting a fresh thread, because comments are difficult to find and keep track of on this blog. Also, in this blog, "comments" don't appear in the daily email.

Stu said:


However, the phrase that pushed my hot button was " a researcher proposes to recreate a living organism inside a computer." And while the following sentence mitigates that image somewhat, the actual article in the Economist further perpetuates the myth that it is possible to create life within the computer itself:

"David Harel of the Weizmann Institute in Israel ... proposes to recreate living organisms inside a computer."

And so it seems to me that even intellectually respectable organizations such as the Economist and Sigma Xi can fall prey to muddy thinking. A computer simulation of a process is not the process itself just as the pictures in Playboy are not really flesh and blood human beings. As John Searle has pointed out in his essay, "Is the Brain's Mind a Computer Program?" (short answer: NO!!!






Joel responds:

I'll take the opposite position and say that the brain's mind IS a computer program. To understand what I mean, we first need to refine the difference between a simulation and an emulation. A SIMULATION is an mathematical approximation of a real physical system which receives inputs and sends outputa. The accuracy of the outputs is based upon how good a model we are using. An EMULATION is a perfect substitute for a real physical system. In computer science, an emulation might be computer software which substitutes for a proposed piece of computer hardware so that it can be tested before it's actually manufactured. An example of a simulation might be a model of Earth's energy balance which can predict global warming and cooling trends. The simulation is clearly not the planet itself any more than a FAX of a pizza is a pizza. However, an emulation of a pizza would be the same as a pizza in all important regards. Clearly, edibility is an important aspect.

Suppose I produce an imitation of a pizza which looks like a pizza, smells like a pizza and tastes like a pizza. In this application it's made out colored polystyrene and seasonings, so I can sell it as a zero calorie pizza substitute. In effect, let's say that it's an emulation that passes the Turing Test. Many would say that it's not a pizza, because in their definition, pizza is a kind of flat bread made out of flour, water, salt, tomatoes and cheese. These people can never be satisfied, because they are looking at the micro level not the macro level. Even if my emulation satisfies every MACRO property of the system, by definition it cannot satisfy the MICRO properties without becoming the system itself.

The same is true of thought experiments like The Chinese Room. Even though the room emulates a chinese speaker in every external or macro regard, there will be those who must look inside and insist that the room contain a "chinese knowing mind." Since it does not, it passes the Turing Test, but fails to contain the expected intelligence. As at least one philosopher has called this kind of reasoning "argument by misdirection." With respect -Joel

3 comments:

Ira Glickstein said...

Joel:

Thanks for providing your views on Stu's posting.

For those readers who may not be familiar with American philosopher John Searle's "Chinese Room Thought Experiment" let me summarize it briefly:

1) An intelligent American who does not understand Chinese at all is hired to work in a small building called the "Chinese Room." The building is located in a Chinese city or the Chinatown of a US city.

2) Native Chinese speakers come by and insert written Chinese questions into an Input slot in the Chinese Room. For example: "Where can I get the best egg rolls in Peking?"

3) The person in the Chinese Room takes the slip of paper and looks at the first (Chinese) character. It is all "squiggles" and he or she does not understand it at all. He or she goes to a set of books and laboriously scans through them until he or she finds the "squiggles" that match those in that character. The book has English instructions on what to do next.

4) Without understanding anything about Chinese or the characters on the slip of paper, the person follows the English instructions, which tell him or her to go through each Chinese character in turn. Eventually, the English instructions tell him or her to write a particular set of "squiggles" on a slip of paper and push it through the Output slot of the Chinese Room. (The person in the Chinese Room doesn't know it, but the message he or she has written says "The best egg rolls in Peking are at Fat Wongs at 123 Mao Tse Tung Street.")

4) The native Chinese speaker comes by to retrieve the written Chinese answer and, sure enough, he or she concludes that the Chinese answer is not only grammatically correct, but also quite responsive and valid. He or she goes to Fat Wongs and orders egg rolls and they are the best!

In Searle's thought experiment the American person and the English instruction books in the room are analogs of a programmed computer that responds to written Chinese questions with valid Chinese answers. (Such programs exist today in English and other languages and are called "chatbots". Today's versions have acceptable performance in some restricted domains of knowledge and I believe they will improve at some time in the future to the point they will be indistinguishable from humans and will therefore pass the "Turing Test" Joel referred to.

Of course, a real person in the "Chinese Room" would die of boredom before he or she could replicate what a programmed computer could do, making Searle's thought experiment impractical.

However, that is not the point!

Searle asks us to assume it could work and could fool native Chinese speakers into thinking the Chinese Room really understands Chinese (and also knows alot about restaurants and so on).

Then, he asks WHERE IS THE UNDERSTANDING?

In Joel's emulation vs simulation dichotomy, where does the Chinese Room (or a chatbot) stand?

When I first heard about Searle's thought experiment, I concluded that the Chinese Room (taking the person plus the English instruction books together) REALLY *understood* Chinese! After all, *you* understand English (else you would not have read to this point) but your eyes and your skin and your muscles and other cells do not! In fact, the very neurons in your brain that are interpreting this sentence, have no idea they are working together to *understand English*!

My initial reaction is called the "System View" in the very well-developed academic literature surrounding Searle's idea. The individual parts of the system do not understand Chinese, but, taken together, the SYSTEM does!

So, I concluded the Chinese Room (or a chatbot), in Joel's dichotomy, EMULATES a thinking Chinese person, at least in the limited realm of understanding Chinese related to some domain of knowledge.

However, after several years (and with the gentle influence of Howard Pattee :^), I have come to the conclusion it is no such thing!

The Chinese Room SIMULATES a Chinese person. All the understanding of Chinese (and restaurants, etc.) is in the brains of the persons who programmed the English instruction books in the Chinese Room and the native Chinese speakers who push slips of written Chinese into the slot and retrieve Chinese answers out of the slot!

A clearer example I like is the difference between a cubic zirconium and a synthetic diamond. A synthetic diamond is made by subjecting carbon to great pressures until it assumes the crystaline structure of a natural diamond. A cubic zirconium is made of totally different material. While a cubic zirconium may have wonderful reflective properties and be indistinguishable when worn as jewelery, it is not a diamond at all!

Similarly, the Chinese Room (or a chatbot) may have wonderful "intelligent" properties in some domains of human knowledge, but, even if it was extended to many domains of knowledge and was indistinguishable from a living human, it still would not *understand*.

I used to call persons who accept the conclusion I now put forward "protoplasm bigots" because, like "racists" they judge a person, paraphrasing Martin Luther King, not by the content of his or her character, but by the color of his or her skin.

The problem, in my mind, is not the material out of which a programmed computer is made -- I (and Searle) accept that it is possible a machine made of silicon and copper could be really *intelligent* and really *understand* and so on.

However, the current generation of programmed computers do not meet that standard. ALL the "intelligence" and "understanding" and so on they exhibit (and much of it is quite impressive today and will be much more so in the future) is in the protoplasm brains of their human programmers and users.

Ira Glickstein

Claire said...

From Stu: Here's my take on the Chinese Room scenario: I believe the total system exhibits intelligence (as Ira used to) but my definition of the "total system" includes not only the programmer's intelligence but everything else in the universe.

I believe everything is connected in much the way described by the Diamond Sutra wherein every object in the universe can be thought of as an infinitely-faceted diamond which reflects every other diamond.

If in fact intelligence is distributed throughout the universe and not localized in special places called human beings, it makes things much simpler but possibly not as interesting; it can even be viewed as a copout...

Ira Glickstein said...

The emulation vs simulation Topic posted by Joel long ago (July 2007) and based on an earlier Topic by Stu continues to get hits according to our "Live Traffic Feed" in the right hand column of the Blog.

Those hits caused me to reread the posting. I noticed that one of Joel's points was not properly responded to. Joel wrote:

"Suppose I produce an imitation of a pizza which looks like a pizza, smells like a pizza and tastes like a pizza. In this application it's made out colored polystyrene and seasonings, so I can sell it as a zero calorie pizza substitute. In effect, let's say that it's an emulation that passes the Turing Test. Many would say that it's not a pizza, because in their definition, pizza is a kind of flat bread made out of flour, water, salt, tomatoes and cheese. These people can never be satisfied, because they are looking at the micro level not the macro level. Even if my emulation satisfies every MACRO property of the system, by definition it cannot satisfy the MICRO properties without becoming the system itself."

Like Searle's "Chinese Room", Joel's Plastic Pizza fools (or less perjoratively satisfies) everyone at the MACRO (or "system") level. However, Joel's example has an advantage over Searle's in that Joel's is a PHYSICAL thing you can touch, feel, smell, taste, eat, and digest. Indeed, if it was fortified with vitamins and minerals and proteins it could even supply the same nutrition as a normal pizza! (Searle's Chinese Room only emulates the MENTAL aspect of an intelligent human being.)

After thinking about it for a while, I agree with Joel that it is an emulation of pizza, not a mere simulation. I would allow Joel's Plastic Pizza to be marketed as a "pizza snack" so long as the label added "artificial color, flavor, and everything else". It is in the same category as some diet soft drinks that contain only artificial ingredients. They are real "soft drinks" with "artificial lemon flavor" but not real "Lemonade". (While writing this I checked a package of "Crystal Lite" that says "Natural Lemonade flavor" on the container. The ingredients are all artificial but it does contain "less than 2% of natural flavor. lemon juice solids, ..." I believe other brands of diet soft drinks do not contain any natural matreial of the fruit flavors they emulate.)

So, just as Cubic Ziconium is real "jewelery", but not natural diamond, Joel's Plastic Pizza is a real "pizza snack" but not a natural pizza.

Ira Glickstein