• Quick note - the problem with Youtube videos not embedding on the forum appears to have been fixed, thanks to ZiprHead. If you do still see problems let me know.

Detecting intelligence via algorithm

JamesM said:
Wowbagger, what is the basis for you thinking that it would work on a random string of characters as well as Emma? If the algorithm is compression-based, I would be very surprised if it worked as well on, for example, the characters of Emma randomly permuted.
Click to expand...
I admit I do not know very much about the algorithm, which is why my bet will only be a small amount of money.

You are also partly correct about the "compression-based" concern: it would deliver more results from organized text rather than random strings of characters. However, what few patterns happen to occur in the random characters may still get picked up. This means that the algorithm "found meaning" in otherwise obviously random characters.

In other words, The claim that this algorithm provides a "hint" of design is invalid, if can also find such "hints" in utter randomness.

Someone may then try claim that the quality of results is also better than that from random gibberish, and that the results from randomness can be explained away under GIGO (garbage in, garbage out).
However, this is now getting into subjective territory. The computer just operates the algorithm. If it finds patterns in random stuff, than you must dismiss the idea that the algorithm can determine design. Simple as that.


ETA: This posts assumes it is compression based. If it is not compression based, then the chances are even better that it could find patterns in randomness.
 
Last edited:
Wowbagger said:
If it finds patterns in random stuff, than you must dismiss the idea that the algorithm can determine design. Simple as that.
Click to expand...
I don't agree here. The quality of the results could be assessed in a objective manner. I would also be surprised if the algorithm found no patterns in random text. But if a suitable null hypothesis was constructed, you could test for significance, by a permutation test if necessary. Finding patterns in random stuff isn't a showstopper.

Applications of information theory and string based algorithms are ten a penny in bioinformatics. Compression-based algorithms have definitely been used, I've used them myself (in a related field).
 
Iamme said:
Well...we know that hieroglyphics were also structured and that took scientists years to unravel and...didn't that take the Rosetta Stone to do it? Modern computers are now our Rosetta Stone...as was the case with The Bible Code.
Click to expand...

The difference here is that they're solving a much easier problem with computers than cracking an unknown language. They're just looking for patterns and not what the patterns mean which can only be solved with contextual information. It's a case of getting the computer to hum along even if it doesn't know the words.
 
T'ai Chi said:
You seem to be hung up on the implications of the word 'design' in a major way. The correct way to state it is that structure does not necessarily imply design. In this case however, we already know for an absolute fact that the book which was written that was analzed was indeed designed.
Click to expand...

You might wish to study logic. You might learn that the statement "A does not imply B" does not imply the statement "A implies not B." It is a short step from that realization to realizing that the word "necessarily" there has no logical import. (It has rhetorical weight, but rhetorical only.)

T'ai Chi said:
If the algorithm works well on things that are known to be designed, then if it operates on object X and performs similarly it can hint that design is present.
Click to expand...

Just like white tables are evidence that all crows are black.

If you studied some logic you might get the reference.

Cheers,
Ben
 
JamesM said:
I don't agree here. The quality of the results could be assessed in a objective manner. I would also be surprised if the algorithm found no patterns in random text. But if a suitable null hypothesis was constructed, you could test for significance, by a permutation test if necessary. Finding patterns in random stuff isn't a showstopper.
Click to expand...

You could be right. But, the tests for significance would have to be based on some objective measure:
* For English text, the tests would be based on rules of human grammar and stuff.

* For DNA, the tests would probably have to be done in the lab, more than in the computer: Experiments would have to be set up to see if these divisions are useful for expressing traits in embryology, etc.

* Where would objective tests for "implying intelligence" (as a form of significance) come from? That would be a whole separate algorithm than the one for chunking up strings of data.

After working for decades to find such a definitive, objective test for intelligence, in general, ID has never come close to finding one. Not even SETI has such a test.

A few more points to consider and to reiterate:

The intention of this algorithm was not to demonstrate ID in DNA design. Its intention was merely to find patterns in the DNA, to see if it can be chunked up better. The title of this thread is, indeed, misleading. As is the title "DNA as the Repository of Intelligence" on the Dembski blog. (And, anyone on Dembski's blog who thinks otherwise is grasping at straws.)

Given its original intention, they might not have bothered testing it with random characters, because it would be a meaningless exercise in GIGO. (Of course, hypothetically, if the intention was really to infer design, then, obviously, testing with samples of randomness would very much matter.)

It was only 80% successful in the English text, where the correct positions of spaces was known by the operators. So, even if it were to be utilized as a test for intelligence, right off the bat, it would not be so perfect at it.

DNA is probably less organized than English text, but each of its "characters" is only 2-bits, which makes pattern-finding easier.
However, it will take more testing before we find out if these divisions it finds in DNA are useful or not. Since we can not determine if the divisions it finds in DNA are useful, right away, that makes it a poor choice for inferring ID.

My guess is that it will be less than 100% successful at finding useful divisions in DNA, anyway. Rendering it more useless for judging ID.

Random characters could be any number of bits each (usually 8 or 16). This, initially makes it seem like pattern finding is hopeless. However...

My guess, and it is only a small-money bet, is that if they did run the algorithm on random strings of characters, it will succeed in chunking them. The numbers of chunks may be fewer than an English book or a strand of DNA. But, if the algorithm can still do it, then the claim that it, alone, can be used to infer design is ludicrous. Also, the idea that its output is formed in such a way as to aid in testing for implied intelligence becomes ludicrous.

(So, perhaps my statement "If it finds patterns in random stuff, then you must dismiss the idea that the algorithm can determine design." was oversimplified.)

It also is worth reiterating that this program is NOT a useless tool, given its original intention. I am quite sure computers can aid in determining the chunking of DNA strands quite nicely. However, it would be useless to think such a tool could also be used to imply hints of a designer.

And on another subject mentioned in this thread:
No one has really found any statistically significant codes in the Bible Code. The math used to fool those who claimed otherwise was always very shaky. You can read about that, here: http://skepdic.com/bibcode.html
Or look up stuff written about it by David E. Thomas.
 
T'ai Chi said:
If the algorithm works well on things that are known to be designed, then if it operates on object X and performs similarly it can hint that design is present.
Click to expand...
Well, not really. You have to demonstrate that the algorithm doesn't find structure in a random jumble of letters. And also if it does find the words in a random jumble of words, what does that do to your theory?

This is why we are skeptical. We have only heard about this extraction of the book Emma. We have not heard about repeated trials with other books (or with other languages). We have not heard about controls with random lists of words and random collections of letters. Almost anything can be implied from a single positive result, without knowledge of what controls were used to exclude all the other possible explanations. That is what the scientific method is.
 
T'ai Chi said:
No one said that the work as presented was complete. Why do you pretend someone did?
Click to expand...
I think you are still mistaking the intent of the algorithm: It is not intented to perform anything close to a measurement of intelligent design. All it does is chunk stuff up.

I will bet* that even the completed work (if there ever is one) would still work well finding chunks in random characters. Therefore, it could never be used to "hint" at design.

(* a small amount of money.)
 
I work in statistical data processing, so maybe I can clear something up here.

This is quite clearly an algorithm for detecting patterns and structure in data. The specifics of what types of patterns it looks for, and how it does it, aren't even important to the "ID" question being raised here.

Any pattern finding algorithm will, when applied to enough randomly generated data, find what it is looking for in that data. The interesting question (in this context) is whether or not it found it more often than would be expected if some null-hypothesis were true.

The thing is, the hypothesis that "the data is not the result of intelligent design", is not a valid null-hypothesis. It is not that there is anything special about ID which invalidates it as a null-hypothesis. It is simply that it is too general.

Null-hypothesis testing can be effectively used to reject specific claims about data. For example, in my own work I am often applying statistical tests to neuronal measurements to study correlations between measurements at different locations. We are interested in finding a very specific type of correlation called "phase-locking", but the hypothesis "These correlations are not due to phase-locking" is not a valid null-hypothesis. Instead, we have to use other scientifically gathered information to narrow it down to a small set of possible mechanisms, and then use hypothesis testing to rule out all but one. Needless to say, this is not easy.


Dr. Stupid
 
You must log in or register to reply here.

Back
Top Bottom