Re: Pi Forums

This is to about 6 people. Many thanks for your efforts on my behalf and your replies. I'm feeling rather weary as I haven't had my after lunch nap and will be getting ready for bed after this. I'll read your replies more carefully tomorrow and hope to get the RPi connected up again. It's not easy as SWMBO gets very concerned when she sees me doing anything that requires mental or physical effort, so I try to do it very early in the morning. (I had a slight stroke recently.)

I'm a bit of a nit-picker, so I like John's comment about the spelling of "complement".

Frank

I can never aspire to be as good a writer as Mark Twain, Thomas Jefferson, W. C. Fields or any number of worthy gentlefolk who despised spellling. Although the actual provenance of the genre of quotes that contain the meme "I have nothing but contempt for anyone who can spell a word only one way" is unknown, I take it as a truism. At best it demonstrates a lack of imagination.  I do believe that it is a skill [?] which will go the way of the buggywhip industry as spell checkers continue to encompass an ever increasing share of the written word.

something else that as an engineer can be quite interesting (because of the logical reasoning)

I have a different reaction reading judicial decisions like this one.  In mathematics, if the reasoning

is logical, then you ought to have unanimous agreement.  In law, you can apply the same law to the

same undisputed facts, and arrive at a majority decision and a contradictory minority decision.

Even within the majority or minority, they may arrive at the decision for different and inconsistent reasons.

I see here a judge making a somewhat arbitrary decision on whether to follow previous precedent that

terms printed on a ticket stub are binding even if nobody reads them, or whether to distinguish the

current case on the basis that it involves an automated ticket machine rather than a human ticket agent.

Even though the decision is explained in logical terms, it could have just as easily gone the other way

and also have been explained in logical terms.  So there is something outside of logic that is actually

swaying the decision. 

coder27 wrote:

 

I see here a judge making a somewhat arbitrary decision on whether to follow previous precedent that terms printed on a ticket stub are binding even if nobody reads them, or whether to distinguish the current case on the basis that it involves an automated ticket machine rather than a human ticket agent.

 

Even though the decision is explained in logical terms, it could just have easily gone the other way and also have been explained in logical terms.  So there is something outside of logic that is actually swaying the decision. 

Well observed!

I think it's quite easy to see what's wrong with that situation:  the premises are unsafe, and might easily be flawed.  Using impecable logic one can arrive at nonsense conclusions by arguing from false premises.  I suspect that this applies also in the non-boolean space of value judgements.

Lord Denning might well have argued differently simply by choosing a different set of premises from case law as precedent.  Law is really just opinion even when argued logically, and rests on cherry-picked foundations.

Engineering has a solution for the above kind of difficult problem:  fuzzy logic.  It doesn't attempt to derive logical conclusions from incomplete information or questionable premises, but instead seeks to encode practical utility in human-digestible chunks, and assign degrees of importance to them.

When a problem domain is non-analytic or is simply too hard to analyse in reasonable time, fuzzy logic applies multiple overlapping weighted quality functions to the data, including heuristics, all aimed at evaluating inputs from a range of different perspectives to apply "commonsense" to a problem in piecewise-linear fashion.  This typically directs process control towards reasonably effective operation (often not far from a practical optimum), and the approach is less susceptible to problems of local maxima that are common in analytic solutions because individual functions can contribute non-local figures of merit.  It  tends to be safe and robust as well, because safety constraints are easily encoded in yet more piecewise functions.

Perhaps in an enlightened and more logical future (one in which humans are still relevant, not the Skynet one), judges will be replaced by open source fuzzy logic automata, controlled by millions of simple piecewise merit functions contributed by interested citizenry?

@ Morgaine,

Won't your fuzzy judges end up like real ones, some good, some bad and most indifferent (perhaps depending on which distro they are based on) ?

My friends who are well versed in classical and modern control theory (a very small subset of the group) don't like fuzzy logic because there is no reasonable mathermatical process to prove stability of a fuzzy controller.

I've always had a soft spot for it myself and still pay the maintenance on my MATLAB Fuzzy Logic Toolbox (althought I haven't used it much for real).

MK

Michael Kellett wrote:

 

Won't your fuzzy judges end up like real ones, some good, some bad and most indifferent (perhaps depending on which distro they are based on) ?

The automaton logic and computation subfunctions would need to be written in a formal and verifiable specification language I think, to avoid that kind of problem.  That wouldn't provide any guarantee that it does what we want, but it would be a fairly strong guarantee that it does what we have specified, which is to compute decisions based on the external ruleset.  It must itself contain no domain-space semantics nor built-in value judgement functions internally of course, nor weights --- the whole idea is that those must be external, provided by the citizenry, and subject to public scrutiny by non-programmers.

Needless to say, this is in the realm of pure fiction at the present time.

My friends who are well versed in classical and modern control theory (a very small subset of the group) don't like fuzzy logic because there is no reasonable mathermatical process to prove stability of a fuzzy controller.

Your friends are entirely right!  However, there is no existing social nor legal system that is analytic anyway, let alone with provable stablity, so their worry doesn't mean that such a fuzzy automaton would be no good.  As coder27 pointed out above, the current legal system rests on a foundation of questionable precedents as premises, so an automatic system with a large enough number of safeguard constraints wouldn't have to compete against a very high bar.  By evaluating social utility of a situation rather than taking precedents as axiomatic, it might well do a far better job than humans through simple brute force.

The early days of such a system would be especially interesting, as people compare the performance of human judges against the automaton running in parallel.

Note that evaluating utility would be stillborn if the automaton were being fed with prior human value judgements as input data, so that has to be disallowed.  If you recall the concept of Fair Witness in Heinlein's Stranger in a Strange Land, that would be the kind of input required --- pure facts.  Humans are really bad at leaving out their own assessment of a situation.

coder27 wrote:

Even though the decision is explained in logical terms, it could just have easily gone the other way

and also have been explained in logical terms.  So there is something outside of logic that is actually

swaying the decision. 

This is true. It's interesting that although the facts are similar and only subtly different (ticket machine vs a human agent in older times could be described as just a small difference) the outcome may swing the other way if the facts make it different enough to allow exploration. Also new precedent is something that maybe judges don't want to do too often!

By evaluating social utility of a situation rather than taking precedents as axiomatic, it might well do a far better job than humans through simple brute force.

If it were coded, I hadn't thought of it, but it is quite novel that possibly the system would not need to consider the frequency of near-similar situations occurring again (e.g. is it rare or common for automated ticket machines to exist) and not have to try to make the current decision such that it is a 'fair' outcome for some future decisions too on similar-but-different facts. I wonder where we'd end up, or if we'd approximately end up in the same position. Someone should write a work of fiction based on that :-)

Sadly I'd not heard of Stranger in a Strange Land, but it's now on my to-read list.

Ja die zijn zeker groot neem er vier of zo proberen we het uit ik weet het ook niet kon niet uit Hamelen  

Verstuurd vanaf mijn iPad

Op 6 jul. 2013 om 11:12 heeft morgaine <messages@element14.com> het volgende geschreven:

 

Re: Pi Forums

created by Morgaine Dinova in Raspberry Pi - View the full discussion

Morgaine Dinova wrote:

 

I think it's quite easy to see what's wrong with that situation:  the premises are unsafe, and might easily be flawed.  Using impecable logic one can arrive at nonsense conclusions by arguing from false premises.  I suspect that this applies also in the non-boolean space of value judgements.

 

Lord Denning might well have argued differently simply by choosing a different set of premises from case law as precedent.  Law is really just opinion even when argued logically, and rests on cherry-picked foundations.

That's exactly the problem.  From incorrect premises you can make any conclusion.  I heard that Bertrand Russell made this claim in a lecture and was challenged by a member of the audience to prove that "if 2+2 = 5 then Bertrand Russell is the Pope".  Dr. Russell replied (paraphrased from memory) that "if 2+2 = 5, then 4 = 5.  Subtracting 3 from both sides we get 1 = 2.  Bertrand Russell and the Pope are two people, therefore they are one".  Loud applause ensued.

We get lots of this kind of reasoning in the USA, particularly in Congress about whom our friend Mark Twain had this to say: "Suppose you were an Idiot.  And suppose you were a member of Congress.  But I repeat myself."

shabaz wrote:

 

I wonder where we'd end up, or if we'd approximately end up in the same position.

I think it would create a completely different world for so many epic reasons (such as immunity from corruption and vested interests unless those are what the population wants) that its smaller effects like there being no fixed points in jurisprudence (since it's unaffected by precedent) would go quite unnoticed.

Sadly I'd not heard of Stranger in a Strange Land, but it's now on my to-read list.

If you like SciFi, it's a must-read!  Although I'm biased, since I have almost all of his books.

It must itself contain no domain-space semantics nor built-in value judgement functions internally of course, nor weights

I am perplexed by the requirement of "nor weights".

The judicial system is entirely based on the notion of weights.

See for example "Lady Justice"

http://en.wikipedia.org/wiki/Lady_Justice

the personification of balancing the scales of truth and fairness.

You have probably heard the saying, "if the facts are on your side,

pound on the facts; if the law is on your side, pound on the law;

otherwise pound on the table".

http://quoteinvestigator.com/2010/07/04/legal-adage/

Aside from the humor, the message is that facts count more than

legal technicalities.  If you are a defendant with a good alibi, don't

argue legal technicalities, but if you were caught red-handed,

then by all means argue any legal technicality you can come up with,

but don't count on the court ignoring that you were caught red-handed.

Another example is literal statutory language vs. legislative intent.

The plain language of a statute counts more than the intent of the statute.

If the plain language is on your side, argue that.  If not, argue the intent

of the statute, but don't count on it overcoming the plain language.

Another example in US constitutional law is that different constitutional

amendments have different weights associated with them.  A novice

mistake is to ignore that.  The 1st amendment (free speech) counts

a whole lot more than the 10th amendment (states' rights).   I don't

think anyone has won a case solely based on the 10th amendment.

Lady Justice's balance scale actually should be a balance beam,

where weights can be attached at different distances from the fulcrum.

A strong amendment such as the 1st is reduced in effective weight

based on how tangentially it is applied.  For example, it still has some

effective weight when applied to privacy cases, but not as much.

Similarly, precedents are weighted according to how factually similar

they are.

Sometimes weights are combined additively, where a lot of small

weights can add up to a large weight, and sometimes thresholds

are used in the analysis, where there has to be enough weight to

meet a certain threshold before continuing with the analysis.

Some arguments with little weight are not worth mentioning, because

they make you look like you are grasping at straws.

The early days of such a system would be especially interesting, as people compare the performance of human judges against the automaton running in parallel.

I believe Leibniz proposed applying computers/AI to the field of law.

You could test such a system's ability to guess the outcome of prior cases

before letting it loose on current cases.  But predicting the stock market is

more lucrative than predicting legal outcomes. 

It would be interesting, and maybe someone has already done it, to see

what percentage of the time a case is won that implicates each of

the US constitutional amendments.  It would also be interesting to have

a comprehensive list of legal technicalities that have won cases that were

otherwise hopeless.

I think that to predict supreme court decisions you would benefit most by

studying the voting patterns of each individual justice, because who is appointed

to the court seems to matter more than the legal merits of the case.

It must itself contain no domain-space semantics nor built-in value judgement functions internally of course, nor weights

I am perplexed by the requirement of "nor weights".

The judicial system is entirely based on the notion of weights.

See for example "Lady Justice"

http://en.wikipedia.org/wiki/Lady_Justice

the personification of balancing the scales of truth and fairness.

You have probably heard the saying, "if the facts are on your side,

pound on the facts; if the law is on your side, pound on the law;

otherwise pound on the table".

http://quoteinvestigator.com/2010/07/04/legal-adage/

Aside from the humor, the message is that facts count more than

legal technicalities.  If you are a defendant with a good alibi, don't

argue legal technicalities, but if you were caught red-handed,

then by all means argue any legal technicality you can come up with,

but don't count on the court ignoring that you were caught red-handed.

Another example is literal statutory language vs. legislative intent.

The plain language of a statute counts more than the intent of the statute.

If the plain language is on your side, argue that.  If not, argue the intent

of the statute, but don't count on it overcoming the plain language.

Another example in US constitutional law is that different constitutional

amendments have different weights associated with them.  A novice

mistake is to ignore that.  The 1st amendment (free speech) counts

a whole lot more than the 10th amendment (states' rights).   I don't

think anyone has won a case solely based on the 10th amendment.

Lady Justice's balance scale actually should be a balance beam,

where weights can be attached at different distances from the fulcrum.

A strong amendment such as the 1st is reduced in effective weight

based on how tangentially it is applied.  For example, it still has some

effective weight when applied to privacy cases, but not as much.

Similarly, precedents are weighted according to how factually similar

they are.

Sometimes weights are combined additively, where a lot of small

weights can add up to a large weight, and sometimes thresholds

are used in the analysis, where there has to be enough weight to

meet a certain threshold before continuing with the analysis.

Some arguments with little weight are not worth mentioning, because

they make you look like you are grasping at straws.

The early days of such a system would be especially interesting, as people compare the performance of human judges against the automaton running in parallel.

I believe Leibniz proposed applying computers/AI to the field of law.

You could test such a system's ability to guess the outcome of prior cases

before letting it loose on current cases.  But predicting the stock market is

more lucrative than predicting legal outcomes. 

It would be interesting, and maybe someone has already done it, to see

what percentage of the time a case is won that implicates each of

the US constitutional amendments.  It would also be interesting to have

a comprehensive list of legal technicalities that have won cases that were

otherwise hopeless.

I think that to predict supreme court decisions you would benefit most by

studying the voting patterns of each individual justice, because who is appointed

to the court seems to matter more than the legal merits of the case.

coder27 wrote:

 

I am perplexed by the requirement of "nor weights".

The judicial system is entirely based on the notion of weights.

You misunderstood me there.  It's the automaton that must contain no weights itself, because the automaton is the semantic-free formally specified engine (an executable program) that implements the fuzzy logic by processing the factual inputs of a case against the large external ruleset of piecewise linear merit functions.  It's the external rules which are weighted.  The engine knows how to apply weights and rules, but carries none internally.  This is a case of separation of mechanism and policy not just for general goodness, but as a crucial part of how this would work fairly.

There is no judicial policy in the engine as it's specified and implemented by programmers who have no business hardwiring any judicial semantics into the engine code.  That's why the engine must be open source and visible to all, so that its agnosticism to policy is verifiable by anyone with programming skills, and so that anyone can track the judgements in their own versions of the automaton if they wish.  It doesn't change unless someone finds an error in this specific implementation or in the general specification.

It's the external public database of weighted fuzzy rules that would change, and do so continually as society evolves.  It's by growing and honing and reweighting this external ruleset that the citizenry determines judicial policy.  They don't touch the engine for this.

It's the external rules which are weighted.

OK, but external rules don't normally come with weights,

the weights need to be deduced by some sort of regression analysis.

For example, there is nothing explicit in the US constitutional amendments

that says how much weight to apply to each amendment.

coder27 wrote:

 

 

 

OK, but external rules don't normally come with weights,

the weights need to be deduced by some sort of regression analysis.

You must be thinking of something else.  These fuzzy rules are each weighted by the citizenry to specify their relative importance.  Fuzzy logic doesn't deal in just true OR false.  An input can be both good and bad at the same (across any number of dimensions), but in different measures --- those are the rule weights, and they can't be determined analytically.  Assigning those weights to rules in the public database is where human commonsense gets applied.

they can't be determined analytically

Google translate works by analytically determining weights, based

on having scanned lots of documents, rather than having lots of

English teachers write down lots of grammar rules.  

I think a computerized legal system would have to work the same way.