Re: Abduction and Introspection

Posted by Frank Wimberly-2 on Jan 26, 2020; 3:03pm
URL: http://friam.383.s1.nabble.com/Abduction-and-Introspection-tp7594386p7594407.html

Introspection :  it would be possible to insert a causal inference module into an agent-based modeling program.  The ABM could examine the causal conclusions emerging from the data it's generating.  It seems to me that is close to what "introspection" means to me.

Frank

On Sun, Jan 26, 2020, 7:56 AM Frank Wimberly <[hidden email]> wrote:

Validation:  Bill Reynolds wrote a paper about inferring causal models from observational data to validate, for example, agent based models.  My original idea was that If the same causal edges emerge as are observed in the modeled system then that helps validate the model.  Bill thought it was better to compare the causal model with experts' opinions regarding causation in the modeled system.  Good enough.

Reynolds, W. N., Wimberly, F. C.,

Simulation Validation Using Causal     Inference Theory with Morphological Constraints.  Proceedings of the     2011 Winter Simulation Conference, Arizona Grand Resort, December,     2011.

I mentioned this here recently and Glen asked for a copy so I sent it to him.  I look forward to his comments.

Frank

On Sat, Jan 25, 2020, 11:48 PM Pieter Steenekamp <[hidden email]> wrote:

I would go along with Johsua Epstein's "if you did not grow it you did not explain it". Keep in mind that this motto applies to problems involving emergence. So what I'm saying is that it's in many cases futile to apply logic to reasoning to find answers - and I refer to the emergent properties of the human brain as well as to ABM (agent based modeling) software. But even if the problem involves emergence, it's easy for both human and computer logic to apply validation logic. Similar to the P=NP problem*, it's difficult to find the solution, but easy to verify. 

So my answer to "As software engineers, what conditions would a program have to fulfill to say that a computer was monitoring “itself" is simply: explicitly verify the results. There are many approaches to do this verification; applying logic, checking against measured actual data, checking for violations of physics, etc.   

*I know you all know it, just a refresher, The P=NP problem is one of the biggest unsolved computer science problems. There is a class of very difficult to solve problems and a class of very easy to verify problems. The P=NP problem asks the following: if you have a difficult to solve but easy to verify problem, is it possible to find a solution that is reasonably easy for a computer to solve. "Reasonably easy" is defined as can you solve it in polynomial time. The current algorithms takes exponential time to solve it and even for a moderate size problem that means more time that the age of the universe for a supercomputer to solve it.

Pieter

On Sat, 25 Jan 2020 at 23:04, Marcus Daniels <[hidden email]> wrote:

I would say the problem of debugging (or introspection if you insist)  is like if you find yourself at some random place, never seen before, and the task it do develop a map and learn the local language and customs.  If one is given the job of law enforcement (debugging violations of law), it is necessary to collect quite a bit of information, e.g. the laws of the jurisdiction, the sensitivities and conflicts in the area, and detailed geography.  In haphazardly-developed  software, learning about one part of a city teaches you nothing about another part of the city.   In well-designed software, one can orient oneself quickly because there are many easily-learnable conventions to follow.    I would say this distinction between the modeler and the modeled is not that helpful.   To really avoid bugs, one wants to have metaphorical citizens that are genetically incapable of breaking laws.   Privileged access is kind of beside the point because in practice software is often far too big to fully rationalize. 

 

From: Friam <[hidden email]> on behalf of "[hidden email]" <[hidden email]>
Reply-To: The Friday Morning Applied Complexity Coffee Group <[hidden email]>
Date: Saturday, January 25, 2020 at 11:57 AM
To: 'The Friday Morning Applied Complexity Coffee Group' <[hidden email]>
Subject: Re: [FRIAM] Abduction and Introspection

 

Thanks, Marcus,

 

Am I correct that all of your examples fall with in this frame;

 

I keep expecting you guys to scream at me, “Of course, you idiot, self-perception is partial and subject to error!  HTF could it be otherwise?”   I would love that.  I would record it and put it on loop for half my colleagues in psychology departments around the world. 

 

Nick

Nicholas Thompson

Emeritus Professor of Ethology and Psychology

Clark University

[hidden email]

https://wordpress.clarku.edu/nthompson/

 

 

From: Friam <[hidden email]> On Behalf Of Marcus Daniels
Sent: Saturday, January 25, 2020 12:16 PM
To: The Friday Morning Applied Complexity Coffee Group <[hidden email]>
Subject: Re: [FRIAM] Abduction and Introspection

 

Nick writes:

 

 As software engineers, what conditions would a program have to fulfill to say that a computer was monitoring “itself

 

It is common for codes that calculate things to periodically test invariants that should hold.   For example, a physics code might test for conservation of mass or energy.   A conversion between a data structure with one index scheme to another is often followed by a check to ensure the total number of records did not change, or if it did change that it changed by an expected amount.   It is also possible, but less common, to write a code so that proofs are constructed by virtue of the code being compliable against a set of types.   The types describe all of the conditions that must hold regarding the behavior of a function.    In that case it is not necessary to detect if something goes haywire at runtime because it is simply not possible for something to go haywire.  (A computer could still miscalculate due to a cosmic ray, or some other physical interruption, but assuming that did not happen a complete proof-carrying code would not fail within its specifications.)

A weaker form of self-monitoring is to periodically check for memory or disk usage, and to raise an alarm if they are unexpectedly high or low.   Such an alarm might trigger cleanups of old results, otherwise kept around for convenience. 

 

Marcus

 

============================================================
FRIAM Applied Complexity Group listserv
Meets Fridays 9a-11:30 at cafe at St. John's College
to unsubscribe http://redfish.com/mailman/listinfo/friam_redfish.com
archives back to 2003: http://friam.471366.n2.nabble.com/
FRIAM-COMIC http://friam-comic.blogspot.com/ by Dr. Strangelove

============================================================
FRIAM Applied Complexity Group listserv
Meets Fridays 9a-11:30 at cafe at St. John's College
to unsubscribe http://redfish.com/mailman/listinfo/friam_redfish.com
archives back to 2003: http://friam.471366.n2.nabble.com/
FRIAM-COMIC http://friam-comic.blogspot.com/ by Dr. Strangelove

============================================================
FRIAM Applied Complexity Group listserv
Meets Fridays 9a-11:30 at cafe at St. John's College
to unsubscribe http://redfish.com/mailman/listinfo/friam_redfish.com
archives back to 2003: http://friam.471366.n2.nabble.com/
FRIAM-COMIC http://friam-comic.blogspot.com/ by Dr. Strangelove