# Turing up a level?

**URL:** <http://discourse.iapct.org/t/turing-up-a-level/11814>\
**Category:** CSG1996\
**Created:** [February 12, 1996, 6:15pm UTC](http://discourse.iapct.org/t/turing-up-a-level/11814 "1996-02-12T18:15:49Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![Peter\_A\_Cariani1](http://discourse.iapct.org/letter_avatar_proxy/v4/letter/p/f17d59/32.png) [@Peter\_A\_Cariani1](http://discourse.iapct.org/u/Peter_A_Cariani1)\
**Post date:** [February 12, 1996, 6:15pm UTC](http://discourse.iapct.org/t/turing-up-a-level/11814/1 "1996-02-12T18:15:49Z")

</div>

[Peter Cariani, 960213, 1300]

Bill Powers wrote: [Bill Powers (960213.0300 MST)]

> Do you think that any computer program (as we now conceive programming)  
> could "go up a level?" That is, could the machine make any comments  
> about why it is doing what it is doing? Could it derive a general  
> principle from the interaction? A system concept like "computer" or  
> "machine"?

I think devices are possible that could do such things, but the  
kinds of devices would be different from the kinds of  
digital computers we now have. In order for anything qualitative  
new to evolve in an artificial device, (I think)  
the device must have some degree of autonomy of  
structure/function relative to its designer.  
I think that such a device would  
need to be able to evolve its own hardware  
(or at least to appropriate more/new hardware from somewhere else)  
and to augment its own sensors, effectors, and computational parts  
to do these kinds of things, which require that the "state space"  
of the device be (autonomously) expanded in its dimensionality.

I don't think it can be \<realized\> in a computer simulation,  
unless such concepts are somehow allowed for by the designer  
of the simulation beforehand (in other words they  
would not be created entirely de novo). One can make programs that  
analyze their own structure, but I don't think they really get at  
what I would want (new "systems concepts"). One can also \<represent\>  
the process in a limited way in a simulation, but this will not  
in and of itself, realize new concepts in the machine  
in an open-ended way.

I'm sure you've thought long and hard about various  
analog devices that would evolve new (higher-level) control loops,  
and (as you probably have also thought)  
it might actually be easier to build an analog one than  
a mixed-digital-analog one.

Peter Cariani  
peter@epl.meei.harvard.edu

Please cc: me a copy of any replies for the time being. I'm still  
getting the bugs out of our email return address. Thanks.