On Sat, 2008-06-28 at 04:21 -0700, Alan Kay wrote: > It was realized that most computing of the 50s and 60s was rather like > synthetic chemistry in which structures are built atom by atom and > molecule by molecule. This gets more and more difficult for larger and > more complex combinations. "Life" generally uses a process quite > different -- instead of building cells and organisms, it grows them. > This leads to seeming paradoxes in the epistemology of making -- i.e. > to make a cell we need a cell., to make a chicken we need a chicken. > However, all works fine in the epistemology of growing. But the > initial bootstrapping is a little tricky. Once the bootstrap is done > to make life then life can assist much more powerfully in making more > life, and to vary life, etc. As mentioned before, the Internet is one > of these, and so is Smalltalk. > > In "biologically inspired" architectures one is much more interested > in how the organism and ecologies of them are to be sustained, and how > the dynamical systems can be understood, fixed, changed, evolved, > reformulated, etc., while they are running and with the help of tools > that work within them. Just as a cell, and especially e. g. a human > baby, is not made by hand, we are more interested in making growth > processes that can be influenced with far more ease than direct > construction. So, most objects are made by other objects acting under > conditions that include the dynamic state in which they will become > part of the ecology.
It seems to me that this analogy is a fairly good one -- although there are definite differences, in that the 'cell' needed to make another 'cell' in the etoys/squeak case has been designed with lots of tools to make it easy to inspect and modify itself, as well as to completely sequence its DNA or produce a new generation on demand. Living systems are however notorious for carrying historical baggage along with them in their genotype, and, since the phenotype cannot easily be recreated without starting with a parent phenotype, the Ken Thompson hack implies that inheritable baggage can (paradoxically) be carried in the phenotype as well. I think the number of somewhat independent tools provided to in[tro]spect a running image would make an intentional malicious Thompson hack in Sqeak quite difficult to maintain for long without discovery; but I would naively guess that there is some 'harmless' baggage that looks reasonable and is allowed to continue just due to inertia. Since the 'DNA' (source code) of etoys/squeak is readily available in a transparent, human-understandable form, it seems to me that the only issue of possible concern is the lesser visibility of the 'paradoxical' inheritance via phenotype/image. Or at least, its lesser visibility if one refuses to run etoys/squeak to use the tools it provides to inspect itself or its images. Continuing with the biological analogy, the folks who want to be able to bootstrap a Squeak/etoys image (starting from 'scratch' without such an image) want literally to be able to make ontogeny recapitulate phylogeny -- not necessarily every time an image starts, possibly not necessarily every time Squeak is 'built' -- but at least with similar frequency and the ease of bootstrapping gcc using a different C compiler. (Like using a turtle egg to hatch a dinosaur ;-) I don't have a strong opinion on this myself, but I do find the discussion interesting. - Dan _______________________________________________ Devel mailing list [email protected] http://lists.laptop.org/listinfo/devel
