I think software and most other engineering products need a much
higher level of coherence than artefacts that are consumed by humans,
like Wikipedia. Wikipedia is full of inconsistencies and even
contradictions. When I browse Wikipedia, I often stumble upon
statements on one page that are contradicted by another page. That's
not a big deal - sometimes I can easily tell which statement is
correct (and fix the other pages), sometimes I can't, but either way,
I am not a computer: I don't follow these statements blindly. In a
computer program, such inconsistencies would lead to erratic behavior,
i.e., bugs. This means that a completely open wiki process will not
work for software.
In a way, a lot of open source software is developed in a restricted
wiki way: someone proposes a change, but before it is merged, it is
checked by people who (hopefully) know all the nooks and crannies of
the existing code. A bit like edit-protected pages in Wikipedia:
everyone can propose changes on the talk page, but only admins can
actually make these changes.
Christopher
On 7 July 2013 04:16, Michael Hale <hale.michael.jr(a)live.com> wrote:
I'm glad you mentioned that the same issue applies
to electronics. I suppose
I could have just referred to Moore's law instead of the relatively recent
increasing size of datacenters. I like asking computers to work hard, but I
find it hard to think of valuable things for them to do. You can play a new
game or donate time to BOINC, but not very many great games are produced
each year and BOINC typically runs algorithms that benefit humanity but not
specifically you. For example, my genetics tests say I have an increased
risk of prostate cancer, so I'd like to be able to tell Folding@home to
focus on the proteins that are most relevant for the diseases I'm most
likely to get.
I still have hope that a more wiki-like model could work for developing
software libraries though. The problems of technical design in software and
hardware are similar, but software can be developed more fluidly and rapidly
due to the lower barrier to entry and non-existent manufacturing costs.
Essentially all electronics are designed and simulated with software prior
to constructing physical prototypes these days.
I've thought about the integration problem some, but I haven't ironed out
how it would all work yet. I think standard object-oriented programming and
modeling techniques have been absorbed by enough programmers that it might
be worth a shot though. Essentially, each article would have a standard
class and supporting data structures or file formats for the inputs and
outputs of its algorithms. It would be like the typical flow chart or visual
programming languages you can use with libraries like Modelica, but on a
larger scale and the formats would often be more complex. So, like, you
would have a class representing a cloud, with flags for different
representations (density values in a cubic grid, collections of point
particles, polygonal shape approximations, etc) which are used for different
algorithms. So then you would have code that can convert between all of the
representations, code for generating random clouds (with potentially lots of
optional parameters to specify atmospheric conditions), code for outputting
images of the generated clouds in different styles, and algorithms for
manipulating them through time. Then if I wanted to see the effects on a
specific cloud I've made drifting over the ocean in different atmospheric
conditions, I could grab the code to instantiate 3D Euclidean space with a
virtual camera, add some gravity, add some ground, add some water, add an
atmosphere, add my cloud, and then simulate it with adjustable parameters
for the accuracy and speed of computation. Now, there are a lot of details
that leaves out, but I don't know of another way to easily mix capabilities
from high-end graphics software and various specialized simulation
algorithms in lots of ways. Graphics software typically gives you some
simulation capabilities, and simulation software typically gives you some
graphics functionality, but I want lots of both.
I think having more semantic annotation tools will be great, but I don't
spend most of my time doing searches. There is an astounding amount of
information, data, and media on the internet, but it's not hard to find the
edge if you really try. It's pretty crazy if you search for images of "blue
bear" how many results come up, but if you search for "blue bear and green
gorilla" you don't get anything useful. Then you get to face the craziness
of how many options you have for combining a picture of a blue bear and a
different picture of a green gorilla into one picture. I think it's
interesting what they are trying with the Wolfram Alpha website, but they
will always have to limit the time of the computations they allow you to do
on their servers, so that's why I think we need better libraries to more
easily program the computers we have direct control over.
________________________________
From: dacuetu(a)gmail.com
Date: Sat, 6 Jul 2013 17:49:41 -0400
To: wikidata-l(a)lists.wikimedia.org
Subject: Re: [Wikidata-l] Accelerating software innovation with Wikidata and
improved Wikicode
Thanks for sharing your thoughts Michael, it is also something that has been
bothering me for a while and not only in programming, also in other
technical domains like electronics.
In my opinion, the reason why programming (or technical design in general)
couldn't follow the wiki world is because it has some structural differences
that require a different approach. To start with, there is the problem of
integration, where code solutions are usually part of a larger system and
they cannot be isolated or combined with others blocks as easily as you
would combine text fragments like in Wikipedia. I'm sure that all those 10
open file examples have some particularities about the operative system,
method, supporting libraries, etc.
The part of scavenging and gluing will be always there unless you follow the
approach used in hardware design (wp: semiconductor intellectual property
core).
Since that kind of modularity trend is hard to set up at large scale other
than what is already stablished, it would be more practical to focus on what
can be improved more easily, which is the scavenging. Instead of copying
code fragments, it would be better to point to the fragment in the source
code project itself, while at the same time providing the semantic tags
necessary for describing that fragment. This can be done (more or less) with
current existing semantic annotation technology (see thepund.it and Dbpedia
Spotlight).
If this has not been done before it is maybe because semantic tools are now
in the transition from "adaptation of an emerging technology" into
"social
appropriation of that technology". For the wiki concept it took 6 years for
it to be transformed into wikipedia, more or less the same amount of years
between SMW and Wikidata. Semantic annotation of code will eventually
happen, how fast it will depend on interest in such a tool and the success
of the supporting technologies.
Micru
On Sat, Jul 6, 2013 at 3:10 PM, Michael Hale <hale.michael.jr(a)live.com>
wrote:
I have been pondering this for some time, and I would like some feedback. I
figure there are many programmers on this list, but I think others might
find it interesting as well.
Are you satisfied with our progress in increasing software sophistication as
compared to, say, increasing the size of datacenters? Personally, I think
there is still too much "reinventing the wheel" going on, and the best way
to get to software that is complex enough to do things like high-fidelity
simulations of virtual worlds is to essentially crowd-source the translation
of Wikipedia into code. The existing structure of the Wikipedia articles
would serve as a scaffold for a large, consistently designed, open-source
software library. Then, whether I was making software for weather prediction
and I needed code to slowly simulate physically accurate clouds or I was
making a game and I needed code to quickly draw stylized clouds I could just
go to the article for clouds, click on C++ (or whatever programming language
is appropriate) and then find some useful chunks of code. Every article
could link to useful algorithms, data structures, and interface designs that
are relevant to the subject of the article. You could also find data-centric
programs too. Like, maybe a JavaScript weather statistics browser and
visualizer that accesses Wikidata. The big advantage would be that
constraining the design of the library to the structure of Wikipedia would
handle the encapsulation and modularity aspects of the software engineering
so that the components could improve independently. Creating a simulation or
visualization where you zoom in from a whole cloud to see its constituent
microscopic particles is certainly doable right now, but it would be a lot
easier with a function library like this.
If you look at the existing Wikicode and Rosetta Code the code samples are
small and isolated. They will show, for example, how to open a file in 10
different languages. However, the search engines already do a great job of
helping us find those types of code samples across blog posts of people who
have had to do that specific task before. However, a problem that I run into
frequently that the search engines don't help me solve is if I read a
nanoelectronics paper and I want to do a simulation of the physical system
they describe I often have to go to the websites of several different
professors and do a fair bit of manual work to assemble their different
programs into a pipeline, and then the result of my hacking is not easy to
expand to new scenarios. We've made enough progress on Wikipedia that I can
often just click on a couple of articles to get an understanding of the
paper, but if I want to experiment with the ideas in a software context I
have to do a lot of scavenging and gluing.
I'm not yet convinced that this could work. Maybe Wikipedia works so well
because the internet reached a point where there was so much redundant
knowledge listed in many places that there was immense social and economic
pressure to utilize knowledgeable people to summarize it in a free
encyclopedia. Maybe the total amount of software that has been written is
still too small, there are still too few programmers, and it's still too
difficult compared to writing natural languages for the crowdsourcing
dynamics to work. There have been a lot of successful open-source software
projects of course, but most of them are focused on creating software for a
specific task instead of library components that cover all of the knowledge
in the encyclopedia.
_______________________________________________
Wikidata-l mailing list
Wikidata-l(a)lists.wikimedia.org
https://lists.wikimedia.org/mailman/listinfo/wikidata-l
--
Etiamsi omnes, ego non
_______________________________________________ Wikidata-l mailing list
Wikidata-l(a)lists.wikimedia.org
https://lists.wikimedia.org/mailman/listinfo/wikidata-l
_______________________________________________
Wikidata-l mailing list
Wikidata-l(a)lists.wikimedia.org
https://lists.wikimedia.org/mailman/listinfo/wikidata-l