Semantic Web awol from comp.ai?
Jorn Barger
http://groups.google.com/groups?scoring=d&q=group:comp.ai+semantic.web
--shows almost no discussion of the Semantic Web project
on comp.ai, ever? Why isn't it even mentioned in the
'FAQ'? What does the academic-AI community think of it,
in terms of practicality or theoretical interest?
Aren't businesses spending tons of money implementing
XML data-standards, and don't they hire AI-grads for
expertise on this? Or haven't AI grad-programs been
adequately training students in these semantic skills?
[ comp.ai is moderated. To submit, just post and be patient, or if ]
[ that fails mail your article to <com...@moderators.isc.org>, and ]
[ ask your news administrator to fix the problems with your system. ]
JXStern
>Can someone explain to me why this Google Groups search--
> http://groups.google.com/groups?scoring=d&q=group:comp.ai+semantic.web
>--shows almost no discussion of the Semantic Web project
>on comp.ai, ever? Why isn't it even mentioned in the
>'FAQ'? What does the academic-AI community think of it,
>in terms of practicality or theoretical interest?
>
>Aren't businesses spending tons of money implementing
>XML data-standards, and don't they hire AI-grads for
>expertise on this? Or haven't AI grad-programs been
>adequately training students in these semantic skills?
You ask many interesting questions. I will speculate on the answers.
>Can someone explain to me why this Google Groups search--
> http://groups.google.com/groups?scoring=d&q=group:comp.ai+semantic.web
>--shows almost no discussion of the Semantic Web project
>on comp.ai, ever?
The amount of discussion on comp.ai ever on any topic is pretty
limited. How many threads have more than a handfull of messages? How
many have even two?
> Why isn't it even mentioned in the 'FAQ'?
The word "semantic" appears only a modest number of times. Neither
"ontology" nor "xml" appear at all.
> What does the academic-AI community think of it,
>in terms of practicality or theoretical interest?
Res ipsa loquitur?
>Aren't businesses spending tons of money implementing
>XML data-standards, and don't they hire AI-grads for
>expertise on this? Or haven't AI grad-programs been
>adequately training students in these semantic skills?
No, no. No.
Businesses may be installing some newer products that aspire to use
XML and provide standards, but frankly, I couldn't name them, and I
can't say there is much of a push on in business today to push the
state of the art in any direction at all.
I doubt you can find fifty instances of businesses hiring AI-grads to
provide expertise on XML projects in the U.S. in a given year. I base
this skepticism not because I have any private (or public) source of
relevant data, but because of the widespread trend today to hire very
marginally qualified developers for all projects.
I am also unaware of any AI program that emphasises semantic skills.
I'm not even sure what "semantic skills" means. Something like
"knowledge engineer" once did, I suppose. I would speculate that most
AI programs concentrate on technology and research, and consider
ontologies and "semantic skills" as mere practices.
Now, I am sympathetic to your question. I suggest that a focus on
applicable semantics and domain-specific ontologies are great
frontiers for AI. However, I am unaware of where if anywhere these
exist today.
Joshua Stern
Eray Ozkural exa
> I am also unaware of any AI program that emphasises semantic skills.
> I'm not even sure what "semantic skills" means. Something like
> "knowledge engineer" once did, I suppose. I would speculate that most
> AI programs concentrate on technology and research, and consider
> ontologies and "semantic skills" as mere practices.
>
> Now, I am sympathetic to your question. I suggest that a focus on
> applicable semantics and domain-specific ontologies are great
> frontiers for AI. However, I am unaware of where if anywhere these
> exist today.
It's because there are no "killer-apps" for ontologies. In other
words, nobody has ably demonstrated that these ontologies actually
have some hint of intelligence that could be used to improve people's
lives in a profitable way. ;)
So, although computational ontology research is interesting with
respect to its connections to formal logic, philosophy, programming
languages, visualization and HCI among other fields I don't know if it
will ever be widespread.
What I'm trying to say is, google is a killer-app for IR. It does
parallel IR reasonably well and everybody uses it. Can you imagine a
counterpart of google for ontology editors, etc.? I think it could be
used in a lot of domains like user interfaces to begin with but I also
think there are quite a few obstacles. First thing, I know that nobody
has the temper to edit an ontology or truly understand what ontology
is. Second, it's almost impossible to make ordinary people collaborate
using a formal language. Therefore, one needs automated methods to
make ontologies simply work, and then interact with the user in a very
natural and visual way. And XML syntax, IMO, hasn't much to do with
the main problems. In fact, I would see it as one of the obstacles
itself (being an XML-disliker myself)
I once did an ontology project for net repositories, it contains some
nice links, an introduction to ontologies and a preliminary ontology
description language design. You might find it interesting.
http://borg.cs.bilkent.edu.tr/~exa/ontology/
Regards,
__
Eray Ozkural (exa) <erayo at cs.bilkent.edu.tr>
baylor
Never heard of it
> >Aren't businesses spending tons of money implementing
> >XML data-standards, and don't they hire AI-grads for
> >expertise on this? Or haven't AI grad-programs been
> >adequately training students in these semantic skills?
>
> No, no. No.
>
> Businesses may be installing some newer products that aspire to use
> XML and provide standards, but frankly, I couldn't name them, and I
> can't say there is much of a push on in business today to push the
> state of the art in any direction at all.
>
> I doubt you can find fifty instances of businesses hiring AI-grads to
> provide expertise on XML projects in the U.S. in a given year. I base
> this skepticism not because I have any private (or public) source of
> relevant data, but because of the widespread trend today to hire very
> marginally qualified developers for all projects.
i considered ignoring this, but i'm guessing no one else on this group
is going to say anything so i might as well
First, my bias - i was until recently head of AI, R&D and standards of
a Fortune 500 company. i've funded research labs at fancy AI schools,
offered to sponsor AI interns, gone to AI conferences, underwrote (to
the tune of tens of millions) AI-based companys and all that sort of
thing. i was also partially responsible for deciding who we hire,
setting up the training class we put all hires through and grading
their assignments (which takes a while given the 1,000 IT people we
had). i also taught classes on emerging technology to our marketers
and buyers
Do businesses spend money on XML? Sadly, yes. It's mostly wasted
money, but money gets spent
Do businesses hire AI people? Of course not. Except for a couple of
ancient departments in engineering companies, practically no one in
business hires AI people. There's no reason to, and frankly how many
AI PhDs want to work for a retailer or financial company or one of
your common businesses. i can tell you i've never had a single app
from a person with a PhD in AI
Would we hire AI people for XML? Why on earth would we want to do
that? i don't understand the connection. Maybe i'm missing something,
but to me it sounds similar to "why don't you hire sociologists to
make mashed potatoes?". i just don't understand the question
Have AI schools been properly preparing their students for work in
business? i realize this is a big argument in a lot of places and so i
don't want to fan the flames here, but my personal opinion having seen
several AI schools, research labs, spin off companies and professors
is that AI schools don't prepare students for anything other than
teaching AI
Here's a quick critque you'll hear from a lot of people like me who
purchase products & services - AI (or name your other favorite field)
doesn't understand business
i'd go one step further and say that the field of computer science
that some people call AI doesn't actually understand anything useful.
That's been my experience watching literally every AI project we've
attempted (all run by PhDs from top 10 AI schools) fail miserably,
from conversations with other sponsors/business people at AI
conferences/symposia, from discussions with people in other fields and
is even the opinion of a few people who make their living at AI (a
decent write up of this is in Jon Doyle's "Big Problems In AI" in AI
Magazine around 1998; Doyle was a CMU AI prof and i think might
currently by an MIT one; his critique, sponsored by the US air force
was on the egos in AI and the field's unwillingness to work with other
fields)
If you could better explain the question, maybe i could give you
specifics. i've spent a lot of time over the years giving pretty
detailed explanations to vendors as to what's wrong with what they
want to do. Business isn't normally complex (yes, we have a few weird
requirements, but most are pretty obvious). But my experience with
people who make products, tools and services is that they don't have a
clue how we make money. With PhDs and academics specifically, i've
found that the people focus *so* much on one tiny little topic (object
constraint languages, say) that interests them that they never get
around to working on the parts that would have made the tool useful to
us. Do you know how many times i've heard a sales pitch from someone
wanting to sell me a stand-alone lisp-based system or require users to
enter data in text files (with no error checking, of course) or
couldn't handle stateless architectures or couldn't work with
relational data or couldn't work with objects or couldn't be updated
while running or any of a hundred unrealistic limitations
As for the semantic Web, i've never heard of it. But i did at one
point start a list of AI algorithms and concepts and approaches for an
AI presentation/class i taught to senior management. Then i gave up
because the field cranks out tons of tiny, unknown,
heading-in-separate-directions "stuff", few of which achieve the
critical mass necessary to make the the things useful
There were maybe five people full time and plenty of people in their
spare time combing the magazines, going to conferences, talking to
professors, etc and we were inundated by ideas an information, most of
it worthless. So if i hadn't heard of a semantic web two years ago,
there's a good chance that it wasn't worth hearing about
Sorry to sound harsh, but i hate hearing "what's wrong with people
that they don't recognize the obvious superiority of XYZ" or "business
people are morons who only hire other morons". i'm hoping it helps to
not preach to the choir and get some feedback from those of us who
doing the hiring of marginal developers
In regards to this specific comment:
> I doubt you can find fifty instances of businesses hiring AI-grads to
> provide expertise on XML projects in the U.S. in a given year. I base
> this skepticism not because I have any private (or public) source of
> relevant data, but because of the widespread trend today to hire very
> marginally qualified developers for all projects.
First, the choice is not between AI-grads and maginally qualified
developers. Those two things are hardly mutually exclusive. i've met a
few AI PhDs i would have hired, but most were exceedingly unimpressive
and nowhere near the top of the candidate list
Second, is it a widespread trend to hire marginally qualified
developers? Yup. It's also true that we hire the very best people
available (well, not always; a lot of HR people screw up and many
people don't know how to interview people, but the best do tend to get
hired before the less qualified). Both things are true. Consider that
for a second
Here's another personal opinion - the IT industry is run worse than
any other industry. While i might make my living at it, i personally
think companies spend FAR too much on computers and pay far too much
attention to the cutting edge. i'm not sure how much we really need
new technology. Pretty much everything we need to be able to do we are
able to do. Our problem isn't ideas and potential, it's execution. We
constantly buy new technology and then use it poorly. We don't need
more technology, nor do we need more technical people. We need better
managers to decide what our priorities are and to apply the technology
more intelligently. And we need to ignore the technical people more
since the tendency of many (probably most) tech people is to want to
so something because "it's neat" rather than profitable
No department in the companies i've worked for is run perfectly, but
some are run pretty well. IT almost never is. Not when you look at how
consistently we miss deadlines, go over budget, deliver the wrong
functionality, approve projects of no value and create things that are
lacking in quality. If high school dropouts can put up a house for a
fixed cost in a specified period of time and be profitable, and if
Hollywood can create movies that have far more variables and
creativity than your average IT system and still come close to hitting
deadlines most of the time, why can't a bunch of PhDs and MBAs manage
to create a cash register program within 100% of the estimated budget
and deadline?
Well, there are reasons, but they take us off topic. The point i want
to make is that, when you need to hire 1,000 or 3,000 or 7,000
computer people to build simple systems, guess what, you sometimes
hire people who aren't going to single handedly change the world. For
different reasons. The energetic, "smart" ones are often arrogant and
undisciplined and thus create useless garbage. They also tend to goof
off (they work 60 hr weeks, but 40 of those are on personal projects).
The ones who understand the business and what the system needs to do
often aren't visionary (which isn't such a bad thing) and the things
they do right often go unnoticed and unrewarded. Some people don't
like the kind of systems they work on and are thus doing the minimum
possible. Some good ones make the mistakes of going into management
because the lowest paid manager tends to make more than the highest
paid tech person (i have no idea who came up with that genius idea). A
lot of managers are undisciplined cowboys and make programmers work on
useless things. And managers don't hold vendors accountable nearly
enough, and so we spend millions every year on some of the most
reprehensible junk you've ever seen. And probably the worst problem in
business computing, we don't talk about out problems (other people's,
yes, but not our own) and we don't hold people accountable. AI (and
maybe all academia; i don't know) is like that too. It's too easy to
do poor work and pass it off as visionary or important or useful.
People are too polite, too scared to hurt someone's feelings while
they're in the room (it changes when you leave)
One last thing people tend to forget. No AI researcher has an
inaliable right to someone else's money. If you want to do your own
thing and work on something with unproven value, that's fine, but you
can hardly expect me to give you $300,000 a year to do it. The
military funds basic research, not business. If i thought i'd make
money off something you were doing, i'd cut you a check for a couple
million no problem. My peers have done it time and again with people
(AI and non-) who didn't have a chance in hell of delivering what they
were promising. i'm a bit more of a hardnose, but the money's out
there. But it's investment money, not grant or charity money
-b
Jorn Barger
> > I am also unaware of any AI program that emphasises semantic skills.
> > I'm not even sure what "semantic skills" means.
If AI-schools took semantics seriously (as they should), a
minimal curriculum would have to look, in effect, at each of
Roget's categories and how that class of words/concepts has
been or might be represented in silicon.
A history of ontologies since Aristotle would be useful, along
with the evolving toolkit of semantic nets. I've made a start:
http://www.robotwisdom.com/ai/timeline/0000.html
Many of these ontologies were tracked down by Fritz Lehmann:
http://www.robotwisdom.com/ai/fritz.html
> > Something like "knowledge engineer" once did, I suppose.
This is definitely one subset of the domain, yes.
> > I would speculate that most
> > AI programs concentrate on technology and research, and consider
> > ontologies and "semantic skills" as mere practices.
(Ie, they search for the carkeys where the light is good, not
where they dropped them.)
> > Now, I am sympathetic to your question. I suggest that a focus on
> > applicable semantics and domain-specific ontologies are great
> > frontiers for AI. However, I am unaware of where if anywhere these
> > exist today.
er...@bilkent.edu.tr (Eray Ozkural exa) wrote in message news:<b9kmm5$ndf$1...@mulga.cs.mu.OZ.AU>...
> It's because there are no "killer-apps" for ontologies. In other
> words, nobody has ably demonstrated that these ontologies actually
> have some hint of intelligence that could be used to improve people's
> lives in a profitable way. ;)
The _theoretical_ utility is clear, it's just the practice that
falls short.
> So, although computational ontology research is interesting with
> respect to its connections to formal logic, philosophy, programming
> languages, visualization and HCI among other fields I don't know if it
> will ever be widespread.
Lots of people are beating their heads against it, but until
someone makes a breakthru, there's no reason for it to become
more widespread. (I offer fractal thickets as a breakthru:
http://www.robotwisdom.com/ai/thicketfaq.html )
> What I'm trying to say is, google is a killer-app for IR. It does
> parallel IR reasonably well and everybody uses it. Can you imagine a
> counterpart of google for ontology editors, etc.?
Google is always evolving, so the question could be rephrased:
"Can you imagine Google with more-explicit ontological
features?" Some likely possibilities:
- categorisation of webpages (etext, faq, image-gallery, etc)
- extraction of persons, places, things, and dates
- composition of search-results into sorted subcategories
(Altavista is doing this now)
> I think it could be
> used in a lot of domains like user interfaces to begin with but I also
> think there are quite a few obstacles. First thing, I know that nobody
> has the temper to edit an ontology
Supposedly the topic-maps crowd is making progress by attacking
from a very practical angle.
> or truly understand what ontology is.
It's trivial to show examples-- what's hard is to explain why
they fail (and fix them).
> Second, it's almost impossible to make ordinary people collaborate
> using a formal language.
The Semantic Web will be the great testcase for this theory.
(If the businesses can get their grommet-ontologies working
smoothly together, at least that will be a foot in the door.)
> Therefore, one needs automated methods to
> make ontologies simply work, and then interact with the user in a very
> natural and visual way.
The fractal-thicket startingpoint is to enumerate the
'players', eg: person person person place place thing.
> And XML syntax, IMO, hasn't much to do with
> the main problems. In fact, I would see it as one of the obstacles
> itself (being an XML-disliker myself)
What's most freakish about XML, it seems to me, is that it
pretends to be for marking up documents (prose paragraphs)
but it's utterly useless for that, and is instead being
used for databases, where it's absurdly non-optimal.
> I once did an ontology project for net repositories, it contains some
> nice links, an introduction to ontologies and a preliminary ontology
> description language design. You might find it interesting.
> http://borg.cs.bilkent.edu.tr/~exa/ontology/
Yes, very good. (The subpages need 'next subpage' links,
though.)
Jorn Barger
> > Semantic Web project
>
> Never heard of it
If you've heard of XML, I don't understand how you missed
the Semantic Web, which is XML's _raison d'etre_.
[...]
> Do businesses spend money on XML? Sadly, yes. It's mostly wasted
> money, but money gets spent
The goal being to get diverse databases (internal and external)
to talk to each other-- but are you saying the money's wasted
because the goal is never achieved, or because XML is an
unnecessarily awkward (but somewhat adequate) tool?
> Would we hire AI people for XML? Why on earth would we want to do
> that? i don't understand the connection.
For starters, try: Cyc -> Guha -> RDF -> XML
The last I heard, Lenat was trying to market the Cyc
ontology to business as a way to get databases talking to
each other (see above). Additionally, he offered the
knowledgebase as a way of detecting bad data-items (eg
birthyears too recent or too ancient to be valid for a
given class of persons).
> [...] his critique, sponsored by the US air force
> was on the egos in AI and the field's unwillingness to work with other
> fields)
(I'm starting to think you have an identical blindspot of your own,
about semantics...?)
> if i hadn't heard of a semantic web two years ago,
> there's a good chance that it wasn't worth hearing about
The big Scientific American hype-job was exactly two years ago:
http://www.sciam.com/article.cfm?articleID=00048144-10D2-1C70-84A9809EC588EF21
but Tim Berners-Lee had been touting it since 1999 at least.
> [...] And we need to ignore the technical people more
> since the tendency of many (probably most) tech people is to want to
> so something because "it's neat" rather than profitable
I believe this also applies to the W3C, and their XML
juggernaut.
> The ones who understand the business and what the system needs to do
> often aren't visionary (which isn't such a bad thing) and the things
> they do right often go unnoticed and unrewarded.
Wal-Mart supposedly has gotten this right, somehow. (They got
their internal systems talking to each other long before XML,
but they apparently like XML for B2B (external) communication.)
baylor
> the Semantic Web, which is XML's _raison d'etre_.
A good question i guess. Point still remains, i've never heard of it
:)
Is "The Semantic Web" just that semantic web (lower case, not a
project name) that Berners Lee talks about, this pipe dream where, in
the future, every Web page had a contents tag that says what it is so
we can seach and find meaningful data? If so, i guess i've heard of it
but don't much care about it since i can't buy it, can't develop it,
can't deploy it, can't make money on it and won't see it's existence
in my lifetime. It's like EDI and CORBA and a bunch of other things
that came and went without much impact, only much vaguer and much less
useful since it really requires everyone to use it, and standards are
tricky things
> The goal being to get diverse databases (internal and external)
> to talk to each other-- but are you saying the money's wasted
> because the goal is never achieved, or because XML is an
> unnecessarily awkward (but somewhat adequate) tool?
"XML is a watse" in general i can answer, but i don't undestand the
motivation for Semantic Web, so it's hard for me to say anything about
that
Why don't i like XML? A couple of reasons
First, it's nothing special. It's one more text file format, like
comma separated values or fixed length fields or pipe delimited files.
Who needs a book and a couple of conferences on something like that?
Second, XML doesn't solve any problem i can think of. Specifically, it
doesn't solve any problem that isn't already solved. It can store data
in a known format. So does CSV, PS, PDF, Btrieve, BSP, registry, INI
file, etc. It can store data in a heirarchy. So can a couple dozen
other things. It's technology agnostic, except for technologies like
ASCII and Unicode (and for some parsers, the CR/LF issue), which makes
it just like CSV, Sylk, the Mastercard format, etc.
Third, i can't think of a single instance where two business systems
were unable to talk where XML would have helped. Can AOL Instant
Messenger not talk to Microsoft IM? That's because they don't want to
- they refuse to make their data public and wouldn't put it in XML
whether it was an option or not. Can my Web site not talk to my brick
and mortar site? That's because my data and you data *CAN'T* match.
Give you and example from the real world. My B&M site used 7-digit
SKUs. The Web site, which had "deep catalog" (ie, more products) used
an 8-digit SKU (a SKU is the unique ID for a product). The B&M system
couldn't just write the SKU to an XML file and send it to the Web side
- the Web side wouldn't "understand" the data
Is there value in getting systems to talk to each other? Sure. And it
was solved about a billion years ago with EDI. People complain about
EDI, but XML makes it worse, not better. i don't think XML can do
anything EDI couldn't (not sure of that though). XML lacks a common,
everyone-agrees-on-this schema, whereas that was the whole point of
EDI. EDI had a schema. It was huge. Almost as tall as me. Then
everyone talks XML and they act like it's going to replace EDI when
XML refuses to use the EDI schemas and where armies of people are
running around inventing their own "standard" schemas, most notable
OASIS and whatever Microsoft decided to do on their own (big surprise
there)
As a note, i've spent millions and millions on ETL tools like
WebMethods, Mercator, BizTalk and other data mapping tools. Sometimes
they use XML, sometimes they don't. i don't care either way so long as
the results are correct. And, obviously, XML isn't a big help here -
all the work is in data mapping, not transmission formats
> The last I heard, Lenat was trying to market the Cyc
> ontology to business as a way to get databases talking to
> each other (see above). Additionally, he offered the
> knowledgebase as a way of detecting bad data-items (eg
> birthyears too recent or too ancient to be valid for a
> given class of persons).
That doesn't sound like he's selling XML, nor would any AI people need
to be involved for the XML piece. Now, a handful of prolog or logic
people might write the core inference engine that all those facts get
entered into, but you don't need AI people to enter facts or write a
program to dump a table into XML and post it somewhere
But Cyc is hardly a business-kind of thing. Most business people have
never heard of Cyc. For good reason. When we buy info, it's from
people like Bloomberg, Fair-Issac, West, FNMA, Experian, places like
that
> (I'm starting to think you have an identical blindspot of your own,
> about semantics...?)
It's possible
> Wal-Mart supposedly has gotten this right, somehow. (They got
> their internal systems talking to each other long before XML,
> but they apparently like XML for B2B (external) communication.)
i don't know if that's true or not. i've never seen it discussed, and
i read most of the Walmart articles (or did when i was in retail).
Pretty much every company has systems that talk to each other. The
only question is how much work was required to make it happen.
Normally it isn't much work at all. And it certainly *never* needs
XML. You do it with standard data exports and imports (BCP, that sort
of thing). For most large companies, most of the processing is done in
COBOL and JCL on an IBM (or clone) mainframe and all the data is
stored there. The data only leaves the system to populate SOD (start
of day) tables in Oracle or whatever for the little tiny systems to do
some front end work. They then batch up their data and send it back to
the big iron to do the real work
That's how you talk inside a company. Outside, you use EDI. It's
pretty simple
Does Wal-mart use XML for B2B? Maybe. Almost certainly they have a
pilot that uses it (*all* large companies use *all* technologies
somewhere). i've done it. Never on a core system - if you don't use
EDI, you're insane. But on some small, new systems. Did we use XML?
Yeah, maybe. i don't remember since it wasn't important. The important
part is that we didn't use an EDI VAN. The cost in EDI is in the VAN,
not EDI (XML would replace the EDI frame, not the VAN). We use VANs
where it's important, bypass them otherwise because they're really
expensive. But nice
So i know what i do in the business world. What i don't understand is
what "The Semantic Web" is and where it would overlap with the
business world
-baylor
JXStern
exa) wrote:
>It's because there are no "killer-apps" for ontologies. In other
>words, nobody has ably demonstrated that these ontologies actually
>have some hint of intelligence that could be used to improve people's
>lives in a profitable way. ;)
Just to quibble, that there is no need for there to be intelligence in
an ontology, I'm not sure how there could be, and yet it could be an
enabling element. Just this is the issue, perhaps, where the Cyc
project now sits, trapped. You need (a) a problem, (b) an
architecture/theory, (c) domain knowledge, (d) competent
implementation, and the chain is only as strong as the weakest link,
traditionally (c). Note how I leave it distinct from (d). OTOH, no
step is completely distinct from the others.
....
>think there are quite a few obstacles. First thing, I know that nobody
>has the temper to edit an ontology or truly understand what ontology
>is. Second, it's almost impossible to make ordinary people collaborate
>using a formal language. Therefore, one needs automated methods to
>make ontologies simply work, and then interact with the user in a very
>natural and visual way.
....
Well hey now, maybe things are not quite the way you portray them.
Perhaps just what any and every software developer does *is* editing
an ontology, at least in part, but they are using poor tools and
absent appropriate theory. Perhaps just what the semantic web project
*is*, is factoring out that work to make it cleaner and simpler and to
develop the appropriate (semi-)automatic tools to get it done.
So, as they say, you R1, and if you'd like to be a better one, you
might want to keep an eye at least on the semantic web.
</marketing_hype>
Joshua Stern
Frederick Crabbe
> Can someone explain to me why this Google Groups search--
> http://groups.google.com/groups?scoring=d&q=group:comp.ai+semantic.web
> --shows almost no discussion of the Semantic Web project
> on comp.ai, ever? Why isn't it even mentioned in the
> 'FAQ'? What does the academic-AI community think of it,
> in terms of practicality or theoretical interest?
I guess the FAQ question can only be answered by me, and I must admit
that is at least partially due to some of my own biases, although, for
the record, if there is no discussion of the Semantic Web in the
group, it can hardly be thought a Frequently Asked Question.
So to the real answer: I am extremely skeptical of the Semantic Web,
the claims of its proponents, and that anything that could be called a
semantic web will ever exist. The last Semantic Web talk that I sat
through was at IJCAI in Seattle (2001- I think it was Jim Hendler).
My opinion at the time was there were more "and then a miracle occurs"
sections than any talk of that profile I had ever attended. I can
also report that I was not the only one grumbling afterward. My take
is that within the AI community, Semantic Web is not (in its current
form) thought as that useful for AI.
I should note that I have nothing against the basic idea, and concede
that something similar might eventually be a big win, but when I look
at Semantic Web right now, I don't see it.
*That's* why there's no mention in the FAQ.
cheers,
ric
JXStern
wrote:
>> Semantic Web project
>
>Never heard of it.
I can understand that, neither have most people I've ever mentioned it
to. I read the Scientific American article in 1999, laughed, and
forgot about it, until about a year ago when I was solicited to review
a publisher's manuscript about it. I agreed, then googled up
sufficient info to figure out what was up, ... such as it was.
>i considered ignoring this, but i'm guessing no one else on this group
>is going to say anything so i might as well
>
>First, my bias - i was until recently head of AI, R&D and standards of
>a Fortune 500 company.
....
>Do businesses spend money on XML? Sadly, yes. It's mostly wasted
>money, but money gets spent
>
>Do businesses hire AI people? Of course not.
>... i can tell you i've never had a single app
>from a person with a PhD in AI.
Back in the day (circa 1988) you would have at least seen the resumes,
and even now you might trip across some cs degrees that focus on it.
>Would we hire AI people for XML? Why on earth would we want to do
>that? i don't understand the connection. Maybe i'm missing something,
>but to me it sounds similar to "why don't you hire sociologists to
>make mashed potatoes?". i just don't understand the question
As you say in this and the other message, that XML is a nice
(all-ascii) protocol and no more, I quite agree. Yet, I can see that
a place would (in theory) hire AI people who would find themselves
using the protocol, and more semantic ... stuff.
>Have AI schools been properly preparing their students for work in
>business?
....
>Here's a quick critque you'll hear from a lot of people like me who
>purchase products & services - AI (or name your other favorite field)
>doesn't understand business
>
>i'd go one step further and say that the field of computer science
>that some people call AI doesn't actually understand anything useful.
All interesting talk, but it's another discussion. The question is
whether the semantic web is just the thing that might be both AI and
useful. The odds are long, but let's discuss it.
>That's been my experience watching literally every AI project we've
>attempted (all run by PhDs from top 10 AI schools) fail miserably,
....
I'd love to buy you lunch weekly for a year to hear the stories, tho I
have my own to tell as well.
>If you could better explain the question, maybe i could give you
>specifics. i've spent a lot of time over the years giving pretty
>detailed explanations to vendors as to what's wrong with what they
>want to do.
Waste of time, that, comes way too late in the cycle.
> Do you know how many times i've heard a sales pitch from someone
>wanting to sell me a stand-alone lisp-based system ...
Yes, that was me, at one time. While I was quite good at keeping a
straight face, I guess I do owe some people apologies, those who
actually bought the product, especially those who actually unpacked
the boxes and tried to use it!
>There were maybe five people full time and plenty of people in their
>spare time combing the magazines, going to conferences, talking to
>professors, etc and we were inundated by ideas an information, most of
>it worthless. So if i hadn't heard of a semantic web two years ago,
>there's a good chance that it wasn't worth hearing about
Harrumph. I'll address that below.
>In regards to this specific comment:
>> I doubt you can find fifty instances of businesses hiring AI-grads to
>> provide expertise on XML projects in the U.S. in a given year. I base
>> this skepticism not because I have any private (or public) source of
>> relevant data, but because of the widespread trend today to hire very
>> marginally qualified developers for all projects.
>
>First, the choice is not between AI-grads and maginally qualified
>developers. Those two things are hardly mutually exclusive. i've met a
>few AI PhDs i would have hired, but most were exceedingly unimpressive
>and nowhere near the top of the candidate list
>
>Second, is it a widespread trend to hire marginally qualified
>developers? Yup. It's also true that we hire the very best people
>available (well, not always; a lot of HR people screw up and many
>people don't know how to interview people, but the best do tend to get
>hired before the less qualified). Both things are true. Consider that
>for a second
Your point is clear, but I meant it the other way. If you
concentrated on bleeding edge projects, you may have been somewhat
exempt or distant from actual practices, or you may have moved in more
rational circles than I over the past five years. Especially since
the coincident dot.com bust and recession and outsourcing craze, the
hiring process has gone to ... but that's another topic, except
insofar as career prospects in AI are relevant to the newsgroup. I
shall in any case leave it here.
>Here's another personal opinion - the IT industry is run worse than
>any other industry.
Also OT, and I agree insofar as it makes sense to talk about an IT
industry as opposed to IT practices within business organizations, but
just whose fault is it exactly? Too bad it's OT or we could both say
so much more ...
> While i might make my living at it, i personally
>think companies spend FAR too much on computers and pay far too much
>attention to the cutting edge.
Aha. Well, I might argue that they spend far too little, but they
would obviously have to do it much more smartly than they have in the
past. I suppose we could swap stories of ghastly stupid expenditures
and strategies as business organizations try to move into new
technologies, even well behind the leading edge. I'm reminded of some
early relational database stories ... but the failure was not the
fault of the technology, as such, which was solid and did, eventually,
get adopted and used somewhat correctly.
> We need better
>managers to decide what our priorities are and to apply the technology
>more intelligently.
Quite.
> And we need to ignore the technical people more
>since the tendency of many (probably most) tech people is to want to
>so something because "it's neat" rather than profitable
While this is far from unknown, one could level equal or greater
tendencies to failure to nontechnicals -- that they want to dig the
Panama Canal with a technological teaspoon, that they won't let the
specs for a project be fixed, that they refuse to take on minor
projects that would save major money because they've just never done
it that way, etc. I think the answer in the book is that we should
all work together and make our best contributions and respect all
parties, as difficult and rare as this is to see practiced in the real
world. But knocking technical people especially is a sure recipe for
failure. In my humble opinion.
>No department in the companies i've worked for is run perfectly, but
>some are run pretty well. IT almost never is. Not when you look at how
>consistently we miss deadlines, go over budget, deliver the wrong
>functionality, approve projects of no value and create things that are
>lacking in quality.
Yes, well, see my previous point about hiring practices, and then
watch me shrug.
> If high school dropouts can put up a house for a
>fixed cost in a specified period of time and be profitable, and if
>Hollywood can create movies that have far more variables and
>creativity than your average IT system and still come close to hitting
>deadlines most of the time, why can't a bunch of PhDs and MBAs manage
>to create a cash register program within 100% of the estimated budget
>and deadline?
I propose the answer may involve a valuable insight into the real
nature of computing and software: they're hard problems. OTOH, we
might have to seek some clarification on the "creativity", much less
the on-budget performance (much less the initial size of those
budgets!) regarding the Hollywood analogy ... one of my stories.
>Well, there are reasons, but they take us off topic. The point i want
>to make is that, when you need to hire 1,000 or 3,000 or 7,000
>computer people to build simple systems, guess what, you sometimes
>hire people who aren't going to single handedly change the world. For
>different reasons. The energetic, "smart" ones are often arrogant and
>undisciplined and thus create useless garbage.
....
Managers at MacDonalds have the same problem and they get the
hamburgers out. Ref. your point about IT management and my point
about hiring the unqualified. What if, now, virtually every software
system *is* a project to change the world?
> If i thought i'd make
>money off something you were doing, i'd cut you a check for a couple
>million no problem. My peers have done it time and again with people
>(AI and non-) who didn't have a chance in hell of delivering what they
>were promising. i'm a bit more of a hardnose, but the money's out
>there. But it's investment money, not grant or charity money
But you see, this is the issue that I see central to Barger's original
post. Business wants to buy some magic beans (those recent IBM ads
are excellent in those regards, tho I doubt many business execs
actually get the jokes). They are unwilling and unable to track a new
technology, evaluate it reasonably, do some trial applications, even
contribute to its development, and thus gain the benefits ahead of
their lazier competitors. That's all a long-term project, with a high
risk-reward ratio, and high-risk projects are way out of favor in IT
today, as I suppose you already know.
Now, that's not to say that any past, present, or future AI ideas are
guaranteed winners, and indeed, most of them were pretty much
guaranteed losers from before the beginning. Back in the day I
watched plenty of companies put their money into AI, with the poor
controls and lack of success you describe. Should any business today
pour a couple of million bucks a year into the semantic web?
Surely not. However, one would expect a little more low-level
activity out there, than there seems to be. OTOH, people are burned
out. And I daresay the actual products that shops like Microsoft dump
out there to show the value of XML and ontologies and whatnot, only
throw more cold water on the situation. Yet, corporate early adopters
to the web were generally eventual winners, and one could cite that as
an example of why one would want to be involved with this next-gen
version of the same.
OTOH, Barger's question was not only to business enterprises, but to
the academy as well, where the risk/reward ratio is computed entirely
differently. I find it interesting (excuse me if I use this
noncommital word too much here, I'm typing as fast as I can!), because
I have a little nascent theory of my own about the place in theory
that these various components have, and I suggest that ontology work
simply has no place in the current academic structure, that the
balance needed to bring AI technologies into beneficial use depend on
a kind of ontological development (and my specific view on this may be
different than Barger's btw) which is being ignored.
IOW, just maybe some aspects of the semantic web are that missing
piece of the puzzle that would let AI-ish technologies finally earn
their keep. Sure, you can make this argument about any theory(x),
but, well, I guess I just did, so there!
Joshua Stern
Jorn Barger
types, but if semantic-AI has any future, then business
semantics like this is where it must start, imho.]
baylor <baylor@no_spam.ihatebaylor.com> wrote in message news:<b9nen9$imn$1...@mulga.cs.mu.OZ.AU>...
> Is "The Semantic Web" just that semantic web (lower case, not a
> project name) that Berners Lee talks about, this pipe dream where, in
> the future, every Web page had a contents tag that says what it is so
> we can seach and find meaningful data?
Alas, they're determined to put the tags on bits of text
*within* documents rather than in the document header.
I blame this on "Goldfarb's conjecture" in the earliest
days of SGML, which hypothesized that formatting-markup
could be profitably moved outside the document, replaced
by semantic or structural markup.
But the examples offered by the Semantic-Web crowd are
things like book-titles and -authors, addresses, names
and dates, product-info, etc. And they never seem to
notice that these are almost never associated with
distinctive formatting-styles, so embedded semantic
tags are not an efficient solution.
> [...] It's like EDI and CORBA and a bunch of other things
> that came and went without much impact
I don't know anything about EDI (Electronic Data
Interchange), but isn't it widely-used? Does it
recommend any sort of semantics/ontology?
[re XML]
> First, it's nothing special. It's one more text file format, like
> comma separated values or fixed length fields or pipe delimited files.
> Who needs a book and a couple of conferences on something like that?
A thousand books and 100 conferences, I think. It
may be the document-oriented applications that make
it so complicated...?
> [...] It's technology agnostic, except for technologies like
> ASCII and Unicode (and for some parsers, the CR/LF issue), which makes
> it just like CSV, Sylk, the Mastercard format, etc.
CSV = comma-separated values
Sylk = Microsoft spreadsheet (Symbolic Link) format
Mastercard format = CDF or Common Data Format?
On first glance, none of these look interested in
establishing semantic standards, which is what XML
is all about.
> Third, i can't think of a single instance where two business systems
> were unable to talk where XML would have helped.
Scary supporting anecdote: Google started offering
search-results in XML, and people found it harder
to use than the standard HTML format (iirc).
> [...] EDI had a schema. It was huge. Almost as tall as me. Then
> everyone talks XML and they act like it's going to replace EDI when
> XML refuses to use the EDI schemas and where armies of people are
> running around inventing their own "standard" schemas, most notable
> OASIS and whatever Microsoft decided to do on their own (big surprise
> there)
Google hints that this is called 'X12'? Here's a page
on X12 semantics:
http://www.rawlinsecconsulting.com/x12tutorial/x12sem.html
Again on 1st glance it looks like it kludges some
very elementary semantic standards (eg 'location')
which might explain why the XML crowd thinks they
_must_ re-invent this wheel (but not why they think
they _can_).
> As a note, i've spent millions and millions on ETL tools like
> WebMethods, Mercator, BizTalk and other data mapping tools.
ETL = extraction, transformation, and loading
http://www.sas.com/technologies/dw/etl/index.html
"extract data from different locations,
transform raw operational data into consistent,
high-quality business data, and
load the data into a data warehouse"
Again I'm most interested to know if there are any
semantic standards in this.
> > The last I heard, Lenat was trying to market the Cyc
> > ontology to business as a way to get databases talking to
> > each other (see above). Additionally, he offered the
> > knowledgebase as a way of detecting bad data-items (eg
> > birthyears too recent or too ancient to be valid for a
> > given class of persons).
>
> That doesn't sound like he's selling XML,
If he's pursuing the path of least resistance, XML
is it, I think.
> nor would any AI people need
> to be involved for the XML piece.
If the Cyc end worked out of the box, no. (Heh!)
> But Cyc is hardly a business-kind of thing. Most business people have
> never heard of Cyc.
Cyc is a GOFAI thing, and cash-starved, so it's
hoping to find a foothold in the business world,
and just because it hasn't caught on doesn't
mean it won't ever.
> For good reason. When we buy info, it's from
> people like Bloomberg, Fair-Issac, West, FNMA, Experian, places like
> that
(Any semantic standards there?)
> [...] The important
> part is that we didn't use an EDI VAN.
VAN = value-added network
At 1st glance, any proprietary alternative to the
Internet for B2B data-exchange, that often includes
expertise in EDI 'translation' (so the need for this extra
translation is presumably what XML hopes to finesse).
Eray Ozkural exa
I must say this is one of the few moments I have heard intelligible
comments on technology from a prominent manager.
baylor <baylor@no_spam.ihatebaylor.com> wrote in message news:<b9nen9$imn$1...@mulga.cs.mu.OZ.AU>...
> Why don't i like XML? A couple of reasons
>
> First, it's nothing special. It's one more text file format, like
> comma separated values or fixed length fields or pipe delimited files.
> Who needs a book and a couple of conferences on something like that?
>
Good point. XML is simply a text file format with a certain (tree)
syntax. It was claimed by proponents of XML that "nobody would ever
have to write a syntactic analyzer again". However, that is a moot
point. First, it's easy to write a syntactic analyzer. Writing a
*semantic* analyzer is hard. Second, that verbose, ugly and mostly
inadequate syntax is suitable for only a very small fraction of
languages imaginable. In particular, it is not suitable for any
language design that should be readable or writable. BTW, I think
those books and conferences are largely caused by people's desire to
invent new things to make a living from.
> Second, XML doesn't solve any problem i can think of. Specifically, it
> doesn't solve any problem that isn't already solved. It can store data
> in a known format. So does CSV, PS, PDF, Btrieve, BSP, registry, INI
> file, etc. It can store data in a heirarchy. So can a couple dozen
> other things. It's technology agnostic, except for technologies like
> ASCII and Unicode (and for some parsers, the CR/LF issue), which makes
> it just like CSV, Sylk, the Mastercard format, etc.
>
As a matter of fact, in vast quantities of data "a forced" format as
XML can only generate overhead. This is simply because any structure
should be stored properly in an efficient data structure with fast
query and update algorithms instead of a dumb text file. I will not
mention the usually unjustified overhead of using XML and DOM where
not necessary.
> Third, i can't think of a single instance where two business systems
> were unable to talk where XML would have helped. Can AOL Instant
> Messenger not talk to Microsoft IM? That's because they don't want to
[snip]
XML and a computational ontology can help here, but I'm not sure if
the additional effort required to get it running is justified in a lot
of scenarios.
What could really help businesses would be letting an AI handle some
of their tasks. I have never thought what could help businesses
thoroughly but I think one could develop weak AI programs that could
help HCI and automation for home and small office users. Taking
personal assistance to a new level. Such a project might actually be
worth your investment.
Regards,
__
Eray Ozkural
Eray Ozkural exa
> Well hey now, maybe things are not quite the way you portray them.
> Perhaps just what any and every software developer does *is* editing
> an ontology, at least in part, but they are using poor tools and
> absent appropriate theory. Perhaps just what the semantic web project
> *is*, is factoring out that work to make it cleaner and simpler and to
> develop the appropriate (semi-)automatic tools to get it done.
>
I suggest you skim the parts about general ontology and the design of
my description language in the site I referred to at URI
http://borg.cs.bilkent.edu.tr/~exa/ontology/
It's actually very brief (by design) and answers some of these
concerns.
What I'm saying there is that "ontology of code" is quite different
from general purpose ontology. In fact, it's a LOT easier because it
has such different purposes (despite what your favorite PDL manual
might say) For instance, in code there is no such thing as "sortal
predicates" or other sophisticated kinds of relations that exist in
real life.
The problem is although ontology of code is a simple thing, most
programmers, who are supposed to be familiar with formal languages and
systems, can't get it right. Now will you tell me how the web users
are supposed to deal with a more complex problem?
And if you have seen a slightly large computational ontology, I think
you will appreciate what I mean.
In retrospect, I think such representation languages (like my design)
might even be redundant. It's the AI behind knowledge that matters.
[*] That is what I meant by "automated tools".
Regards,
[*] "Exchange of knowledge" that's a bold claim but I doubt we have
the correct theory of knowledge in the first place. I'm quite
skeptical about the use of a subset of FOL for such purposes. I think
it doesn't matter whether knowledge is in any way digestible by
humans, so the whole enterprise might be shortsighted.
__
Eray Ozkural
Eray Ozkural exa
> > http://borg.cs.bilkent.edu.tr/~exa/ontology/
>
> Yes, very good. (The subpages need 'next subpage' links,
> though.)
Thanks for the comments and suggestion. I haven't updated that site
for a long time. The idea of DL there is simple but I think it could
find some uses especially in repositories where thousands of instances
are gathered and a modular ontology is required. It only needs to be
implemented ;)
Regards,
__
Eray Ozkural
JXStern
exa) wrote:
>JXStern <JXSternC...@gte.net> wrote in message news:<b9pc5j$5g5$1...@mulga.cs.mu.OZ.AU>...
>> Well hey now, maybe things are not quite the way you portray them.
>> Perhaps just what any and every software developer does *is* editing
>> an ontology, at least in part, but they are using poor tools and
>> absent appropriate theory. Perhaps just what the semantic web project
>> *is*, is factoring out that work to make it cleaner and simpler and to
>> develop the appropriate (semi-)automatic tools to get it done.
>
>I suggest you skim the parts about general ontology and the design of
>my description language in the site I referred to at URI
>http://borg.cs.bilkent.edu.tr/~exa/ontology/
>It's actually very brief (by design) and answers some of these
>concerns.
http://borg.cs.bilkent.edu.tr/~exa/ontology/html/problem.html
"One obvious solution is to make computers as smart as humans and tell
them to continuously organize the web efficiently"
I do not know that humans organize knowledge efficiently. I know that
they access it sufficiently, and that the search process is
effectively massively parallel. It's all vague beyond that. I doubt
that any two people organize all their knowledge in exactly the same
way.
http://borg.cs.bilkent.edu.tr/~exa/ontology/html/background.html
"In AI, ontology takes on a cognitive perspective"
I like this statement. Ontology for its own sake seems to me an
impossible and obsolete metaphysical project.
"This is achieved by an abstract theory of knowledge which formalizes
the knowledge in terms of objects, classes, relations, functions and
axioms."
I dislike this one. I hold that any ontology is a specific theory,
not an abstract one. This is a complex point.
>What I'm saying there is that "ontology of code" is quite different
>from general purpose ontology. In fact, it's a LOT easier because it
>has such different purposes (despite what your favorite PDL manual
>might say) For instance, in code there is no such thing as "sortal
>predicates" or other sophisticated kinds of relations that exist in
>real life.
Rather than argue metaphysics, let me ask if this has any implications
for whether the semantic web (SW) project can or cannot work, and
whether it is or is not related to AI? If anything, I would read this
as being on the side of some kind of SW.
>The problem is although ontology of code is a simple thing, most
>programmers, who are supposed to be familiar with formal languages and
>systems, can't get it right. Now will you tell me how the web users
>are supposed to deal with a more complex problem?
Same way they deal with reality, yes? Surely that's more challenging!
>And if you have seen a slightly large computational ontology, I think
>you will appreciate what I mean.
I think you have missed my point, that any program must have
ontological aspects, implicit or explicit.
I understand that explicit ontology projects quickly lose their
bearings and collapse under their own weight. So does any mundane
program that lacks a design and architecture, though, so I cannot
blame it on the nature of just being an ontology.
>In retrospect, I think such representation languages (like my design)
>might even be redundant. It's the AI behind knowledge that matters.
>[*] That is what I meant by "automated tools".
While Barger did come to an AI newsgroup to ask about interest in the
semantic web, it also seems that his interest is in applying some kind
of semantics to mundane applications. So, to say "it's the AI behind
knowledge" seems to me possibly irrelevant -- though I think Barger
may be more on your side on this matter.
Joshua Stern
baylor
> have my own to tell as well.
i live in Minneapolis and am easily bribed with food
i thought i was going to IJCAI in August, but realized my schedule
won't allow it. Bummer. But maybe we'll run into each other and swap
silly corporate stories
> >If you could better explain the question, maybe i could give you
> >specifics. i've spent a lot of time over the years giving pretty
> >detailed explanations to vendors as to what's wrong with what they
> >want to do.
>
> Waste of time, that, comes way too late in the cycle.
Yup. i normally state requirements, the sales guy and sales engineer
ignore them (oh how they ignore them), i decide not to toss them a
half million and suddenly i have a dozen engineers and regional
managers wanting to know what's wrong. So i give them the test cases i
used and the scorecards and the post audits that explain in black and
white what features needed to be there and the tests they failed. And
they thank me and i never hear from the again. Well, not always,
Sometimes the comments are taken in and used. But more often than not,
they're not
> > Do you know how many times i've heard a sales pitch from someone
> >wanting to sell me a stand-alone lisp-based system ...
>
> Yes, that was me, at one time. While I was quite good at keeping a
> straight face, I guess I do owe some people apologies
Shame on you, shame :)
i had a WebLogic sales engineer tell me that they hardcoded passwords
in plain text in the startup INI and couldn't change them without
taking down the server because "our customers don't consider security
very important"
i'll give you the benefit of the doubt and assume you were better at
sales than that :)
> Your point is clear, but I meant it the other way. If you
> concentrated on bleeding edge projects, you may have been somewhat
> exempt or distant from actual practices, or you may have moved in more
> rational circles than I over the past five years
i was responsible for all projcets by our 1,000 developers, including
help teach the class the IT people had to go through and grading their
homework (odd job i had). And i can assure you that while our tech
people often didn't know how the business worked and weren't too up on
life cycle issues, many (not all, but maybe most) managers were
just... Oh, what's a stronger word than irrational, myopic and prone
to hair-trigger opinions. It's not that they didn't understand the
tech people - one of the best managers i knew was a sales guy who was
being rotated through IT. But many managers didn't ask for options and
didn't think critically. Never a good sign
> Especially since the coincident dot.com bust and recession and outsourcing craze
The dot com days were insane (can you breathe? Here's a job!), the Y2K
days were silly, and i don't know if outsourcing is a craze right now
but i know my old company is firing all of their tech people except
the architects and the project managers. The rest is headed to India
or, in some cases, Russia. Which i wouldn't be that disapproving of if
our pilots had gone well. But our outsourcing pilots were disasters.
So we decided to double down. It's been a nightmare in so many ways...
> Also OT, and I agree insofar as it makes sense to talk about an IT
> industry as opposed to IT practices within business organizations, but
> just whose fault is it exactly? Too bad it's OT or we could both say
> so much more ...
Whose fault is it? My opinion: the problem is with people failing to
hold those under them responsible. The manager should know what he
wants and hold his people to that (i was going to say he should know
what's possible, but i take that back and believe only that he should
ask for more options so he or she can make a decision). The VP should
hold the managers accountable. The CEO should hold the CIO
accountable. The share holders the copmany, etc.
And what does this have to do with AI? i remember reading somewhere a
big debate on the value of AI competitions and benchmarks and how many
AI people (and assumedly people in all fields) would tailor their
problem to the solution they had rather than vice versa, meaning you
couldn't compare apples to apples. A comment was made that setting
specific, measurable, comparable, meaningful goals and standards
forced people to search for solutions to problems rather than problems
for their solutions and that the field had advanced as a result. Which
is my experience with business. Set something measurable and objective
and people tend to get better. Let them wander in the desert and,
well, you get what you get
OK, maybe that doesn't completely bring it on topic, but it was worth
a try :)
> Aha. Well, I might argue that they spend far too little, but they
> would obviously have to do it much more smartly than they have in the
> past
Being smarter in how they handle these things is certainly what i
consider the key. i worked for a company flush with cash once. It
should have been nice, but we were actually less successful with the
rampant spending. And our .com side, the business model was *never*
projected to be successful but did we hand them money to burn... Who
spends $500m on a Web site projected to do $10m annual _revenue_. Oh
right, we did :)
Going back, spend more money on the projects that will succeed, cut
the ones that have no actual vision or goal, ask for a cost-benefit
analysis and set specific goals and hold people to them (we've done
the former but not the latter; accountability in the companies i've
worked in was never enforced except on the unpopular managers)
Part of my job was as the internal AI evangelist. i talked a lot on
what could be done if we just thought differently about how we did
business. What if we had a more intelligent, solution-focused Web
site. What if we could let you ask "what's the difference between
these products?" or reformat the product data to say "get XYZ for just
$50 more" (product confusion was a huge issue for us). What if you
could tell us what you wanted the product for and we could recommend
the best thing for you? When what you wanted sold out, what if we
could recommend what to do next (go 5 miles to the next store, wait 3
days for the next shipment, suggest an alternate product, etc.)? Or if
we noticed that you bought some products (say a PC and a martial arts
video game) we recommended related products in other categories (a
martial arts movie and maybe alt metal CD)? What if we could look at
what you bought at checkout and suggest an extra item at a discount
(ie, bundle pricing)? It was a lot of simple stuff you could pull off
with some data extraction and reformatting, a little nearest neighbor,
a bit of model-based recommendation and a 3rd party collaborative
filtering product. It wasn't expensive but it was worth investing in.
But it required us to think about things differently
In other words, in our case, we didn't need more money (i could have
funded all of the proposed projects except maybe the CF stuff with the
money we would have saved axing the $20m constraint-based reasoning
recommendation engine project), we needed to spend what money we had
better
But that requires a bit better management and, hey, when you're
staffing a company with 5,000 computer workers and 200,000 non-IT
people, you try to do the best with who you have rather than hire the
brightest 20 people. Hiring bright people is for small companies. It's
a pain to pull off when you're hiring thousands. Which is when you
need better processes and a better leader
> I'm reminded of some early relational database stories ...
> but the failure was not the fault of the technology
We had plenty (plenty!) of failures with things like UML and objects
and Java. Sometimes the problem was with the technology (actually, it
often was), but it was also very common that we bought something
because it got a good review and we just used it badly
For example, we purchased a not-very-good call center app. We wanted
more functionality, so the seasoned tech lead decided to modify the
core code of the product. The next version of the product came, we
upgraded and suddenly all of our "enhancements" were gone. The manager
told his boss the vendor was to blame, but come on...
We've also been burned by poor product functionality, but it was often
in features that we _knew_ we needed and could have easily tested. But
we didn't. Whose fault is that? Ours of course. And while i'm no big
fan of Microsoft, is it really their fault if someone decided to build
a 100GB database in Access and it's not working well?
As to failures of early RDBMSes, i would like to believe that i would
never purchase a product until it had gone through its growing pains
(or unless we could really justify absorbing the risk, as in the
potential payoff was huge). We once spent 3 years writing a system for
an in-development DEC platform that DEC eventually decided not to
release. We are such geniuses...
> I think the answer in the book is that we should
> all work together and make our best contributions and respect all
> parties, as difficult and rare as this is to see practiced in the real
> world
i've actually seen this happen. Small company (~500 people), stable
production system, the working relationship between the business
people and the IT developers who enhanced the system was wonderful.
And it was almost entirely based on a)these people knew each other and
b)the IT people had worked on this one system for years. They had
gotten past the "brand new everything, my consultant doesn't
understand you business" phase and got to the part where everyone got
to know each other. The people who had to *support* the system were
much better at thinking through the support implications of decisions
(to them, "neat" technology was something that kept them from coming
in over the weekend to fix an AbEnd) and, for whatever reason, they
all understood and cared about and were kinda excited about the
business end of things
But in the world of .com and Y2K migrations and system overhauls and
"what does our business do again?" and "my project is so large, who
are you again?", this didn't happen nearly enough. i was plagued with
"achitects" who felt it would be cool and "better" to write their own
search engines, random number generators, shopping carts, load
balancers and the like. Oh man, the pain...
AI does whatever AI does. That's fine in the sense that it doesn't
affect me (well, i guess it does now that i'm in PhD school and
looking to enter academia, a place i don't really know or understand).
But if AI people want people like me to give them money, they are less
likely to get it romancing me with BS smoke and mirror terms like
situation calculus, non-monotonic reasoning and collaborative auctions
than if they can tell me why anything they did would make my life
better. It's like the contractor who wants to build your house showing
up and saying, "hey, look at my toolbox, this is a hammer, this is a
screwdriver, these are pliers, are you impressed yet?"
AI has generally done better when it's sold as "this manufacturing
system will save you money, and by the way, it has a planning system"
rather than "buy AI tools, we use Rete!". i've noticed the AI
pure-plays didn't seem to get very big. Too much focus on tools, not
enough on solutions. And as someone posted on this group a few weeks
back, to build and sell real-world systems that have AI in them,
you're going to spend 5% of your time on AI code and 95% on non-AI
work. People like me don't buy AI, we buy solutions and just really
don't care if the solutions came from statisticians, OR guys or AI
ones
So on Semantic Web, if it's something i'm supposed to be buying,
someone needs to tell me why. If it's a concept like CORBA or DCE,
someone's gotta tell me why i should be looking at it. More
importantly, for companies like mine that are retailers or banks or
manufacturing companies but _not_ software companies, i need to know
_when_ to start paying attention. Ideas don't grab me. People close to
bringing real (non-toy) solutions to life grab me
> I propose the answer may involve a valuable insight into the real
> nature of computing and software: they're hard problems
You know, i hear that, but i don't think it's true. As a manager, i'll
give you a list of things i want. Some will be hard, but a whole lot
will be easy. In the last too-many years, something like 80% of the
systems i've seen were variations on cash registers - a collection of
line items on which we perform some action (ring up sales, track
inventory, dispatch repair people, etc.). The number of basic data
entry screens we create relative to the number of scheduling systems
is really quite large. And maybe it was hard to write that first
customer lookup screen, but we've been writing the same systems over
and over. It ought to get easier with practice. We ought to have
learned some things. And sometimes that's true, but often it's not,
especially when everything has to be in the latest and greatest
language (what, you can't do that in COBOL, you need to do it
Smalltalk, now C++, now PowerBuilder, now Visual Basic, now ASP, now
Java, etc.). One of the best systems i ever saw was a screen builder
where you just typed in the names of the fields and it generated the
screens. Change a field in the DB and the screen(s) changed. It was
written by a guy frustrated with building the same type of screens
over and over
That's why i'm decently excited about that whole field of design
patterns. A friend of mine (also getting his AI PhD) is hoping to
bring the concept of reusable design patterns to AI. We don't know
what that means yet, but it sounds good :)
> Business wants to buy some magic beans (those recent IBM ads
> are excellent in those regards, tho I doubt many business execs
> actually get the jokes). They are unwilling and unable to track a new
> technology, evaluate it reasonably, do some trial applications, even
> contribute to its development, and thus gain the benefits ahead of
> their lazier competitors
This is soooo true
> Now, that's not to say that any past, present, or future AI ideas are
> guaranteed winners, and indeed, most of them were pretty much
> guaranteed losers from before the beginning. Back in the day I
> watched plenty of companies put their money into AI, with the poor
> controls and lack of success you describe. Should any business today
> pour a couple of million bucks a year into the semantic web?
When i look at companies and products, what i ask is "do these people
have any idea what it is we do? Do these people understand the type of
problems we face? Do these people know what it means to have a
production system and all that comes with the care and feeding of a
running system?". A lot of ideas are poor because the people building
them don't get what we do. For example, STRIPS is cute, and some
people even extended STRIPS to allow conditional operators and
negation in Effect clauses, but i have a lot of real-world issues i
have to deal with. Steps can be done in parallel but i only have 4
people who can do it. And certain steps can only be done by certain
people. And one of those people is on vacation in June. And i have a
calendar date i have to hit and need to know how many more people i
need to hire to hit that date. And it requires me to use one of 3
machines whose availability is limited (and which are shared with
another department). And these five specific functions need to be done
by this date for the first phase audit. Etc etc. i realize some real
world planning systems can do this (i assume; the intelligence aspect
is lacking from main stream project management tools like Microsoft
Project), but it wasn't in the early planning stuff and so wouldn't
have been of interest to me back then
As for the semantic web, i don't know enough about it to say why i
should care. Gartner hasn't mentioned it, it's not talked about at
TechEd, it's not in InfoWeek and i've got a stack of papers 4" thick
on my desk of things i'm supposed to be evaluating. Maybe once i
figure out if any of these COM/RMI bridges actually work, how to model
system performance and find bottlenecks in a scaled back integration
environment, what the switch from thread to process orientation in
IIS6 means, how to monitor .NET applications, how to gather real time
system performance numbers on a legacy ActiveX-based application,
whether there's an easy way to build a self-service Web based
performance tuning expert system and how to properly structure tens of
thousands of business rules and data objects to be able to find and
recommend the best home loan and interest rate in under 5 seconds,
maybe then i'll look into this semantic web thingy. If i'm not
conducting a system audit or trying to improve our "processes" (such
as they are)
Which gets back to the point of the original thread. Why aren't
business people paying more attention to ontologies? First, because
you used the word "ontology" (vocabulary seems to be a big problem for
AI people; here's a hint: if you want a normal person's attention,
don't use words like ontology, semantics, canonical, simulated
annealing, wavelets or anything that would make your mom go "huh, what
are you talking about"), and my mind turns off when people refuse to
speak English (or worse, show me formulas with lots of arbitrary Greek
letters). Second, because we have just sooooo much we have to look at.
There's not time for everything and we need to make triage decisions
quickly. Nothing personal, but you have no idea how many things get
tossed at me and the other architects/R&D staff, managers, team leads,
etc.
So for someone in the business world to notice semantic Webs and
ontologies and stuff, i hate to say it, but it really needs a better
marketing arm and some concrete examples
> And I daresay the actual products that shops like Microsoft dump
> out there to show the value of XML and ontologies and whatnot, only
> throw more cold water on the situation.
:)
> OTOH, Barger's question was not only to business enterprises, but to
> the academy as well, where the risk/reward ratio is computed entirely
> differently
Yup, but i have no information for/knowledge of academics. i don't
know who would study this stuff, but one assumes it takes time away
from other projects people are knee deep in. Which is fine if the big
funds are there or the idea is really exciting. But my limited
experience with my professors shows that they seem pretty darn busy
Again, without knowing what specifically the Semantic Web is and what
people are predicting for it's chances of success and value, it's hard
for me to say. Not that ignorance has ever shut me up before :)
-baylor
JXStern
wrote:
>But that requires a bit better management and, hey, when you're
>staffing a company with 5,000 computer workers and 200,000 non-IT
>people, you try to do the best with who you have rather than hire the
>brightest 20 people. Hiring bright people is for small companies. It's
>a pain to pull off when you're hiring thousands. Which is when you
>need better processes and a better leader
Oh, I agree, and go further -- big companies hate to hire smart
people! It's a rational decision, of sorts. Say that innovation
depends on smart people. An innovative project might make the company
millions of dollars more productive. Or, a screwup might cost the
company billions. Innovation being risky, they go with the expected
value, and hire people who (they believe, more or less accurately)
won't rock the boat. Problem is they also fail at even simple
projects. Hence my comment about the preference for minimally
qualified (eg, cheap) employees. Where that leaves the market for AI
people in general business employment, I leave you to evaluate.
>> I propose the answer may involve a valuable insight into the real
>> nature of computing and software: they're hard problems
>
>You know, i hear that, but i don't think it's true. As a manager, i'll
>give you a list of things i want.
....
>That's why i'm decently excited about that whole field of design
>patterns.
Programming is easy, it's the world that is difficult.
And managers are difficult, too. That manager's list is unlikely to
be coherent and complete, but in modern IT environments, it is usually
taken as sacrosanct. This is a big change from even ten years ago, in
my experience. One thing the AI people discovered in the 1980s, is
that even an expert's best-effort view of his own domain tends to be
incomplete, let us say. You don't want to be your own lawyer, you
shouldn't be your own knowledge engineer, either -- not even for
mundane software projects, btw. The world is too difficult, and there
are necessities, conventions, and technicalities of all kinds related
to building a computational entity in the world.
Do design patterns address this? Perhaps they address the programming
side, and maybe the world side, just a bit. So, guess what we need to
address the world side?
(pssst -- the semantic web!)
Now, if we only actually *had* it!
Joshua Stern
Tim Finin
> Can someone explain to me why this Google Groups search--
> http://groups.google.com/groups?scoring=d&q=group:comp.ai+semantic.web
> --shows almost no discussion of the Semantic Web project
> on comp.ai, ever? Why isn't it even mentioned in the
> 'FAQ'? What does the academic-AI community think of it,
The semantic web community is drawing people from a number
Computer Science subareas, with AI, web technology, and databases
being the main ones, I suppose. There are a lot of AI people
who are very active in semantic web research, mostly those with
an interest in knowledge representation and reasoning. Take
a look at some of the projects mentioned at http://www.daml.org/
or http://semanticweb.org/ or the papers at the 2002 semantic web
conference (http://iswc.semanticweb.org/). You will see a lot
of the usual AI suspects from places like (in the US) Stanford,
CMU, SRI, ISI, Yale, BBN, Maryland, etc.
Jorn Barger
[...]
> I have a little nascent theory of my own about the place in theory
> that these various components have, and I suggest that ontology work
> simply has no place in the current academic structure, that the
> balance needed to bring AI technologies into beneficial use depend on
> a kind of ontological development (and my specific view on this may be
> different than Barger's btw) which is being ignored.
I'm guessing the view you're distancing yourself from would
include this earlier comment of mine:
> If AI-schools took semantics seriously (as they should), a
> minimal curriculum would have to look, in effect, at each of
> Roget's categories and how that class of words/concepts has
> been or might be represented in silicon.
> A history of ontologies since Aristotle would be useful, along
> with the evolving toolkit of semantic nets. I've made a start:
> http://www.robotwisdom.com/ai/timeline/0000.html
This Roget-centric perspective has seemed intuitively
_necessary_ to me since highschool, but I'm painfully
aware that it's anything but obvious to the academic AI
community. My working hypothesis is that science-types
have a sort of Freudian blindspot about the language of
human emotion and motivation, which constitutes a vast
part of Roget's system. But I can't imagine any
realistic alternative curriculum...?
(Back c1976 I ran across a book <Googles> which must
have been "Language and Perception" by Johnson-Laird
and Miller, that took certain Roget-style categories
and tried to do something AI-like with them...
Obviously the intention impressed me, even if the
execution didn't.)
Jorn Barger
> The semantic web community is drawing people from a number
> Computer Science subareas, with AI, web technology, and databases
> being the main ones, I suppose. There are a lot of AI people
> who are very active in semantic web research, mostly those with
> an interest in knowledge representation and reasoning. Take
> a look at some of the projects mentioned at http://www.daml.org/
I've been web-surfing pretty continually since 1997, and
reading netnews since 1989, and I've developed an
instinctive _shudder_ whenever I'm offered a URL like this,
that's obviously just the homepage... because it usually
means I'm doomed to waste many false steps trying to find
the actual buried content there.
You recommend "projects" but it's not at all obvious which
of the ~50 links on the homepage count as projects-- I
guessed "Publications" which got me a scraggly little
runt of a page with two sub-choices: WebScripter report or
citation list. The first included no links, the second
has ~100 links in no evident order with no descriptions
except the titles: http://www.daml.org/publications/cite.html
The Ontology-Library link gives me another list of non-
evident subchoices, the first few of which were useless
'gotchas'. The 'keyword' subchoice at least gave me a
hint of what's being treated, but I'm not willing to
invest exploration time without better hinting:
http://www.daml.org/ontologies/keyword.html
The 'roadmaps' link includes a semantic-web-researcher
roadmap that throws me back to the basic 'About Daml'
page-- http://www.daml.org/about.html --which consists
of a very abstract one-paragraph summary and a half-
dozen heterogenous links. Trying the first couple, I
get more borderline-unreadable ineptly-formatted slaps-
in-the-face... so I give up.
The Web is supposed to be about communicating with
humans. The XML crowd thinks human-communications
should be secondary to machine-communications, and
consequently views webpages as databases.
This approach is appalling, and doomed. (If you want
me to keep trying, give me one URL with appropriate,m
clear content.)
JXStern
>> I have a little nascent theory of my own about the place in theory
>> that these various components have, and I suggest that ontology work
>> simply has no place in the current academic structure, that the
>> balance needed to bring AI technologies into beneficial use depend on
>> a kind of ontological development (and my specific view on this may be
>> different than Barger's btw) which is being ignored.
>
>I'm guessing the view you're distancing yourself from would
>include this earlier comment of mine:
>
>> If AI-schools took semantics seriously (as they should), a
>> minimal curriculum would have to look, in effect, at each of
>> Roget's categories and how that class of words/concepts has
>> been or might be represented in silicon.
>> A history of ontologies since Aristotle would be useful, along
>> with the evolving toolkit of semantic nets. I've made a start:
>> http://www.robotwisdom.com/ai/timeline/0000.html
Well, y'know, yeah.
>This Roget-centric perspective has seemed intuitively
>_necessary_ to me since highschool, but I'm painfully
>aware that it's anything but obvious to the academic AI
>community. My working hypothesis is that science-types
>have a sort of Freudian blindspot about the language of
>human emotion and motivation, which constitutes a vast
>part of Roget's system. But I can't imagine any
>realistic alternative curriculum...?
The thing is, if you want Roget's thesaurus, you've got it, already.
Link it to a dictionary. You're done! But what do you have?
>(Back c1976 I ran across a book <Googles> which must
>have been "Language and Perception" by Johnson-Laird
>and Miller, that took certain Roget-style categories
>and tried to do something AI-like with them...
>Obviously the intention impressed me, even if the
>execution didn't.)
Well sure, but this is the naivete of the semantic web idea, that
these things are easy, and haven't been tried before. People have
been looking for the perfect way to represent knowledge and the world
since, well, forever, as philosophy, as metalanguage, as
encyclopedias. The most popular theory is that logic is that
metalanguage, from Boole to Frege to Russell to Wittgenstein to
today's neo-Fregeans (tho I've never quite figured out what that's
supposed to mean) and modal logic fans, with Chomsky's syntax-based
theories of cognition, or Schank's more semantics-driven conceptual
dependency theory and focus on scripts, and various brute-force
attempts to parse and generate language, with varying degrees of
success, back to keyword driven bots like Eliza and Parry, and modern
theories of social interfaces, ...
A thesaurus may constitute an interesting point in a very large
hypersphere of theoretical and pragmatic issues, and then what?
Joshua Stern
Jorn Barger
> The thing is, if you want Roget's thesaurus, you've got it, already.
> Link it to a dictionary. You're done! But what do you have?
It's not an endpoint, just a startingplace. But it has
the _appallingly unique_ quality of bringing together
_all_ categories of meaning in one comprehensible package.
(Dictionaries are not comprehensible in the same sense.)
> >(Back c1976 I ran across a book <Googles> which must
> >have been "Language and Perception" by Johnson-Laird
> >and Miller, that took certain Roget-style categories
> >and tried to do something AI-like with them...
> >Obviously the intention impressed me, even if the
> >execution didn't.)
>
> Well sure, but this is the naivete of the semantic web idea, that
> these things are easy, and haven't been tried before. People have
> been looking for the perfect way to represent knowledge and the world
> since, well, forever, as philosophy, as metalanguage, as
> encyclopedias.
Yes, this is the main theme of my timeline, and the fact
that the SW crowd is so naive argues for including some
semantic-curriculum at a much broader/earlier level of
higher education. http://www.robotwisdom.com/ai/timeline/0000.html
> The most popular theory is that logic is that
> metalanguage, from Boole to Frege to Russell to Wittgenstein to
...Lenat. So Cyc's strengths and weaknesses should also
be much more widely studied.
> today's neo-Fregeans (tho I've never quite figured out what that's
> supposed to mean) and modal logic fans, with Chomsky's syntax-based
> theories of cognition, or Schank's more semantics-driven conceptual
> dependency theory and focus on scripts, and various brute-force
> attempts to parse and generate language, with varying degrees of
> success, back to keyword driven bots like Eliza and Parry, and modern
> theories of social interfaces, ...
If this list could be 'tamed' and extended, that would
be very useful. My own intuition is that scripts based
on 'what usually happens next' will be much more useful
than logics based on 'what can we deduce next'.
> A thesaurus may constitute an interesting point in a very large
> hypersphere of theoretical and pragmatic issues, and then what?
Picture an intro AI-text with a chapter on semantics
that re-sorts Roget's categories and considers how they
can be represented for computers-- the hard,
psychological categories especially. Enumerations of
motives, including "The Sims"' object-driven approach,
would have to be included. Military war-games that
factor in morale and diplomacy and (eg) infiltration-
by-spies are another example of this semantic fringe/
frontier where compsci types start to go into denial...
imho.
Gordon Joly
Jorn Barger <jo...@enteract.com> wrote:
>[...]
>(Back c1976 I ran across a book <Googles> which must
>have been "Language and Perception" by Johnson-Laird
>and Miller, that took certain Roget-style categories
>and tried to do something AI-like with them...
>Obviously the intention impressed me, even if the
>execution didn't.)
>
And a little Googling reveals...
Language and Perception. (George A. Miller and Johnson-Laird, P.N.)
Cambridge: Cambridge University Press. Cambridge, Mass.: Harvard
University Press, 1976.
Gordo.
Jorn Barger
> Picture an intro AI-text with a chapter on semantics
> that re-sorts Roget's categories and considers how they
> can be represented for computers-- the hard,
> psychological categories especially.
Another way to look at this:
The world of physics has a well-publicised buzzword, "the
theory of everything" (ToE) that gets 29,300 hits at
Google (or 12,500 at Google Groups). But even if this ToE
were solved, all it would be is some formulae for particle
physics, and would have negligible impact on ordinary
lives.
Lenat's Cyc Project is an attempt at a _semantic_ ToE, but
gets only 2660 hits (or 942)-- less than one-tenth the
popular visibility. And Lenat (last I heard) had decided
to skip the hard psychological categories for now. But
the theoretical impact of succeeding even on those easier
categories could revolutionise every software application
anyone uses... and if the psychological categories can be
added, a new world now begins!
I've been exploring the semantics of web-timelines for the
last few years, most recently this logarithmic jobby:
http://www.robotwisdom.com/science/logarithmic.html
Because it's logarithmic, the top tenth is astronomical,
the next tenth is geological, followed by biological etc.
I _assume_ Cyc's ontology can handle the formation of
galaxies and planets, and the early stages of life,
allowing these early timeline entries to be represented
comparatively simply using Cyc semantics. (Eg, gravity
condenses interstellar gas into spinning disk.)
But the last half of this timeline is almost all
psychological, and trying to represent it in the same
fashion will demand a level of subtle semantics that Cyc
can't begin to approach.
Gordon Joly
Jorn Barger <jo...@enteract.com> wrote:
>I wrote in message news:<ba7rtr$knl$1...@mulga.cs.mu.OZ.AU>...
>> Picture an intro AI-text with a chapter on semantics
>> that re-sorts Roget's categories and considers how they
>> can be represented for computers-- the hard,
>> psychological categories especially.
>
>Another way to look at this:
>
>The world of physics has a well-publicised buzzword, "the
>theory of everything" (ToE) that gets 29,300 hits at
>Google (or 12,500 at Google Groups). But even if this ToE
>were solved, all it would be is some formulae for particle
>physics, and would have negligible impact on ordinary
>lives.
>
And would be philosophically difficult. A "ToE" could not be tested;
it would be a theory in glass cage.
Like AI?
:-)
Gordo
P.S. Semantic web discussions are AWOL from many forums, thanks to Tim
Berners-Lee.... and W3C?
Jorn Barger
> The world of physics has a well-publicised buzzword, "the
> theory of everything" (ToE) that gets 29,300 hits at
> Lenat's Cyc Project is an attempt at a _semantic_ ToE, but
> gets only 2660 hits (or 942)-- less than one-tenth the
> popular visibility.
For comparison I should have added "semantic web" which gets
an astonishing 186,000 hits on the Web, but only 1490 at
Google Groups.
> > Picture an intro AI-text with a chapter on semantics
> > that re-sorts Roget's categories and considers how they
> > can be represented for computers-- the hard,
> > psychological categories especially. [...]
> http://www.robotwisdom.com/science/logarithmic.html
> [...] But the last half of this timeline is almost all
> psychological, and trying to represent it in the same
> fashion will demand a level of subtle semantics that Cyc
> can't begin to approach.
Pondering what the AI-chapter would look like, and rejecting
Roget's ordering, I started to wonder if Roget's categories
could be re-arranged according to the historical-evolutionary
scheme of the logarithmic timeline... and it looks like that
could work pretty well. I can't post my draft to the Web
until my Mac gets re-connected, but the basic outline is:
big bang, inflation, condensation, stars-etc, dry planets,
wet planets, chemical evolution, life, metabolism [and
here the psychological complexities intrude] sensitivity,
motivation, planning, memory, thought, sexuality,
sensation, space, travel, place, time, vision, reasoning,
society, hierarchy, language, possessions, religion,
technology. The url for the draft will eventually be:
http://www.robotwisdom.com/ai/etymogeny.html
For comparison, Roget's 1852 scheme and Wilkins' 1668 are at:
http://www.robotwisdom.com/ai/roget.html
http://www.robotwisdom.com/ai/wilkins.html
Jorn Barger
> Pondering what the AI-chapter would look like, and rejecting
> Roget's ordering, I started to wonder if Roget's categories
> could be re-arranged according to the historical-evolutionary
> scheme of the logarithmic timeline... and it looks like that
> could work pretty well. I can't post my draft to the Web
> until my Mac gets re-connected, but the basic outline is:
>
> big bang, inflation, condensation, stars-etc, dry planets,
> wet planets, chemical evolution, life, metabolism [and
> here the psychological complexities intrude] sensitivity,
> motivation, planning, memory, thought, sexuality,
> sensation, space, travel, place, time, vision, reasoning,
> society, hierarchy, language, possessions, religion,
> technology. The url for the draft will eventually be:
> http://www.robotwisdom.com/ai/etymogeny.html
A vast simplification, obvious in retrospect:
Since the psychological categories are supposed to be
introduced in the order they evolved, they can be named
after the nearest surviving 'relative' at each point on
the tree of life, eg: germs, worms, bugs, fish, mice,
monkeys, chimps.
I wonder is there an authoritative synthesis like Wilson's
Sociobiology that covers animal behavior of all sorts
(not just social)?
(I actually posted that question to sci.bio.evolution
recently in a different context, in reference to a
timeline of human 'universals':
http://www.robotwisdom.com/ai/universals.html
And Polti's 36 plots can also be rearranged by this
scheme-- to a worm, a sidewalk is an 'enigma' (Polti's
#11).)
Returning to the idea of an intro-AI chapter on semantics,
arranged this way, it would be easy to take the categories
one by one and examine how thinking about that category
evolved, and what simulations have been attempted-- which
is the point of my timeline of knowledge representation:
http://www.robotwisdom.com/ai/timeline/
So before the Big Bang was conceived, there were Babylonian
and Egyptian creation myths, pre-Socratic models (eg all
is water), the Book of Genesis, etc. All the way up to
the WMAP team's latest model for testing fluctuations in
the Cosmic Microwave Background.
The alife comunity has had only a little success with worms
and bugs, but _aspects_ of complex social behavior have
been modelled in all sorts of interesting ways. (Compare
eg MATLAB for physics with Conway's 2-D gliders.) The intro
should collate all these approaches and sketch their
underlying algorithms.
KR = AI + semantics + simulations!