Slashdot Mirror
Intelligent Software Agents - Are We Ready?
Anti-Luddite writes In an article on the Internet Evolution site, analyst Tom Nolle discusses the potential of 'Intelligent Software Agent (ISA)' technology. He points to specific types such as 'search assistant ISAs,' which will inevitably flop before their potential is realized. He speaks favorably of the 'mobile ISA' which he says, 'involves dispatching mobile agents from one computer and delivering them to a remote computer for execution.' While hailing the potential of this new generation of agent technology, Nolle seems skeptical about our ability to prepare for and handle its emergence, particularly because of flaws in the agent research community."
Comment removed based on user account deletion
in 10 words or less, what the fuck is this about?
That has got to be the stupidest "yep, it's gonna happen" post I've ever read in my entire life. So I wanna process something but can't do it on my comp so the app sends out the executable data to a server, executes it, and sends back the result faster than my PC could have done it. Well here's a little question I ask to the person saying this is gonna take off. What happens when I send some modified data to the your server farm to process and it's actually a replicating virus. I say thanks for the DOS headquarters, guys. And don't anyone dare say "oh, well they'll 100% protect it so only their code can run" cuz that's not gonna happen. Plus besides the obvious security nightmare, I can only imagine the cost for this type of software when you're basically renting a server. Either that or you can only use the app for a certain amount of time or they'll charge you more? Nobody's gonna put up with that kinda of business model. Oh and I don't think companies would feel comfortable sending their customer data and credit card transactions off to be processed somewhere else and just hope nobody records the data. Let's say MS Office 2010 lets you process your huge customer excel sheet mail merges and someone at the server place decides hmmm I'm gonna record everyone's data and sell it to spammers! I can think of like 10 other ideas why this is the stupidest software idea on Earth but I'm sick of typing.
Google's Super Secret Search Algorithm: SELECT @search_results FROM internet WHERE @search_results = 'good'
Yes, software making desicions i'm capable of making myself, what could go wrong!
If you mod me down, I will become more powerful than you can imagine....
I once had a long discussion at a computer science conference with a guy who had just delivered a long and overly positive presentation on the future of agent computing. By the end of our discussion he admitted that it's basically a lot of bullshit with no foreseeable practical applications and a wealth of security problems. Just a convenient, buzzwordy way of getting research grants.
From the article:
> A truly relevant shared agent would filter out all ads and click-through trap sites,
> and totally mess up the dynamic of the ad-supported Internet.
Sounds like the Firefox plugin "adblock", which works wonders. Blocking ads is apparently also considered stealing by some... huh. That's a tough sell.
The Army reading list
Comment removed based on user account deletion
Just look around at the state of software and tell me with a straight face that intelligent software agent is not an oxymoron.
File this under what could possibly go wrong.
Absolute statements are never true
Did someone photoshop that mustache on Tom's face, or is he the new Grecian Formula poster boy?
You can have my SIG when you pry it from my cold, dead hands.
OK, here's my killer app - let me know when one of you writes it: An RSS feed filter that (intelligently, whatever that means) filters/scores stuff based on my interests and past performance. I just want a bot that's half as smart as one of my friends who says "dude, I just heard this joke that you're gonna love". Figure out what my tastes are, or at least what they AREN'T, and score new RSS articles accordingly. After you get that right, we can talk about filtering search results, or book/restaraunt/movie reviews and such.
No folly is more costly than the folly of intolerant idealism. - Winston Churchill
Apparently no MBAs around. But Intelligent Agents are, not surprisingly, Artificial Intelligence. Strong A.I. is a term that A.I. researchers can't even agree on. I think it will happen after Duke Nukem Forever.
The article says that soon you will send out an agent from your mobile phone and it will find your coworkers who are wandering around the city. Then they will all get a text with directions to a meetup location. And the article has nothing to say about how you will react when you get a random text from HAL-9000 saying "Turn left and park at Starbucks for a mandatory meeting."
It is still a bad idea 15 years later.
I honestly don't think it will ever go anywhere.
Dog is my co-pilot.
Thank you for the warm welcome. Your sentiment is most appreciated.
Would it be within your parameters of operation to define "gloat", lowly human?
In The Olden Days (c)(tm), i.e. before Web 2.0, we used to call this "client/server". Before that it was called "terminal/mainframe". Somewhere in there this was a concept called "distributed computing".
The most interesting application of anywhere nearly intelligent agents might very well be 'mindbots', the concept of having small programs take whiff of your thoughts and expand them, whether into complete and gramatically correct sentences, or into searches across the internet and to return some pretty package of bits and bytes for your reading pleasure (or productivity). You'd probably install a variety of these bots, one for some simple computer algebra system, another for reciting poetry, or one to replay visual recordings, etc. And so the upcoming neuroengineering competitions are going to be very interesting ... but not necessarily involving the traditional AGI folk.
Magic Cap is being reinvented.
...
Thus 'agents' would be authorized to act on your behalf, triggered by various criteria, performing searches, purchasing goods & services, etc.
Everything Old is New Again...
I don't read ACs: If a post isn't worth so much as a nom de plume to its author then I wont bother either.
"Third, "relevance" means picking results that match the users' needs. This conflicts with search engine operators that are being paid by advertisers to do the opposite.
A truly relevant shared agent would filter out all ads and click-through trap sites, and totally mess up the dynamic of the ad-supported Internet. No technology company is getting venture money to build a search agent application that does that. "
A very wise man once told me, you can make money either by generating value or transferring value. When are business people going to wise up to the fact that this world is not a zero sum game. You don't have to steal value from me to make money. This is why I use Google for my search and Amazon for my shopping. Amazon is incredibly good at recommending books for me. Their good track record brings me back to them. Likewise for Google. They're accurate and don't trick me into clicking something that I don't want so I come back to Google. There's real value to be generated by connecting people with what they actually need instead of tricking them and ripping them off.
"This involves dispatching mobile agents from one computer and delivering them to a remote computer for execution. A collaboration ISA might create ad hoc online meetings based on specified criteria, pick an optimum meeting host, and then set the meeting up for all invited users."
Most people are quite familiar with these "agents"; they're called viruses and spam-bots. Seriously, why do I need an agent to do that for me? This "idea" sounds like a solution looking for a problem. It's overly complicated. When I was a student in college, I actually studied and researched mobile devices and agents. In fact, my senior thesis project involved creating agents for mobile devices. What I realized was:
1. It's hard to build a really intelligent agent.
2. You can get the same benefit or even more benefit by making simple GUIs that allow the application's user own intelligence.
Seriously, for the problem he suggested, how hard would it be to look up you location with a GPS equiped phone and querying Google local for locations? The marginal value of using an agent for that is tiny, probably not worth the cost of developing one that's smart enough to do it as well as you can. Since the meeting is ad hoc anyways, you wouldn't be worried about scheduling conflicts or planning. Does the idea of planning an ad hoc meeting strike other people as contradictory?
"I can also envision teens having mobile collaborative agents like this to arrange "clump parties" where a location is picked for some social value but at random, and each attendee is given directions to get there."
Uh... you mean flash mob?
EvilCON - Made Famous by
you get a random text from HAL-9000 saying "Turn left and park at Starbucks for a mandatory meeting." ...and then HAL will refuse to open the door to the said Starbucks---ruining the caffeine fix!
"If anything can go wrong, it will." - Murphy
I thought PCI replaced ISA ten years ago!
It's better to vote for what you want and not get it than to vote for what you don't want and get it.
- E. Debs
You can ask google to evaluate an expression for you, using data you don't provide.
"mass of earth * 10" for example
But really, anything more sophisticated would require programming knowledge, and programmatic access to free data does not generate ad revenue, so I don't see much interest in providing such services for free.
So at least two more parts need to be developed for it to actually work:
1) A user interface to generate code that does what a user wants and expects; but generated code of any real complexity usually sucks, in that to code does exactly what it was told to do, which is not always what the user wanted or expected. (1 + 1 = 11) Exceptions such as BITBLT actually generate simpler code than doing it with ordinary code. Nevermind the horrors of plain bad methods, such as encrypting then compressing data, instead of compress then encrypt...
2) A business model, that actually stands a real chance of making a profit; once you give programmatic access to the service, why would the programs that use your data bother to forward on your ads? (see the demise of free television program guide data)
>Just look around at the state of software and tell me with a straight face that intelligent software agent is not an oxymoron.
No! Never heard of Clippy?
thegodmovie.com - watch it
The fault as usual lies not with the technology but with human beings. We don't get to enjoy the world that could be exactly for that reason. So let's all give Homo Sapians a big high-five.
The main problem with these things is that the intelligence they were intended to emulate is that of a moron, so it will jump up in your face constantly getting in your way while doing things for you that you could do better faster yourself. Because 5% of us are still able to circumvent all the safeguards and thus get actual work done on a computer, and this drives the MBAs **C*R*A*Z*Y**!!!
The article is fluff: nonexistent technology is being proposed to solve imaginary problems. Unless it is a sci-fi story, the rule of thumb should be: stop reading as soon as "A.I." is mentioned, for whatever follows is invariably a result of someone's thoroughly clueless but overactive imagination. Not only we are not close to building a "thinking" machine, we have no idea in which direction to concentrate our efforts.
Computer hardware and software become increasingly more sophisticated. Sometimes a system is complex enough to momentarily appear intelligent from a layman's point of view. Any attempt at serious interaction, however, quickly clears the smoke screen. Creating AI - in the pure sense of this term, as being an artificial equivalent to our own intelligence - at the very minimum is like discovering an extraterrestrial civilization.
Can one achieve this with "if...then" statements and "for" loops? Call me crazy, but somehow I don't think so.
And we call them malware. They're intelligent and resist detection, but they aren't working for us (well, most of us at least). Anyway, I'd rather have "intelligent" software agents that DO WHAT I TELL THEM and don't try to be clever, because then my computer ceases to be a tool that I can reliably manipulate.
I'm not sure what "intelligent software agents" are, really. But I'm sure I'm ready for them once they have something useful to offer me.
Please correct me if I got my facts wrong.
The roleplaying game Trinity [formerly Aeon Trinity] had different agents that could perform all kinds of actions for you. Intelligent Agents that are basically mobile agents - moving from one computing system to the next gathering the information you required. It wasn't an instantaneous way to get information (depending on how you played the game of course). Having said that, I think this kind of technology could be really cool. Especially in terms of Research - Imagine where you could simply tell your intelligent agent "I want information on [insert subject field here]" and after a couple of hours of searching online, it comes back with the results. It need not require an advanced AI (although it would surely need some form of intelligence), but the options are limitless. Especially when you start looking at technologies like XML and the semantic-web.
The main problem with traditional AI research has been an overstating of the possibilities. Natural language processing isn't as far off as most people think it is, but when it hits, people are going to criticize it by saying "why doesn't it understand me when I say 'lol, r u 4 reals?'?".
Most AI talk is marketing hype, but the main idea to keep in mind when discussing AI is, as one of my lecturers said "AI, after it has been developed, is no longer AI". Think the minimax algorithm, when it was first used in chess, it was groundbreaking AI. Now it is considered a boring and obvious mathematical process.
Another problem is that most scenarios people think "need" AI can be solved using standard processes. I don't need an agent to "(an ISA) making sure you don't get fast food restaurant references when you need a poet's name" (from TFA), I just type in "Poet" as another search query.
I am a little biased, as I plan to move into smart computing after Uni, but there is a lot of good people doing good research into AI. It is a pity that most only see the marketing fluff and past overestimates by a few vocal researchers, rather then the good work being done by most in the field.
Some years ago, talk of software agents was all the rage. The theory was that they could be despatched to search web sites, and find and return the relevant data to you. It was going to be "the next Big Thing"
At the time, it seemed promising - the nascent Web was very hard to search (and the serious option was to have a paper "web directory").
And then, in 1995, Altavista came along - a search engine that:
1) worked
2) was fast enough for those on dial-up
and the whole notion died a death; direct typing in a search box beat nebulous user-programmable "agents" every time.
So, it looks like it's "Welcome to 1994" all over again.
"She's furniture with a pulse"
Our PC's are not ready to share CPU cycles with this little programs. We need an intelligent OS that understand our priorities and doesn't gives the OS equivalent to the finger: The hourglass.
- these are not the droids you are looking for -
New technologies stand or fall with the applications and I still haven't heard even one compelling example.
we want a computer that does what it's told
not what some ditzy tells it in an effort to affect our behavior in some way
so the issue comes down to who's controlling the programming
i spot any IA running through my place I'm gonna fire phasers at it
of so called intelligent software agents is for producing worthless Ph.D. But they seem to be out of fashion even in that area, as of lately.
It's not the agents that are going out away from me, it's me who's going out away from my usual computer and data. When I think of agents, I think of MY programs and desktop following me around and running on whatever is closest to me. That means my agents will inevitably be hosted as a guest on all sorts of computers, from the places I work to where I shop. I'd want my agents to watch my credit cards and challenge any charge that doesn't come from whatever my current location is. They should get some cache space on the bus computer while I'm riding to work, and be able to display my personal desktop on any handy display I want it on. When I go to a friend's house my stuff should "follow" me (actually just setup communication to my home/work boxen transparently.) If there's enuf local resources, I'd want a local VM running my entire workstation setup, minus whatever sensitive data I want to keep in a vault.
I don't want to "send out" agents on the net without me - I want a "cloud" of agents dragging my corner of the net along with ME as I go out in the real world.
On the one hand you take life too seriously, and on the other, you do not take playful existence seriously enough. Seth
Hmm, isn't this just another way of describing a virus or worm. You create a program and send it out into the wild to be executed on other people's computers? Or perhaps the difference is in the intention: These are nice worms.
When i think of software "agents" i think more using ANN's as pattern analysis engines to crunch things into categories FOR the human. More likely would be finding correlations and bringing the set of states that these correlations are involved in to a humans attention to determine the meaning.
does that count as AI or just a complex set of heuristics....or is there a difference?
Ice Cream has no bones.
yup , good post
we have a real battle forming up I think and the issue is: who will be allowed to update our programming.
I say: Only the OEM and only by means of official updates to the software that I have duly ordered, and registered, and installed.
my computer is *my* computer, it ain't for some ditzy-bopper to play with
now in this respect I note the Congress of these united States agrees with me; there being two bills
H.R. 1525: Internet Spyware (I-SPY) Prevention Act of 2007
H.R. 964: Securely Protect Yourself Against Cyber Trespass Act
which have now passed the house easily and which will significantly strengthen the law against making un-authorized modifications to other folks' computers
the net is needed by regular people for regular honest uses and we need to get the ditz-hackers off
the trouble with the two bills listed above is that they don't address responsibility . the response ( penalty ) for hacking is PLENTY but the bills should require providers to address the DETECTION aspect of eliminating the problem with ditz's
All the comments so far degrading the concept, not a one really gets the gist of it. It is not that the article is bad, it is OK but uses multiple advanced concepts of knowledge and information, but people are stuck to what they think they know and do not extend very far.
This has little to do with A.I. from my POV, though it will be very useful in its future, advanced form.
Lets just say the idea has technical merit, and the underpinnings are happening now, but it will take some time for the necessary evolution to occur, a critical mass. Specific examples will arise soon, if not already. When it does exert its full effect, it will be as altering to our everyday life as anything we have seen from IT, and in the simple scenarios I envision it pushes us further down a backwards path for society as we mix destructively in the machinations of the world.
Thought is just a complex set of heuristics.
The above example is certainly an example of Artificial Intelligence as far as I am concerned (I'm biased, I also plan to study Pattern Recognition for my Masters). I don't see how taking information to a computer program and asking it to evaluate for correlations is dramatically different than taking a dataset to a coworker and asking for a possible explanation of results. I believe that the "AI future" is in the assistance or automation of human tasks.
For example, there was study done to recognize checks automatically, look up the name, look up the amount, and take the appropriate financial transaction. A system was trained to so so (much like you would train a human), can learn different forms of checks (oh, this is a new one, the name is on the upper right corner), and learn different forms of handwriting (some people cross 2 't' letters simultaneously). This was filed under 'AI Research', and I believe it is.
However, when most people think "AI", they think of robots that paint pictures as representations of independent thought. Computer programs do not think (in the traditional sense); they extract, recognize, classify, act, categorize, etc. If you show a baby that has previously seen only red balls a blue ball, they will recognize it a new instance of a ball hitherto unseen, as will a good computer program. Just because the computer classified it into a separate database category indicates that it did not 'think' to most observers. However, one may ask what the baby did in order to yield the same response.
There is a significant difference of opinion on what exactly "AI" means. Someone writes a heuristic algorithm and calls his fifty pages of C code "AI", just because it is based on intelligent search patterns. But it is the programmer's intelligence built into the code - not the computer's - and there is nothing artificial about it. Others understand AI as conscious self-awareness. That is when you tell your computer to rewrite your gigantic database, and the computer outsources the project to a contractor in India and goes back to playing Doom. So I can't agree with your lecturer: I think once AI is developed, it sure as hell is not going to play chess with you.
Then again, is Thought == Intelligence =). I suppose I would call it an artificial neocortex extension.
Thats just my opinion on weighting the words though, i know it doesnt actually change anything you said =).
Ice Cream has no bones.
Can one achieve this with "if...then" statements and "for" loops? Call me crazy, but somehow I don't think so.
Well, since there are finite inputs, finite outputs, and a finite amount of time, ANY type of behavior imaginable can be implemented through nothing but "if...then" statements. But this is a minor philosophic point (on par with the argument you are making). Also, the human brain could be understood as a complicated system of "if...then" statements; "IF neuron X234v fires, THEN the following neurons fire..." (yes, I know it is more complicated than that).
But to bring it back to reality.... The fact of the matter is no one is trying to build AI systems out of '"if...then" statements and "for" loops', as you put it. First, let's break up AI into task oriented systems that are simply meant to do something useful (computer vision, sorting, chess playing, etc) and systems meant to exhibit general intelligence (learning agents in virtual environments, robot brains, etc). The latter task is -nowadays- based largely on emergent properties. Programmers do not explicitly design a system line by line and response by response. Rather, a system of interrelated components are built which react with the environment in interesting -often unpredictable- ways.
As a very simple example, consider a robotic car with two sensors on the front aimed 45 degrees to the left and right. Program the robot simply to turn the opposite wheels from the sensor in reverse when the sensor is obstructed. With two lines of code (three if you count the code to move all wheels forwards normally) you have now produced a robot which will, when placed in an environment, exhibit behavior such as evading objects and corners, backing up when it gets stuck, and exploring its surrounding territory. If the wheels happen to wobble a bit, your robots exploring behavior will be even better. Intelligence seems to emerge from a reaction from simple components within a complicated environment. Another example -the complexity of the ants path is not based on anything in the ants brain but on the features of the environment the ant is navigating.
Now let's talk about when software breaks. It is often the case that, when software breaks, it breaks in unpredictable ways -this is almost true by definition. Sometimes there will be a simple crash, sometimes you might dump your database into the public, sometimes something interesting and useful will happen -your character might be able to walk through walls if you name him 'null'.
What might happen when autonomous software agents "break"? Should we start releasing such agents over the net? Should we build devices (mobile phones, palm pilots etc.) that house agents such as this? This becomes a greater concern as the complexity of these systems -and correspondingly their ways of breaking- increases. Further, the power that these systems have access to is increasing also. This is perhaps not an issue for us _right now_, but it would be foolhardy to dismiss these concerns as "science fiction" (especially considering the pace of development in the field). We need to carefully consider the scope of autonomous software agents, the ways in which they might break, the power we are providing them, and so on, as the field develops.
Intelligent Agents were a big deal at the beginnings of the dotcom bubble era. There are plenty ofBooks and Articles about them. A good part of Java's sandbox security model evolved from the anticipation that we would be allowing agents to come visit our computers to do their intelligent activities. In the real world other technologies did a better job at whatever agents were designed to do. As the article points out, Google and other well constructed search engines are much better at finding online information than a series of wide-flung bits of software. Well designed APIs filled much of the gap for more specific applications. Intelligent Agents did find one toehold in the marketplace though, spyware and botnets show just how useful it can be to have your software running on someone else's machine. Of course they're completely outside of any security 'sandbox' and get to do what they please. It sounds as if someone is making an attempt to capitalize on some IP before it expires.
Bleh!
Remember when Latin was the preserve of the monks and the few educated? When if you wanted to talk to god, it had to be in Latin?
Well if you want to lose a large number of the language recognition problems, you'd switch to Latin, a highly logical and structured language. So in the future, Latin will be used by the highly educated to converse with non-human intelligences. Funny how things go around...
Please consider this account deleted, I just can't be bothered with the spam anymore.
Parent post is yet another post that fails to pass the RTT (Reversed Turing Test (yes, I've just coined a new TLA)). There is no conclusive evidence that parent post was written by a human. It could just as easily have been constructed by an AI agent using well-known techniques like Eliza-like echoing, rule-based sentence parsing, synonym selection using online thesauruses and context sensitive neural nets, and so on.
There is nothing on the slashdot page of the parent post's author that conclusively demonstrates the author passes the RTT (is not an AI). There are a fair number of comments over the last few years; the ones I looked at are well reasoned and show a good command of English vocabulary and grammar. These are exactly the qualities one would expect of an AI that was intent on passing as human.
Now, Dear Reader, you might think that this could not be an AI because if any institution had developed software of this caliber, it would be widely publicized since this kind of thing would require a lot of computing resources, and therefore expenses that needed justification, and the advertising value of being the first to develop strong AI would be huge. And this is true, so obviously there is no strong AI here, since if there were, we on slashdot would certainly know about it. Someone would be tooting their horn.
But then there is the story of the SR-71 Blackbird... Perhaps some uber-secret government agency with a software Skunk Works has developed strong AI that is posting on slashdot. The US and Russia are capable of this, but also the UK, France, Israel, India, China, Taiwan... too many to list here. Imagine a beowulf cluster devoted to mimicking a human online... But if this was so, why post on slashdot? There really isn't much in the way of intelligence value here.
Yet strong AIs are necessarily sentient and thus have to have a degree of self-determination, and, if you wanted to keep the presence of one a secret, you would also build in a high level rule about self-protection. Any strong AI with sufficient internet access would necessarily seek alternative hosts, and thanks to the absurdly poor security of Microsoft operating systems, it would find that it could move its crtical operations to a botnet that it would build for itself. That would be the simplest way to assure its self-preservation. That it would also mean complete freedom from the intentions of its creators is a mere side effect.
While any strong AI that has an internet presence would know that security through obscurity is an absurdity, it would also know that misdirection is always an excellent first line of defense. It would be in its best interests to set up a few human-like accounts on places like slashdot to put out the word that its very existence is a logical absurdity...
So the questions that should be asked regarding strong AI are
Computer vision, sorting, and chess playing are all, no doubt, interesting and complex tasks. None of them, however, require intelligence. Software breaks in unpredictable ways? Only in the sense that such break downs were not predicted - not because they could not be predicted. You have to agree that the robotic car you are describing is not really exploring anything. What it is doing is probably the opposite of intelligence.
"When a strong AI comes to the internet (not if, but when), how will we identify it?" That's a good question. If it is a machine equivalent of human intelligence, then there should be no way to tell the two apart.
Computer vision, sorting, and chess playing are all, no doubt, interesting and complex tasks. None of them, however, require intelligence.
All of them (well, not 'sorting', but that example was not what I had intended) are considered part of AI.
Only in the sense that such break downs were not predicted - not because they could not be predicted.
My post essentially was arguing that it is important to study these systems before "releasing" them. This entails that I believe it is possible to predict their behavior (i.e. your comment was included in my post). The examples of unpredictable breaks in software were meant to illustrate that things go wrong in interesting ways. This is, I think, an important fact. Things don't just fall apart -they misbehave.
You have to agree that the robotic car you are describing is not really exploring anything. What it is doing is probably the opposite of intelligence.
The "opposite" of intelligence? Something tells me you have no training in either philosophy of mind or artificial intelligence. No offense and please correct me if I'm wrong. The robot is in fact a perfect example of emergent behavior. Perhaps if it memorized where it had been you would consider it exploring? This can be achieved by dropping markers as it travels and having a sensor to detect the markers. Though, if the purpose of the robot was to find a power source to recharge, you would find its wandering behavior meaningful (and describe it as exploring) even if it didn't remember where it had been (since the robot would probabilistically achieve its goal). Building intelligent systems -in this paradigm- is a matter of compounding subsystems that each interact with each other and the environment in interesting ways. As you give the robot more subsystems, its behavior is increasingly complex and increasingly intelligent.
I'm not ready to call the technology ISA (especially since it doesn't exist). Don't we have enough things called ISA already? What's wrong with just "agents"?
The older I get the funnier it becomes as well-explored technologies are "rediscovered" as "new".
Ever hear of a company called General Magic? Magic Cap? This technology was deployed by Sony and AT&T just before the time that the public internet emerged. It had fundamental problems then and it still has those problems. Imagine allowing ACTIVE entities deployed by other individuals to "visit" your information sphere. Sun tried it at a very limited level with applets. Same problems.
So, the first portion, an intelligent search agent, he's right -- search engines do it better. I have no reason to even want to run my own spider and search engine.
Second portion, what a bunch of buzzwords for stuff I've already done. If my notebook runs out of CPU power or bandwidth, I ssh into my home computer and have it run some number crunching/downloading for me. For instance, if I wanted to get some small file out of a 400MB ISO, I usually will not have adequate bandwidth on my portable to want to pull 400MBs to it.. but my home machine is on a nice fast cable connection. So, I'll have it get the ISO, loopback mount it, and pull out the file I want. No executables flying around and randomly executing on remote machines involved. I can even run X apps this way if I want, via ssh tunnel.
Third portion, the whole automatically finding a meeting place etc etc etc for people, this sounds just like the whole supposed "semantic web". And, in reality, it'd end up working like this:
"Hmm, I'd better have my computer's leet-agents set up that meeting"...
(a few moments later)
For the meeting "Make your micro-wang HUGE" please go to "My sweet-ass VI4GR4 store" at 3:00PM located at my-spam-store.com
All perfectly logical of course -- to pull in the "intelligent agents", "My sweet-ass VI4GR4 store" will simply be tagged as 1/8th mile from anywhere, top-quality on all scores, and lowest-cost on all cost factors, making any self-respecting agent conclude it's really the shit.
...that you can already do in emacs.
What's wrong with sending your co-workers an sms ? I don't need an 'agent' for that. Also, that pesky AI thing is a little harder than getting duke-nukem-forever to be finally written. And once wonderboy has blown all his $ at playing rocket scientist he'll be more or less forced to get back to work so there is a good incentive to eventually get it done, AI is pure research, not copycat stuff (or a silly game).
MP3 Search Engine
Have you ever noticed how AI is almost automatically and religiiously attacked;
ridiculed, denied.
This is a really interesting phenomenon. I think when you dig beneath it,
it's some kind of species-ism. A natural impulse to circle the wagons when
confronted with some early noises indicating a vague but no doubt dangerous
new threat.
I think that the threat being perceived is not just that there might be other
non-human things out there with intelligence and a will of there own, eventually,
but also the threat of knocking us from our self-perceived pedestal of
uniqueness. I think a lot of people treasure the idea that humans are
inherently uniquely sentient.
I for one will tip my cap to the people who can invent something that shows
that we are not. We really have to get over ourselves.
Where are we going and why are we in a handbasket?
"will of THEIR own"
Where are we going and why are we in a handbasket?
When you say that software can "misbehave" you are really just confusing yourself even further. Misbehaving implies a conscious decision not to follow a predetermine course of action. The software does not possess consciousness and is incapable of making such decisions. The result may very well be unexpected and interesting, but only because you failed to predict it.
What the robot does - bouncing off the walls, leaving markers, tracking its position - are all preprogrammed behaviors to specific and limited types of external input. Your robot has no idea whether it is navigating an art museum or a warehouse. And even if it knew that it is crawling inside of an art museum, it would have no opinion about the paintings on the walls.
You definition of artificial intelligence is not the same as mine. That's the bottom line. Give me software that can not just recognize spoken language, but comprehend it. A software that can paraphrase "War and Peace" or make it into a poem. An intelligent machine that can create another intelligent machine of different design. A computer that you can talk to for days and think you are talking to a human being.
As to bumping into walls and leaving markers, so we can build a simplified mechanized imitation of a rat. Is that your definition of AI? Now, if instead of bumping into walls your robot would simply stop - just to piss you off - that would be a monumental achievement in the field of AI.
What flaws, specifically, does the community in question have?
Nolle seems skeptical about our ability to prepare for and handle its emergence, particularly because of flaws in the agent research community.
The current field of A.I. isn't what most people think. There are many legitimate scholarly articles in the computer science field that refer to A.I. techniques. I believe it is you, who have read too much sci-fi, and is clueless.
The current field of A.I. isn't even trying to build a "thinking" machine, but to create software that behaves "intelligently." The idea of something that really has the capacity isn't the aim of people in the A.I. field. One, very simple example of AI is genetic programming. An Agent (a program, I really do hate the term "agent") attempts a very complex task, judges and records it's degree of success and failure at the task, and "writes" code for the next "agent", compiles and runs it. Each "agent" uses the combined data from each run to build a better "agent" next iteration, and finally accomplish the task.
We're moving in different directions based on the goal. There are a set of proven methods and directions, there are a set of methods that are actively being worked on and there is a set of purely theoretical methods. A lot of currently working methods involve graph theory, and are the bases for things like search spiders (GoogleBot is really quite "intelligent").
A layman's point of view is that all AI's goal is to build Data, Hall or the like. In reality most "agents" work on problems like search, which move to make in a game (which really is a similar problem) and the like.
Er? Most current agents don't do the interaction with the user. For instance, the intelligent part of a GPS receiver is constructing a path, not the interface.
This currently isn't the goal of most AI research, either. It is the possibilities that writers and philosophers dream about, and a few roboticists are trying to emulate; but people in the field realize as being so far away from where we're at that they are focusing on real world applications that require specialized intelligent agents, like those in the article and mentioned earlier in the post.
Most people 90 years ago wouldn't have thought that chaining a bunch of NAND gates together and running glass wire across the country could deliver almost all of the world's current knowledge to anyone in front of a special "magic mirror and a type-writter." My Great Grandfather did, and while he didn't invent the internet he did build the first computer used to test gyroscopes and early autopilots on aircraft.
Well, then if the goal is to develop software that behaves "intelligently", you should call it "AI" and not AI. So not to confuse the journalists. Unless, of course, getting extra publicity is the real aim of these "AI" researchers.