Slashdot Mirror

Seems like the coder used genetic/evolutionary programming to achieve the "feat" However, it is not truly random since you are comparing end result with an absolute answer, and not some mathematical approximation. In simple words, the user wanted Shakespeare, so he hardwired the code to get Shakespeare. I don't mean comparing the end result to Shakespeare. The program effective weeds out non-Shakespeare out of the result pool at every iteration. Not random. Here is an example demonstrating how anyone could do it. http://www.generation5.org/content/2003/gahelloworld.asp

During our discussion, important value and ethical issues arose in the design and use of such "relational robots." These robots are meant to create bonds and simulate "authentic" relationships. They react to voices, track their owner's eyes, respond and project emotions, and so on. Yet, they remain robots - all these actions and reactions are programmed - pre-determined. So, how do the designers decide what emotions to program and which to omit? In an effort to be realistic, My Real Baby gets happy as well as sad. If you bounce her when she's happy, she gets more happy; if you bounce her when she's fussy, her fussiness only increases. How should she react, then, if she is abused? It is not hard to imagine a child (especially one who is herself a victim of abuse) to violently shake, strike or otherwise "abuse" the doll. How should this evocative object respond? Should she show pain? Begin to cry? Eventually "pass out" or even "die" if the abuse continues? How "real" should the robot be in order to create an "authentic" relationship?

[In the end, the designers wanted the doll/robot to react as a child would, with pain and sadness. However, the company's lawyers stepped in and were concerned that any type of response by the doll might encourage further abuse (stimulus-response theory), and they didn't want to be accused of actually encouraging abusive behavior. In the end, the doll simply did not react to abuse.]

Other ethical dilemmas related to the design of such robots included whether they should be capable of deception or betrayal, two common features of human relationships. Or, should they "die." On one hand, the experience of death as part of the life cycle is an important part of psychological development and would add to the "authenticity" of the relationship. On the other hand, one of the benefits of these robots seems to be the avoidance of the emtional damage that can happen when a "real" companion (whether a human friend, or even a companion dog) dies.

Kohenen Neural Nets are probably the basis for this... I discovered this article only a short time ago. The author does some pretty impressive things with his client. I can't find the source he linked to anymore, as it's dead :( Self Organising Maps (another name for Kohonen NN's) are pretty good at figuring out stuff like this.

[PDF] Artificial Intelligence AI = A + I Replicating human ...
File Format: PDF/Adobe Acrobat - View as HTML
... Strong AI vs. ... terms; a rigorous understanding of intelligence is possible Intelligence
can take place outside the human skull The computer is the ...
www.cs.ucsb.edu/~cs165a/Lectures/9-30.pdf - Similar pages

Artificial Intelligence vs. Human Intelligence
Artificial Intelligence vs. ... manipulate computers to do many tasks that reduce the
human labor. Although a computer appears to do intelligent tasks, it does not ...
www.geocities.com/Paris/Arc/4865/AIvsHI.html - 13k - Cached - Similar pages

Artificial Intelligence VS Human Psychology
Artificial Intelligence vs Human Psychology. ... In French's words, Only a computer that
had acquired adult human intelligence by experiencing the world as we ...
members.tripod.com/edward_mallon/rants/rants26.h tm - 34k - Cached - Similar pages

generation5.org - PenteAI
... vs. human match for the first five moves, then switch to human vs.
computer). Set Biases... What is that?! ... Artificial Intelligence. ...
www.generation5.org/penteai.shtml - 10k - Dec. 9, 2002 - Cached - Similar pages

WH Calvin notes on human memory mechanisms
... You may wish to download without images. Computer vs. Human Memory Mechanisms.
Silicon Valley show-and-tell 23 August 1995. ... Discrete vs. Distributed Memory. ...
faculty.washington.edu/wcalvin/memory.html - 8k - Cached - Similar pages

essays and essays writing essays on Psychology - 176-010
... Computer Intelligence vs. Human Intelligence [ send me this essay ] A 3 page essay
in which the writer analyzes a specific journal article which explains the ...
www.essaypage.com/categories/176-010.html - 15k - Cached - Similar pages

Deep Blue vs kasparov
people.advanced.org/~jaron/deepblue.html - 10k - Cached - Similar pages

Computing Science vs. Computer Engineering
University of Alberta, COMPUTING SCIENCE VS. ... computation, distributed computation,
computer-human interaction, computer ... vocational areas in computer science and ...
www.cs.ualberta.ca/programs/undergraduate/cmpeng .php - 12k - Dec. 9, 2002 - Cached - Similar pages

Profile / Id Blog: The exploration of new ideas and techonlogy ...
... Human intelligence: observation of meetings and attendance at common events. ... the type
of agent making the recommendations (human vs. computer, collaborative vs ...
startsimple.com/trend/profile/ - 21k - Cached - Similar pages

What, there's something beside A*?

Look, I could object to Aibo hacking on the grounds that somebody might develop an attack Aibo that will be programmed to roam the streets, ripping people throats out. But let's try to get a grip, shall we?

There is a famous naysayer who pretends that IA-64 is doomed to be an underperformer. The problem with this naysayer is:

• 1. He's often hard to understand.
  
• 2. He's dead.
  

I'll shamelessly quote myself here. Here's an excerpt of an article
in The Register in which I said:

The problem is that the only way to predict what a program
will do is to run it (that's a consequence of Turing's
theorem). A classical processor executes the program code and
knows what happens with the current data (loops, jumps, etc). It can
infer locality relationship from its instruction flow and make
on-the-fly optimizations. Meanwhile, the compiler of an non-optimizing
VLIW processor such as the IA-64 can only look at the code beforehand
and make assumptions about what kind of data that code will
process. Turing's theorem says that the only way to know what a
program will do with a given data set is to execute the program.

Thus the theorem predicts that the IA-64 pre-compiled
optimization will always be inferior to a good on-the-fly RISC
processor. At best, the precompiled optimization will yield an optimal
path for a given set of assumptions, which may not be true on every
run, since run-time data can defeat these assumptions.

I'd gladly be proved wrong. Can someone please tell me why Turing
is wrong and Intel right? Or is Intel fighting an uphill battle
against Uncle Alan's laws?

Now, if you guys are going to implement something THAT complex, then it probably would warrant a neural network...anything less, and you're kinda using technology for the sake of it. Don't get me wrong, I love neural networks, but they tend to be overused so applications that don't need them.

Perhaps the network wouldn't be used to explictly control the different switches, but it could be used to determine (for example) when you've gone to bed and turn off all the lights. Perhaps over time (as the old age sets in), it realizes you are getting up rather regularly at 1:00am to relieve your ailing bladder, so at around 12:00am, it puts the lights on at a low level, then turns them off 20 minutes after it detects the toilet flush. (I am actually being serious, it is just a funny situation)

Security is always a big issue these days, so the neural network could do some funky stuff there. You are listening to music through headphones, but your computer detects this and when I ring your doorbell, it turns down the music and says "James is at the door, would you like me to let him in for you?" "No, ask him what he wants", computer asks me. "He wants you to show off your new intelligent house" "Ok, let him in...". The computer could use sensors on the windows and doors to keep the house under close constant survelliance, it could turn lights on and off randomly to give the impression someone is in, when they're not (ward off burgulars)...

Again, a lot of this can be done "classically" but perhaps when dealing with a large network of periperhals, you'd want to use a NN. Just my thoughts.

For those of you that don't know what neural networks are, check the Generation5 essays on them. Yeah, shameless plug at the end, sorry :)

James.

Now, if you guys are going to implement something THAT complex, then it probably would warrant a neural network...anything less, and you're kinda using technology for the sake of it. Don't get me wrong, I love neural networks, but they tend to be overused so applications that don't need them.

Perhaps the network wouldn't be used to explictly control the different switches, but it could be used to determine (for example) when you've gone to bed and turn off all the lights. Perhaps over time (as the old age sets in), it realizes you are getting up rather regularly at 1:00am to relieve your ailing bladder, so at around 12:00am, it puts the lights on at a low level, then turns them off 20 minutes after it detects the toilet flush. (I am actually being serious, it is just a funny situation)

Security is always a big issue these days, so the neural network could do some funky stuff there. You are listening to music through headphones, but your computer detects this and when I ring your doorbell, it turns down the music and says "James is at the door, would you like me to let him in for you?" "No, ask him what he wants", computer asks me. "He wants you to show off your new intelligent house" "Ok, let him in...". The computer could use sensors on the windows and doors to keep the house under close constant survelliance, it could turn lights on and off randomly to give the impression someone is in, when they're not (ward off burgulars)...

Again, a lot of this can be done "classically" but perhaps when dealing with a large network of periperhals, you'd want to use a NN. Just my thoughts.

For those of you that don't know what neural networks are, check the Generation5 essays on them. Yeah, shameless plug at the end, sorry :)

James.

1.) Is a lot of your time spent in front of the computer, or do you try to get out or give yourself other hobbies?

2.) How do your friends see you? I went through a period of having ppl playfully teasing me about my "computer skills" - but gradually this gave way to geniune respect (and a lot of phone calls asking what to do when Word crashes!). Do you find anything similar?

3.) What do your parents think of you? I always get a lot of shit from my parents (again, probably more playful than anything)...but they admit they only tease me because they don't understand what I do! How about your parents?

4.) What areas of computer science do you find interesting?

Hmm, that's all I really want to ask. Thank you.

Generation5 - "...at the forefront of Artificial Intelligence..."

Dont go after high profile projects like Mozilla and the linux kernel for your first project. Those type of projects attract a large number of the best programmers, leaving few interesting jobs for the less experienced. If you do want to work on Mozilla, etc, it will be appreciated, i'm sure, but it will not be nearly as satisfying to you (well, thats my experience, at least)

You should decided what you're interests really are. If you are more interested in games, there are plenty of games being developed that could use any help that can be gotten. If you are interested in AI, check sourceforge or Generation 5 there's always something going on.

You'll benifit from a smaller, less important project because the project leaders will be more willing to take you under their wing and help you smooth out your programming skills. There's also the equivalent of a small fish in a big pond, or a big fish in a small pond... you'll be more vital to a smaller project than say Crystal Space or, especially, kernel work.

One of the problems with neural networks (both in biological and artificial systems) is that slightly different setups can yield surprisingly different results. There have been several studies on twins (genetically equivalent), and they exhibit very different brain activity patterns (not that surprising, look how many identical twins also differ in build, personality, etc).

You can try a similar thing with simulated artificial neural networks: set up a simple back-propogation neural net with different initial conditions and give them the same training data. Now watch for subtle differences in the output. Bigger neural networks give even less subtle (ie often rather large) differences. So, the machine would have to be trained for an individual.

Just as your neural network can learn to process its own data uniquely, so should a simulated artificial neural network (See link for a good intro). Our detectors of today (NMRI) would only need to be improved by an order of magnitude (I've used a NMR machine in a lab that was very similar to Purcell's origional experiment and have seen the improvements in commercially available systems for hospitals). Our processing power would have to be improved by several orders of magnitude (I am basing this off of papers I've read on large simulated artificial neural networks ~1,000,000 nodes, which are still dwarfed in complexity by the human brain).

So, given enough training, a computer/NMR device ought to be able to learn your commands and distinguish between subconscious and conscious desires.

On the other hand, some philosophers thought that given enough training that men would eventually understand women...