R&L Thread 4: Semantic Engines and the Turing Test
Reminder: Tonight (Thursday) @ 6pm in Greg 217: I will be giving a short paper entitled “What is the Internet?”
Tomorrow @ 3pm in Greg 213, Bernard Reginster “On being Looked At: Sartre on the Significance of Alienation”
Attend either talk (and say hi to me!) and I will mark you down for extra credit.
Here are the prompts for this week’s R&L thread. Respond to this post by class on Tuesday for participation credit for the week.
1) Is the mind a semantic engine? Why or why not?
2) Argue for or against any of Haugeland’s X-factor objections to the central thesis of cognitive science.
3) Can machines give a damn?
4) Is indistinguishability from human behavior enough for intelligence? Why or why not? Is the imitation game fair?
5) Is the Lady Lovelace objection fatal to the possibility of artificial intelligence?
6) Can machines learn? Are learning machines autonomous?
26 Responses to 'R&L Thread 4: Semantic Engines and the Turing Test'
Leave a Reply
You must be logged in to post a comment.
4. Indistinguishability is enough for intelligence. According to aristotle’s definition of phronesis, intelligence involves science and craft. A formal system is utterly that: the combination of science (/technology), and craft. Thus the intelligence game is fair, because if the product of science and craft can work to fool a human into thinking it’s a human, than it is intelligent.
Elena Solomon
27 Sep 08 at 6:38 pm
Aristotle’s definition of phronesis emphasizes that it is the capability to consider an action or method. While the imitation game demonstrates that technology can fool humans, the machine is not truly considering its actions. The program is such that each input has only one output. There is no thought concerning the action itself.
To play the devil’s advocate to summarize my views of imitation: let it be such that there is a man standing behind a curtain along with a perfect sculpture of himself. They are illuminated from behind, casting shadows on the curtain. To an observer on the opposite side, both shadows would appear to be of the same man. No sane person, however, would say that the statue was intelligent. Technology has improved to the level where we can replace the sculpture with an advanced machine, programmed to utilize optical sensors to follow the man’s movements. The machine, still, is unintelligent - it cannot consider its actions, and performs only as it is instructed. Imitation and intelligence share but one quality: the letter with which their linguistic representations begin.
Luke Kaiser
27 Sep 08 at 8:55 pm
@Luke:
“The machine, still, is unintelligent - it cannot consider its actions, and performs only as it is instructed.”
How can you tell the difference, if you can only observe its behavior?
Daniel Estrada
27 Sep 08 at 9:24 pm
Whether or not I perceive the method by which the machine moves does not change the truth concerning that method. Were I behind the curtain or, better yet, the programmer of the machine in question, I would certainly know that the machine acts not of its own volition but according to a set of rules I have established. We have a window into the machine’s “mind” that we do not possess with respect to humans.
Luke Kaiser
27 Sep 08 at 10:05 pm
First and foremost, see my post “Semantics and (Artificial) Intelligence” I look forward to all the opinions and arguments it raises in relation to this thread.
@ Luke:
I agree with you. However, I see what Daniel is saying. How do you see into the mind of the machine? Let’s say that in 50 years, science has advanced to the point where there are human-machines amongst us. In addition, it is impossible to differentiate the difference between a machine operated being in your class and a human in your class. How do you know what is going on in the mind of the machine? Is it anymore accessible than the mind of a human being?
Lihy E.
28 Sep 08 at 1:23 am
“How do you know what is going on in the mind of the machine? Is it anymore accessible than the mind of a human being?”
Those who write the programming for the machine surely know what’s going on in the machine’s mind. In that respect, it is far more accessible than the mind of a human being. We know some things about nerve cell interactions, but consciousness is still largely a mystery. The robot’s programming is an accessible system that allows us to predict - with perfect accuracy - what the robot will do given a stimulus.
Luke Kaiser
28 Sep 08 at 11:27 am
@ Lihy & Daniel - I guess what really separates humans from machines is the ability to judge based on morals rather than just the circumstance that is presented to it. If you put a computer in a courtroom, and presented it with someone who murdered another man, the computer wouldn’t be able to make a moral decision. It would strictly see the man as a murder of any sort. Say the man in the courtroom is a father who killed a robber as he broke into the house. Would the computer be able to say “Oh well he was protecting his family so he’s off the hook.” I guess the way I look at it is computers can do no more than strictly regurgitate information that we have pumped into them.
Another major thing that separates humans and computers is the “originality factor”
Would a computer be able to write a bestselling book, create a priceless work of art, or compose a symphony? A computer can’t have an understanding of poetry, or appreciation for music. Nothing that a computer does is from the computers own thought or action, it must first be given a set of instructions or rules no matter how complex the computer is, it really has no free will of its own. It’s nothing more than a very sophisticated marionette.
Austin Maske
28 Sep 08 at 6:18 pm
Question 3: Can machines give a damn?
Essentially, what you’re asking is whether or not machines can care. (Haugeland’s third X Factor)I don’t think that a machine can care. If it does, that is just based on the way it was programmed to respond to various stimuli (as Luke was saying). I think that what machines lack is a part of the human psyche that incites feelings such as empathy, longing, etc. that can’t really be programmed. If you can’t empathize with another being, how can you understand them? Interestingly…I feel that there are humans that are lacking in this field as well. Does that mean that they are no better than a machine?
Lihy E.
28 Sep 08 at 7:03 pm
On machines giving a damn: So what exactly is it to “care?” The the majority would say that it is to feel some emotion in response to something. Lihy mentioned that a machine would only care depending on the way it was programmed to. But in a way, our minds were sort of “programmed” as well: that’s why we respond to things the “natural” way, or the common way. If a loved one died, we’d likely be sad, that’s just how most were programmed. So perhaps this is not a good definition for caring. You might say that it has to do with choice, that people can choose to move on or what have you, but ultimately that is simply the way their mind works. Some peoples’ minds are wired to respond one way, while other peoples’ will respond differently, meaning that some people will move on eventually, and some will not. Again, each person does what is natural to them. Is it safe to say then that a robot that is programmed to feel happy all the time is just “naturally” that way? And what if you program them to respond the way humans naturally do? Then maybe they’re just like humans, except that we have direct access to their “minds.”
On a different thought, I think what captures a lot of the difference between humans and robots is the concept of choice. A machine will always be within the boundaries of rules that the programmer designed. Even learning robots will have some set of rules that can’t be broken, and so their decisions are very limited. A human will also have many rules, but can still consciously choose to break them- it happens all the time. Humans can ultimately break any rule that is either set for them or that they set for themselves (outside of breaking the laws of physics…but that would be pretty rad).
Gautam Srikishan
28 Sep 08 at 8:17 pm
Mostly In response to Gautam,
I’m one who likes to imagine the possibilities of future robots as mostly unlimited. If you take time to think about it, there is almost always a way around the problem. To the issue of robots making a choice: robots can be programmed with random possibilities. Say a robot has a random number generator (most graphing calculators have random number generators), to every number that might be randomly chosen, an action or reaction could be assigned. The robot then in fact would be making a choice, in the sense that it picked what action to follow, out of a set of actions the programmers gave it. Now, say the robot can in fact learn from experiences, like iterations with humans, or other robots, or the environment, or anything; and then the robots would assign outcomes from these experiences to a random number or some other system. Now the robot has more actions to randomly choose from (it could get to the point where the choice isn’t random, but it still the robot’s decision. I just like using the example of something like a random number generator, because it makes the idea of a robot’s ability to choose easier to comprehend). A robot could even choose to break the rules it was preassigned, because it has learned to do so (it would be following the rules of learning, but perhaps breaking rules that it was originally programmed with). Now who can say the robot cannot theoretically choose for itself?
Brad Thompson
28 Sep 08 at 10:17 pm
While random number generators enable robots to make choices, it does not enable them to make decisions; the latter requires consideration that a random choice surely does not involve.
Additionally, “say the robot can, in fact, learn from experiences” is an incredible leap. You could program a robot to remember what random decision it made given a certain stimulus, but that would only make its lack of true learning more obvious. Take a robot car, for example: the car approaches a wall, and its frontal sensor detects the wall as an object in its path. The robot has been programmed to use a random number generator to choose an action, and happens to choose “Full speed ahead.” If the robot had been programmed to associate that response with its stimulus, the robot car would accelerate into every wall it encountered. Imagine what this would look like you saw a human performing similarly. The robot hasn’t “learned” anything since its response would be poor every time, instead of actually adapting to its surroundings.
Luke Kaiser
28 Sep 08 at 11:02 pm
*when I said “iterations” in my last post, I meant to type “interactions” *
In response to Luke:
I never said the robot would have to use a random number generator for every single decision. Nor was I equating random number generation with any aspect of the learning programming. I was only trying to use learning as an example (I assumed a theoretical learning robot, and then I gave the robot hypothetical random number generation so that it would be able to make a “choice”) of how a robot would be doing actions it learned from observation, and then turning that into one of it’s list of choices. The random number generator wouldn’t be the deciding factor for every decision, but rather for smaller decision, like randomly picking what shirt to wear that day (yes, robots should wear shirts). So, I wasn’t trying to leap from random number to learning, or say that random generation would be the reason for every decision. It was just an idea of how a robot could make a “choice.” I absolutely agree that this type of choice isn’t a decision however. hopefully there would be some better programming for decisions like driving than random actions.
Also, I found your image of a robot accelerating into every wall it encountered very amusing. good stuff.
Brad Thompson
29 Sep 08 at 2:19 am
First and foremost, after hitting the fifth wall, that robot wouldn’t be accelerating much at all.
It’s very important to make a distinction between choice and decision. Making a choice could be random based on anything random. A decsion, however, is based on understanding. DING DING DING Buzz-word…with Aristotle’s definition of understanding, even an animal isn’t capable of understanding (how can a robot be?).
Lihy E.
29 Sep 08 at 4:38 am
I have a har time believing that animals are not capable of understanding somethings. Such as when a person says “sit” a dog will sit, or “fetch”, or “stay”. The dog has an understanding of the word that it’s being presented with… oh right, and theres always this example http://www.youtube.com/watch?v=R6KvPN_Wt8I
Austin Maske
29 Sep 08 at 2:16 pm
I will at some point address topic 6…
I think the reason why Thursday’s class ended in the middle of a rather heated debate was because we were forgetting that such a discussion about computers must refer to the possibility of intelligence. i.e. There may not be a computer in existence now that can act autonomously, but is there a possible computer that could do this? Or, alternately, is this an impossibility? It is more hypothetical than we were making it out to be. However…
That being said, I can’t conceive of a completely autonomous computer. Yes, Daniel explained that computers can learn certain tasks, but that is based completely off their programming. Computers can only learn in so far as a human at one point predicted the ways in which they could adapt to situations. And because humans are necessary to program these functions into a computer, said computer could never be completely autonomous.
Of course, one could make the argument that in this nitty-gritty definition of autonomy, humans aren’t completely self sufficient either. While we are not directly programmed (at least not in this country…I think), we still cannot operate on our own. A child devoid of human contact will not learn or develop in a healthy manner. He/she needs a different kind of programming, but still cannot exist autonomously.
Juxtaposed in this manner, it would seem that computers can learn in a different manner from humans, but that the definition of autonomous is ambiguous enough to include both classificiations of “intelligent” beings.
I still find it hard to believe that a computer would ever do something unexpected, which is part of being autonomous. Also, making mistakes is part of the genius of autonomy, at least for humans, considering that this often leads to unanticipated new avenues of exploration.
But machines are okay too.
Calli Leventis
29 Sep 08 at 5:05 pm
I will go a step further than Calli and say that computers cannot learn. They are pieces of hardware that have programs written onto them by computers. All a computer is able to do, is what it was programmed to do. It has no free thought. It doesn’t even know what it’s doing!
Now as for a computer learning independently, I again say no. Now in order for something to learn, technically it must remember as well. So a couple instances of a computer ‘remembering’ is programs you use most often, it will have them on the start menu. Another is in a video game and the AI ‘learning’ your moves, i.e. you always jump left insteed of right and the AI picks up on it. Neither of these examples are of a computer learning independently. All they are doing is exactly what they were programed to. No freewill. Just something the computer remembers for a time until you start using other applications/movements.
Jason Blumstein
29 Sep 08 at 8:35 pm
In response to whether or not machines can “give a damn,” I do not feel that they can. Deep Blue did not rejoice over its victory. Even if “giving a damn” is a response to programming (as mentioned in some earlier posts), ie respond in this way when presented with this situation, the machine is simply presenting the required output. There is no emotion.
In response to all this talk of autonomous learning… I think that it is possible for machines to hit a “cascade point,” ie when a machine essentially learns how to learn independently. For example, when the machine has all the programming foundation given to it by humans, and it can take that a step further, to create something new in an intelligent, rather than random, way. Therefore, I don’t think that machines can act intelligently about anything that is not based on the initial rules of its human programming. I don’t have the technical vocab to explain my idea fully, but I hope someone sees what I’m getting at?
Rebecca Spizzirri
29 Sep 08 at 9:32 pm
Machines can give a damn to an extent, but it’s limited by its programmer. The machine can pretend all it wants to, but their is no emotion, no passion. It’s applying the thoughts and suppostions of its creator to any situation that calls for judgement, so I guess you could say that machines that “give a damn” are literally walking plagerisms, but then again, aren’t we all? Do we not take the inherited beliefs of our creators (parents, God, etc.) and apply them to everyday situations much like a robot would? Even if we don’t, we learn and adapt, just like a robot with those capabilities could. However, paraphrasing what Luke and Austin said earlier, there is no sincerity in the robot; the robot will never evolve like a human would, and it can’t be INVENTIVE only INNOVATIVE.
Roy Bell
29 Sep 08 at 11:13 pm
excuse the grammatical errors.
Roy Bell
29 Sep 08 at 11:27 pm
In the sense that Haugeland suggests, I kind of agree that we are semantic engines, but you have to break down each word Haugeland uses to realize his intent at this claim. According to Webster, semantics is the “of, pertaining to, or arising from the different meanings of words or other symbols,” but in every variation of the word engine, the word machine is used to describe it. These are words defined by the human race, not the machine race. Humans and machines may share similarities, but they are not completely synonymous, so I wouldn’t go as far as implying that our minds are engines; that’s a bit extreme, Haugeland.
Roy Bell
29 Sep 08 at 11:50 pm
Machines can’t give a damn like humans do. Machines can be programmed to “feel” things, but it’s artificial. A machine can never feel what it is like to get a cut or bruise because it is not made of flesh and bone. A machine could be programmed to have empathy for somebody who is injured, but would never be able to comprehend what it is like to feel that kind of pain.
Dan Pierson
30 Sep 08 at 1:28 am
I just don’t think we have any way of knowing. It is entirely likely that, as Roy was saying, we are just programmed the way computers are by our creators, and we don’t have the technology yet to understand how that is possible. But even if we had the technology, I can’t think of a way we could be certain that computers “understood” things. There could very easily be some intangible essence that creates consciousness, too. All I know is that brains work a whole lot like computers, and that the more we learn about the makeup of the world and ourselves, the less we attribute to intangible forces.
Katherine Anderson
30 Sep 08 at 1:37 am
Machines cannot give a damn. This is impossible. No matter now hard a programmer works to give a machine “feelings” or “emotions”, they will not succeed. A machine can just do what it has been programmed to do, and then does it without any thought or feeling. It is impossible for any machine to get tired or bored or simply have any kind of emotion at all. Machines cannot care when something goes wrong and, on the other side of things, even when it gets things right, it feels no sense of pleasure or pride, or delight in those accomplishments. This holds true because machines have no spirits or souls, and no wishes, desires, ambitions, or goals. That’s why a machine will just stop when it’s stuck, for example if a program is written and there is a bug the program simply stops trying to run the program and a person must go back into the code and correct the issue. Machines stop trying when there is an error, but a person will struggle to get something done. An obvious difference in the way a machine’s “mind” works and a human being’s mind works. Simply put it is impossible for machines to “give a damn” as they cannot have any emotions, positive or negative.
JD
30 Sep 08 at 1:40 am
A computer is programmed in the same way a human being is socialized. It can only be told what we already know, and it is apparent that both computers and people need a stimulus to act or communicate. What makes a human so superior in our minds is that we are also humans. A computer is a creation of ours, so theoretically it shouldn’t be able to exceed us, but Kasparov lost to Deep Blue. We could calculate pi to the 1 millionth number, but a computer does it for us so much faster. Now due to the integration of the computer as a quintessential component of our society, it seems the computer has taken a part of us away and now is inherently a part of us.
Ben Tondera
30 Sep 08 at 12:09 pm
A computer isn’t programmed the way we are socialized. We are both programmed, but the socialization aspect to a human is like run-time adaptation for a computer. We are all hard-coded with a basic set of instructions from our genes, and we pick up more instructions and rules as we live. It’s like at birth we’re all flashed with whatever firmware we have, and it’s all created to write our new code. The difference between us and computers is the means by which the baseline code is developed, and the medium of computing. We are made of neurons and pulses are charge imbalances..computers are silicon chips and wires. I believe that computers will become fully conscious themselves some day, either before or after we determine conceptually what consciousness is. I don’t believe the ability to explain the concept is a prerequisite for creating it in this case.
Joel Ferm
30 Sep 08 at 12:28 pm
Yea, human socialization is an ongoing process. True, our parents teach the basic rules to us, and the more complicated ones as well, but then we get older and learn from our friends and the media. We constantly adapt ourselves to be the person we want to be, and our methods of interacting with others follow. We begin to teach ourselves how to socialize, which a computer could never do because humans programmed it. Any ‘teaching’ it does would just be a response from a program inside it telling it what to do, a program created by man. A computer can never learn how to socialize well enough to interact with humans.
Elena Solomon
6 Oct 08 at 11:39 pm