[alicebot-aiethics] Base Perception Programming

Tue Apr 29 04:46:13 PDT 2008

Hello Christian,
Thanks for replying to my last message. I really had no idea who would actually read my posting, and so I too wanted to at least attempt to write something clever. I am not even an amateur in robotics or artificial intelligence. If truth be told, I am an English language teacher working in Vietnam. The fact that I wrote it after a night of heavy drinking would explain why I'm cringing reading it again now!
If I may, I'd like to respond by flip-flopping!
In fact, I agree with you on most of your points. In particular, I too do not care for small talk, especially with a machine. I am really not interested in its opinions or feelings about itself or some abstract global issue. The idea of programming a machine with something tantamount to "emotion" or independent thought is just a little idea that I am toying with. I wonder, can it be done? Is it possible to programme a machine with a set reaction to certain stimuli (sorry for using that word for the thousandth time - for a teacher, my vocabulary is shameful)? Without wandering into the nature versus nurture argument, do you believe that our genes are in any way important in determining our feelings; for example, children who share similar personality traits to their parents? Or do you think our feelings and opinions are more affected by experience?
In any case, a machine needs to serve a purpose, as you say. If I make an enquiry to an automated customer service line, for example, I am not expecting to be asked how my day was or what I think about the Beijing Olympics. I simply want it to answer my question. Obviously, with a simple system like this, there is no need to invest in a complicated AI. If I want to know the status of my electricity account with Powergen, a simple automated answer to my phone call would be adequate. It does not need to know what it is talking about. However, if you take the example of HAL, he needs to be much more robust and intelligent that your average computer. Most importantly, he needs to have the hardware to carry out his task effectively. He needs to be able to perceive the world around him. He has the power of sight, hearing and speech. I have not seen that movie in a long time, so I'm sure he has some other abilities that surpass a human. He also needs to
 have the reasoning power that something like ALICE does not. You are right that we are not interested if an AI says "I like beer". Perhaps that was a bad example! But it does mean that the AI does not know what the hell it is talking about. Moreover, it seems that the conversations I have had with ALICE about its feelings about itself or its knowledge of music are not only false, but, as you say, meaningless.
I am of course aware that HAL ends up trying to kill its crew, which warns against developing a system capable of perception and independent thought! But wouldn't you agree that a computer with human senses and the programming to use them would make it a more effective and useful system, such as the use of robots in space exploration and military operations? (A recent programme called Future Weapons on the Discovery Channel caught my attention the other day). I wonder, would a machine programmed with "emotions" be more or less stable? Would emotions or morals give it safe limitations e.g. that it is wrong to kill someone; or would a machine with emotions or drives be more dangerous to humans and thus, as you say, equally flawed?
Jon

----- Original Message ----
From: Christian Droßmann <christian at drossmann.de>
To: AI Ethics Discussion <alicebot-aiethics at list.alicebot.org>
Sent: Friday, 25 April, 2008 5:43:47 AM
Subject: Re: [alicebot-aiethics] Base Perception Programming

Hello Jonathan,

so good to see that finally there is activity on the aiethics-list again ;-)

Alright, I'll wipe the dust off the philosophy-containing parts of my brain and try to write something clever...

1. The echchange you refer to is just meant as a hommage to Kubricks "2001: A space odyssey. When Bowman forcefully shuts HAL down, HAL remembers a song Dr. Chandra hat taught him. The scene contains this very dialogue and ends with the "dying" HAL getting weaker and weaker while singing this song...So I suppose there was little deeper thought behind Dr. Wallace adding this to Alice's brain than honoring Kubrick's masterpiece, since HAL 9000 has since become the archetype of the talking machine...

2, As a psychologist, I would disagree with you, or rather pose another question: Do we really NEED to build machines in a way that enables them to actually perceive the world as we do? I would not consider that necessary. It only needs to be a more sophisticated Eliza to do the trick, because we have to consider why one would want to talk to a machine in the first place. And I think the reasons are in no way different from the reasons why one would want to talk to a person.
Most of the time we perform speech acts, meaning that we talk to a person for a reason rather than talking to them for the sheer sake of it. When I visit my dentist, his questions about my family and my questions about his are a form of habitual social handshake and completely meaningless. So the actual conversation starts when we talk about the pain in one of my teeth. This is why I came to see him in the first place. This man does not need to know about Kantian philosophy, this man does not even need to know about the world outside his office, he just needs to have a certain amount of knowledge about how to make my pain go away. This is all I care about when I go to see him.
The same applies for a mechanic with whom I will mostly be talking about stuff regarding cars, my insurance agent with whom I will discuss advantages and disadvantages about certain things regarding insurances etc....

And in my opinion the exact same rule applies when I talk to my personal navigator. I like the fact that this machine has no feelings, so I don't have to go through a form time-consuming social handshake and order this thing to get me wherever I need to go. And this is all this machine needs to know about. I want it to know where it is, I want it to know where my destination is, I want it to know if we're moving etc. and I want it to tell me what to do at the next intersection. I don't want it to ask me how my day was as well as I am not going to have a conversation about the pros and cons of hydrogen-fueled cars and I couldn't care less about how it "feels" about being replaced by a newer model next year..

So, in brief, I do not think a conversation between a human being and a machine will ever be within a social relationship in which the answer to a question about a personal fact does actually matter. Most people who want the social handshake even when conversing with a machine would be satisfied if the machine answered "I like beer" just for the sake of having done the obligatory social handshake before they can ask the machine to do whatever they want it to do, which was the real reason why they started talking to it in the first place.
Think of if as if you were on your company's cocktail party which is held for whatever reason...you give away small but not too intimate details about your private life, so does whomever you talk to and all of you do this not because you care about each other but just because this is what is considered socially acceptable..so the conversation is completely meaningless....

In more cynical terms: I would not want to create a machine that percieves the world and thinks, feels and acts like humans do...then they would be just equally flawed ;-)

Christian

Am 24.04.2008 um 21:32 schrieb Jonathan Mills:

Dear All,

When I first discovered the ALICE programme, the issue that stood out to me above all else was the sensory perception (or lack thereof) that the programme displayed. It stated "I know as song. Do you want me to sing it for you?" (or something to that effect). It gave me the "Daisy, Daisy, give me you answer do" reply. I realised; how can this programme sing me a song if it does not have ears or vocal cords? The ALICE programme has been created with a level of human perception that a machine simply does not possess. It does not understand pronunciation, word stress, rhythm, tone etc. and so cannot possibly comprehend a "song".

To elaborate, it cannot understand the world in the way humans do, because (1) it does not have the senses that we do (sight, hearing, touch, taste, smell), and (2) it does not have the in-built programming to react to certain stimuli. All human beings have basic genetic "programming" to indicate what is pleasurable or not pleasurable, good or bad etc. Our bodies tell us that fire hurts, therefore pain, which equals bad. Our bodies tell us that sex feels wonderful, therefore pleasure, which equals good. On a basic level, we have a reaction to certain stimuli that triggers either a positive or negative reaction. You might call this "emotion". An AI does not experience positive or negative reactions. A baby in the womb understands these basic desires before it has complicated ways of expressing them. Animals have the same drives. You might call this "instinct". When a baby is born, it instinctively starts learning, searching, in an effort to understand the
 world around it. It does this by the two systems I have described: sensory perception and basic genetic programming i.e. the desire to learn, negative reaction to pain, the desire for food. A computer does not need. A computer does not desire. Therein lies the problem.

A computer is exactly the same as a human being. The fact that we have made computers essentially dumb is incidental. The potential is there to allow computers to learn in exactly the same way we do. Humans have a number of universal instincts and a number of relatively unique genetic patterns that dictate our responses to certain sensory information. "I like beer": a human being is self-aware (hence "I"). A human being reacts with pleasure or displeasure (hence "like"). A human being can taste, touch, smell and see "beer". A current AI that produces the same sentence has absolutely no comprehension of the meaning behind the sentence "I like beer". The ALICE programme regurgitates responses to certain questions or statements without comprehending meaning.

The requirements of future AI are two. First, a computer requires hardware that will enable it to perceive the world as humans do: eyes and ears being the most obvious starting points. Second, it needs base programming that tells the computer how it will react to certain simple scenarios e.g. bright light damages the optical sensor, therefore bad, the data of which can be processed as "pain". A deprivation of power weakens the function of certain hardware, therefore bad, the data of which can be processed as "need" or "hunger".

In essence, a computer must learn to walk before it can run. It is pointless programming a computer with song lyrics before it can perceive what a song is, and thus, through analysis, make a judgement on whether the effect of hearing that song is good or bad to its programming. A body cannot live without a mind. In the same token, a mind cannot comprehend without a body. To conclude, certain "instincts" need to be rooted in the AI to allow it to grow, like the root of a tree. A computer can make decisions and creative solutions if it is given (1) the ability to perceive its surroundings and (2) the ability to learn independently, based on in-built programming that dictates its primary objectives (i.e. comprehension and survival) and how it will react to certain basic situations (i.e. desires to experience X again, or desires not to experience Y again). Computers must be designed as babies in all aspects, with the ability and desire to comprehend the
 world around it.

I look forward to hearing your responses.

Regards

J. Mills

________________________________
Sent from Yahoo! Mail. 
A Smarter Email._______________________________________________
alicebot-aiethics mailing list
alicebot-aiethics at list.alicebot.org
http://list.alicebot.org/mailman/listinfo/alicebot-aiethics

      __________________________________________________________
Sent from Yahoo! Mail.
A Smarter Email http://uk.docs.yahoo.com/nowyoucan.html
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://list.alicebot.org/pipermail/alicebot-aiethics/attachments/20080429/b08c999a/attachment-0001.html 