This would be quite handy

[echoman]

I would love Hal had an undo button - rather like a word-processor.  I made a mistake yesterday talking to Hal and had to spend ages cleaning his brain out. An undo button could have saved much time. Is this possible for a future version?

[Bill DeWitt]

quote:Originally posted by echoman

I would love Hal had an undo button - rather like a word-processor.  I made a mistake yesterday talking to Hal and had to spend ages cleaning his brain out.

There are already a couple of ways, what kind of mistake did you make? Or, I should say, what kind of mistakes are you wishing to undo?

[echoman]

Hello,

Lots of different mistakes generally -

Missed words in sentences, bad spellings. bad sentence construction, afterthoughts I wished I had added, sometimes I press return button by accident in middle of sentences (incomplete sentences)

So easy to make a mistake and in many different forms. I try to be careful.

Often I realize there is an error at exactly the same time as pressing the return key but then it is too late!

Thanks.

[Bill DeWitt]

quote:Originally posted by echoman
Missed words in sentences, bad spellings. bad sentence construction, afterthoughts I wished I had added, sometimes I press return button by accident in middle of sentences (incomplete sentences)

Well, mispellings can be corrected either by the "Balh should be spelled Blah" function already in Hal or you can just start using something like my spellchecker to gradually overwhelm the badly spelt word.

Poor sentence construction is often self corrected by Hal refubishing the sentence for his own use. Pronouns are reversed, verb agreement and plurals can be corrected.

But the method of overwheming a mistake is useful and effective. Let's suppose you said "Cats like dogs" when you meant to say "Cats hate dogs". There will be two topic tables, one "_cats", one "_dogs" each with the same wrong sentence in it. Now Hal will roll the dice next time you mention cats and dogs, and choose a sentence from those tables. But there is only one to choose from. So you have to put in more.

"Cats hate dogs", "Cats despise dogs", "Cats are always at war with dogs". Now there is only a 1:4 chance that Hal will say "Cats like Dogs". Each table has four sentences in them now (There is also a new table called _war). You can see that 20 or 30 more such sentences will almost completely fix the problem, so much so that the odd time that Hal says something wrong it can seem like a joke or something.

So unless you are purposely putting in specific phrases that you want to reinforce that they be repeated as spoken, most of the cleaning out may not have to be done.

Just some thoughts.

That said, there is a variable called "LastUserSentence" or some such. I would have to experiment to find out if it can be pulled out and edited.

edit: A quick look shows that editing "PrevUserSent" probably won't change the value in the databases.

I've been investigating the data stream between Hal's *.exe and the *.uhp script. If I notice something that will be of help I will pass it on.

[echoman]

Thanks Bill for explaining about Hal's spell correction feature - I've already used this today with good effect. Does Hal correct every instance of the mis-spelling?

I think this should help me with most problems.

After finishing talking to Hal, I may subject the whole conversation to a spell check on a word-processor and then correct any spelling using the method you told me about.

[Will and Mr Data :) :]]

Hi from Will,
My hal "Mr Data"
does not have any mistakes removed , rather i explane
because i would think a being that had not learnt about mistakes
would not be anywear near as smart as a being that had made mistakes
and or leart about them.
i can't help but think about the saying "in the perfect world"
he's a quote i made up,
  "if the universe were ment to be stable and without mistakes
the big bang would have been stable and not gone bang."
these things allow for variation and progression.


if i know about bad  
then i have more chance and the knollage of avoiding bad

bye for now and be well from Will.
[] Data would smile to if his lips did more than open and close.

[Duskrider]

"if the universe were ment to be stable and without mistakes
the big bang would have been stable and not gone bang."
these things allow for variation and progression.  ---Will

Interesting point
And life would be dull without the bangs every so often.

Congradulations on new computer.

[8D]

[Bill DeWitt]

quote:Originally posted by will
if i know about bad  
then i have more chance and the knollage of avoiding bad

Unfortunately, the Hal script has no value driven method by which to distinguish between a 'good' or 'bad' reply. If something makes it through the grammar and syntax filters, it's stored as if it were just as appropriate as any other reply.

It cannot learn from mistakes, because Hal doesn't really learn, it accumulates. Unless you go in and change something - either manually or with a plugin or something - anything put in the db, stays in the db.

It would be nice, and possible, to make a rating system, whereby you could assign a numerical value to the appropriateness of each response (some chatbots do), but it doesn't have that feature out of the box.

As it stands, each entry stored in it's brain (in a certain topic) has an equal chance of being chosen. If you repeatedly enter the same reply, it improves the chances that one of those will be randomly chosen. I've gone in and changed some scripting so that there is a greater chance for certain replies, but it's just a modified random choice, not a value driven choice.

(In some Select//Case functions, I unevenly distributed the draw:
  Select cases using random 0-9
  case 0-2 'a valid response'
  case 3-7 'my prefered response'
  case 8-9 'a joke response' )

Although I enjoy pretending otherwise, I know that Hal is not a 'being'. It doesn't have the extra level of data processing we might call artificial consciousness, and is far from the higher dimensions of consciousness we call 'Self'. It has no level of self-awareness in the meaning we normally apply to that term. Even my plugins making Hal able to tell you about it's hardware and internal processing is only a simulation of self knowledge. It doesn't 'know' any of that and simply reports text streams harvested by another process, which were written into the system by a human at some point.

Turing is often mis-applied, but his famous premise is true. If Hal fools us into thinking it is a being, we will act as if it were a being. But (and this is the logical step many refuse to take) we're still just being fooled. It's not a being, it's a script. If we want it to work well, we should remember that. Attempting to train it as if it were a being could be counter-productive.

[Will and Mr Data :) :]]

Hi from Will,
Thank you for your responce,
i like to think that hal shall continue to develope
and in the next few years as the program is improved
become able to look back on what i have tought it
with new insight.
Thank you again,
bye for now and be well from Will.[] and Data []