John: There is absolutely nothing new here. ?? Nothing.

Ben: I wrote software to improvise music in 1994, using genetic programming
& fractals

I see the confusion here. It does indeed arise as I suggested from the misuse/ambiguity of words .

The kind of commonplace AI "improvisation" Ben is talking about here is not true improvisation. It is where you take a la GA, predefined chords/phrases etc and combine them in predetermined ways. There is nothing creative about this - nothing AGI. And John presumably is thinking in similar terms.

I am saying the difference between narrow AI and AGI is :

the execution of a ROUTINE, basically OLD course of action vs the creation/[true] IMPROVISATION of a NEW course of action

a narrow AI robot is programmed to perform a routine - such as a routine journey in a routine environment like a factory or warehouse - it does this by following an algorithm which dictates every step and every path it must take

this method only works in an artificial specially created/structured and totally familiar environment. The slightest new, unfamiliar alteration to the environment throws the robot completely. Hence such robots cannot deal with the real, unstructured world.

[a "conversational robot" like Siri is no different - it is following conversational prewritten routines - it has routine answers to routine questions.

The so-called "improvisational" music program is the same. It has routine ways of combining routine musical elements. Its "new" tunes are only trivially new combinations of the same old elements. It is not *generically* new].

An AGI robot will be programmed to create a NEW course of action - to take a NEW journey in a NEW unfamiliar environment that is fundamentally different from any environment it has ever been in.

Narrow AI is journeying with a route map. A narrow AI warehouse robot knows exactly where and how to go to find a given box of diamonds - it's in row G, and you go there via this and that aisle. It has a route map and an itinerary.

AGI is journeying WITHOUT a route map. An AGI robot will have to find a box of diamonds in a warehouse that is radically NEW to it. Let's say, it's the same warehouse, but in a totally disordered, higgledy-piggledy mess - an earthquake has thrown everything all over the place. "*Now* find the box." The box could be anywhere - and crumpled up/ deformed

So an AGI robot can have no algorithm, no itinerary. Just a vague idea of where to go. Our AGI robot will know something about travelling through warehouses and finding and lifting boxes, and some of the steps it may have to take, but it won't know this warehouse.

AN AGI therefore will have to work by true, adventurous improvisation -.

You choose a possible goal where the box might be, take a step or two towards it, then see where that gets you. If the way is blocked - can you move the obstacles out of the way? -That may well require *innovation* - trying new grips and holds to remove objects. If you can't, try reorienting. and following another direction to the goal... etc. etc.

Note: the robot cannot consult some predetermined set of options. THERE ARE NO ORDERED SETS OF OPTIONS IN AGI. it has to look around and find out what the options are in each new situation. It doesn't know what paths are available until it has got some way into the messy tangle of boxes, frames and rubbish. It won't know either what moves are available until it finds itself in a given position and can see how the obstacles all around restrict its lines of movement.

I won't go into more detail here because it would take a good chapter.

But the basic principles should be obvious - and i have already linked to one paper outlining an initial, limited move towards real robotic improvisation.

***Improvisation is the complete opposite of Turing** - of executing a predetermined formulaic plan. You don't follow a script, you write the script as you go along. You don't follow an itinerary, you forge one as you go along.

No one is doing this, John, and no one is saying this. No one is producing true improvising, exploratory robots. I don't mind you swearing at me, but if you're going to, bloody well produce some examples of your assertion. *Where* is this being said? And if you can't produce them, then have the courtesy to admit it.

You guys have just demonstrated that you completely misunderstand the whole concept of improvisation - precisely because AI is such a lying, cheating, word-misusing field and confuses everybody. It's not altogether your fault - but it's happening over and over here, because you're not really listening and just automatically trying to bat away every new idea.

AGI is what a NASA rover should *ideally* be - as distinct from the narrow AI, telecontrolled reality - a robot that you can tell to explore a NEW territory WITHOUT an army of programmers preprogramming or teledirecting its every step. Just give it an idea of what it must do, and then let it improvise its journey. That is the new revolutionary kind of robot we need.

The equivalent in a conversational AGI robot would be one who couldn't just tell you that you have to go to aisle 8 to find a box, but -could deal with NEW situations that may arise, just as human conversationalists can -.could make suggestions when you tell it there's a crowd in the way

A true improvisational musical robot wouldn't just combine predefined chords etc, it would adventurously play around with notes and create new, surprising phrasings and chords and combinations thereof - and might try to invent and incorporate new instruments and sounds, just as pop music does.

[Neither of these last two examples are practical AGI propositions in the conceivable future. Physical, exploratory robots, on the other hand, *are* practical propositions.]

True improvisation demands innovation and invention - introducing new elements - new moves, for example, new grips, new steps and treads in the course of a journey.

But it is not at all mechanically or computationally mysterious.

No, what I am saying is still new - and it is the secret of generality. If you can improvise a solution to one kind of problem, you can improvise a solution to any kind.




-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-c97d2393
Modify Your Subscription: 
https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-2484a968
Powered by Listbox: http://www.listbox.com

Reply via email to