I am sure any AI system will be more than one program modual. The knowledge base(KB) and knowledge representation(KR) and any programs for transfering the knowledge into the analysis modules... Those might be scored on time or instruction executed.
I prefer to use the instruction executed versus time and then any given AI system would be independent of time on any given platform, for all pratical purposes. What are the Seed AI moduals? And how would you score each for how well each modual performs? If the AI system was doing Stock Market investment buy/hold/sell decisions we could see how any one or any mix of the programs modules performed over any time series of data. The same does not exist for non-investment type performance analysis. Knowing what a good decision is might not become obvious until the two branches are searched. Dan Goe ---------------------------------------------------- >From : James Ratcliff <[EMAIL PROTECTED]> To : [email protected] Subject : Re: [agi] How do you evaluate?... Reward & Punishment? .... Motivational system Date : Mon, 12 Jun 2006 07:25:47 -0700 (PDT) > Dan, > Possible plans considered would be projected forward and given a GoodValue that will try and min/max itself to find optimal paths: > > GoodValue = a*alive + b*health + c*wealth + d*enjoyment + e*learning + f*friends + g*pastplans - h*time > : Where staying alive is paramount right now (a is highest parameter), and each other element has an effect, health is staying healthy and undamaged, welath is money and object accumulated minus cost of an activity, enjoyment is activities that an entity enjoys, learning is a metric for promoting exploration of new experiences, and friends is a general metric for promoting people to like you and keeping from harming people, past plans is an indicator of repeating patterns of actions, and time subtracts the amount of time taken by activity. > > So any system would run best with the highest score fro GoodValue. > > This is simply an intial GoodValue equation as well, that is modifiable in its variables, and can be added to as the AGI goes along. > > I take my inspiration in part from many old style MuDs where there is a fairly rich, yet finite world and set of interactions. I think we should take something like this as our model, even though it would not be a full AGI, and then strap on a very advanced learning system, that would allow an AGI to aquire any new information needed about the world through interactiona dn diretion by humans. > > The GoodValue above is a measure of what action the AGI should take next. > > James Ratcliff > > [EMAIL PROTECTED] wrote: > How do you score any given AI System test run? > > Dan Goe > > ---------------------------------------------------- > From : James Ratcliff > To : [email protected] > Subject : Re: [agi] Reward versus Punishment? .... Motivational system > Date : Mon, 12 Jun 2006 06:13:45 -0700 (PDT) > > Will, > > Right now I would think that a negative reward would be usable for > this aspect. I am using the positive negative reward system right now for > motivational/planning aspects for the AGI. > > So if sitting at a desk considering a plan of action that might hurt > himself or another, the plan would have a negative rating, where another > safer plan may have a higher rating. > > One possible thing here as well is to have asmall random value added, > so that even though a plan has a suboptimal value, it woul dbe possible to > take that route instead. (maybe adding in the value of having a new > experience here as well) > > > > One important thing we will need here, is an entire KR that will > represent all the AGI's past actions, and a way to look back over the > actions and compare them to thier expected outcomes, and see why something > is different. (Reflection) > > IE If the robot proposes to cross the road at a point, sees it as a good > plan, and does it, but nearly gets hit by a car, it needs to be able to > look back over its actions, and determine that something was missing from > his equation, and try to add it back in, or ask a human for assistance, so > in the future, he can better handle this activity. > > > > James Ratcliff > > On Fri, 09 Jun 2006 19:13:19 -500, [EMAIL PROTECTED] wrote: > > > > > > What about punishment? > > > > Currently I see it as the programs in control of outputting (and hence > > the ones to get reward), losing the control and the chance to get > > reinforcement. However experiment or better theory would be needed to > > determine whether this is sufficient or negative reward would be > > needed. > > > > Will > > > > ------- > > To unsubscribe, change your address, or temporarily deactivate your > subscription, > > please go to http://v2.listbox.com/member/[EMAIL PROTECTED] > > > > > > > > Thank You > > James Ratcliff > > http://FallsTown.com - Local Wichita Falls Community Website > > http://Falazar.com - Personal Website > > Hosting Starting at $9.95 > > Dialups Accounts - $8.95 > > __________________________________________________ > > Do You Yahoo!? > > Tired of spam? Yahoo! Mail has the best spam protection around > > http://mail.yahoo.com > > > > ------- > > To unsubscribe, change your address, or temporarily deactivate your > subscription, > > please go to http://v2.listbox.com/member/[EMAIL PROTECTED] > > ------- > To unsubscribe, change your address, or temporarily deactivate your subscription, > please go to http://v2.listbox.com/member/[EMAIL PROTECTED] > > > > Thank You > James Ratcliff > http://falazar.com > __________________________________________________ > Do You Yahoo!? > Tired of spam? Yahoo! Mail has the best spam protection around > http://mail.yahoo.com > > ------- > To unsubscribe, change your address, or temporarily deactivate your subscription, > please go to http://v2.listbox.com/member/[EMAIL PROTECTED] ------- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
