Could we handle an AI?

[Solece]

If only I could use fileWrite to write into my Expression2 folder, I could in theory, use reset() after writing to an existing E2, and make an AI capable of coding itself! Now that I think of it, the check for fileWrite is probably clientside so....
MWUAHAHHAHAAHA!!!

Edit: I was right, the check is clientside, now I can start work on some plan here, assuming reset() doesn't just reset the chip, but also recompiles the file, then it shouldn't be hard to make a self coding AI.

[Vbitz]

ouch, a polymorphic expression 2 ai, add a strong auto coder... use a ragdoll to turn it into a bot, ouch.

[Lyinginbedmon]

Edit: I was right, the check is clientside, now I can start work on some plan here, assuming reset() doesn't just reset the chip, but also recompiles the file, then it shouldn't be hard to make a self coding AI.

First and probably biggest problem is that you need to teach the chip to code E2. That's difficult even for some humans.

I mean, they're usually not very smart humans, but human intelligence is kinda the baseline for AI as it is.[highlight=E2]@name Holopattern Generational Evolution Prototype
@persist [Angles Pop Val]:array Gen State
@persist Start Age Old
#Filtering variables
@persist P Vecs:array
#Reseeding variables
@persist [Res As Bs]:array A B
#Stats
@persist G Kill Born

if(first()){
#Holos at different angles indicate different attributes
AnglesushAngle(ang(45,90,0)) #Visual acuity
AnglesushAngle(ang(135,90,0)) #Efficient diet
AnglesushAngle(ang(225,90,0)) #
AnglesushAngle(ang(316,90,0)) #
AnglesushAngle(-ang(45,90,0)) #
AnglesushAngle(-ang(135,90,0)) #
AnglesushAngle(-ang(225,90,0)) #
AnglesushAngle(-ang(316,90,0)) #

#Starting population
Start=100
#Breeding age
Age=1
#Median age of death
Old=5

M=1
while(M<=Angles:count()){
holoCreate(M)
holoAlpha(M,160)
holoModel(M,"hqicosphere2")
Al=255/Angles:count()
holoColor(M,vec(1,1,1)*(Al*M))
holoParent(M,entity())
M++
}

print("==Function start==") G=1
}
interval(100)
R=10 #Ops management

#Initial population seeding
while(Gen<=Angles:count() & Pop:count()<=Start & !State & R){
X=round(random(0,15))
Y=round(random(0,10))

ValushString(X+"-"+Y)
V=vec(X,Y,0)
V=V:rotate(Angles[Gen,angle])
holoPos(Gen,entity()os()+V)

if(Gen==Angles:count()){
S=Val:concat("/")
PopushString(S+"/"+G)
if(Pop:count()>=Start){
State=1
Kill=0
P=Pop:count()
print("===Beginning Generation 1")
print("=Population: "+P)
}
Val=array()
Gen=1
}
else{Gen++}
P=Pop:count()
R--
}

#Population filtering
M=0
while(M<=min(P,R) & State==1 & Pop:count()>1){
Mem=Pop:count()-P #Population member
Str=Pop[Mem,string]:explode("/")

if(Str:count() & R){
Chance=100
#Modification of Chance variable
#determines the odds of survival
#The higher, the better

#Enter criteria for survival here#
#Good vision
S1=Str[1,string]:explode("-")[1,string]:toNumber()
S2=Str[1,string]:explode("-")[2,string]:toNumber()
Sight=(floor(vec(S1,S2,0):length())/18)*12.5
Chance -= Sight

#Efficient diet
S1=Str[2,string]:explode("-")[1,string]:toNumber()
S2=Str[2,string]:explode("-")[2,string]:toNumber()
Diet=(floor(vec(S1,S2,0):length())/18)*-12.5
Chance -= -Diet

#Old age
Age = G-Str[9,string]:toNumber()
Age = (Age/Old)*50
Chance -= Age

#Use details in Str array for reference#

Rand=random(0,50)
Surv=100-Chance
if(Rand<=Surv){
Pop:remove(Mem)
Kill++
#Failure, the member dies
}
P--
}
R--
M++
}
if(!P & State==1){
State=2
print("=Deaths: "+Kill)
}

#Population reseeding
if(State==2 & !Res:count() & Pop:count()>1){
A=round(random(0,Pop:count()))
B=round(random(0,Pop:count()))

As=Pop[A,string]:explode("/")
Bs=Pop[B,string]:explode("/")

Res=Pop:clone()
Pop=array()
}

while(Gen<=Angles:count() & Res:count() & State==2 & R){
#Parent 1's XY variant
Ax=As[Gen,string]:explode("-")[1,string]:toNumber()
Ay=As[Gen,string]:explode("-")[2,string]:toNumber()
#Parent 2's XY variant
Bx=Bs[Gen,string]:explode("-")[1,string]:toNumber()
By=Bs[Gen,string]:explode("-")[2,string]:toNumber()

#Combination
X=round((Ax+Bx)/2)
Y=round((Ay+By)/2)

#Mutation
Ran=round(random(0,1))
if(!Ran){
X+=(Bx*round(random(-1,1)))/10
}
else{
X+=(Ax*round(random(-1,1)))/10
}
Ran=round(random(0,1))
if(!Ran){
Y+=(By*round(random(-1,1)))/10
}
else{
Y+=(Ay*round(random(-1,1)))/10
}
ValushString(X+"-"+Y)

Vec=vec(X,Y,0):rotate(Angles[Gen,angle])
holoPos(Gen,Vec+entity()os())

if(Gen==Angles:count()){
PopushString(Res[A,string])
PopushString(Res[B,string])

#Birth control
Ran=round(random(0,3))
if(Ran==1){Res:remove(A)}
elseif(Ran==2){Res:remove(B)}
elseif(Ran==3){
Res:remove(A)
Res:remove(B)
}

if(Res:count()==1){
Res:remove(1)
}
#Prevents asexual reproduction
else{
#Generate new breeding pair
A=round(random(0,Res:count()))
B=round(random(0,Res:count()))
}
#New parent variable arrays
As=Res[A,string]:explode("/")
Bs=Res[B,string]:explode("/")

#Breeding age or playful curiosity?
Ag1=As[9,string]:toNumber()
Ag2=Bs[9,string]:toNumber()
Breed=(Ag1>=Age & Ag2>=Age)
Fert=(Ag1<=Old*0.8 & Ag2<=Old*0.8)
if(Breed & Fert){
PopushString(Val:concat("/")+"/"+G)
}
Gen=1
}
else{
Gen++
}
R--
}
#If finished populating, return to filtering
if(Pop:count() & !Res:count() & State==2){
State=1
G++
Kill=0
P=Pop:count()
print("===Beginning Generation "+G)
print("=Population: "+Pop:count())
}

#Last Survivor holopattern
if(Pop:count()==1){
Vecs=array()
while(Vecs:count()<=Angles:count()){
Str=Pop[1,string]:explode("/")

X=Str[1,string]:explode("-")[1,string]:toNumber()
Y=Str[1,string]:explode("-")[1,string]:toNumber()
V=vec(X,Y,0)
V=V:rotate(Angles[Vecs:count()+1,angle])
VecsushVector(V)
holoPos(Vecs:count(),V+entity()os())
}
}[/highlight]I'd call this proof of concept as far as creating genetic development in Wire, albeit it's primacy.

[Vbitz]

Well that is really cool.
Teaching a e2 bot code pattens is easy, it learns from a person whenever it does not know something

[feha]

They always die out :P

Poor lone person it always end at XD.


Neat code anyway.

[Lyinginbedmon]

They always die out :P

Poor lone person it always end at XD.


Neat code anyway.

If they don't die out, they just keep breeding. That means the arrays get huge and the chip takes forever to filter and reseed, that ends in lag.

[Nikita]

Hey guys, I'm using this chance to revive a very old thread to say that today I randomly wanted to try making a chat bot.

I'm beginning with a knowledge base, that contains concepts for some of the most used words of English language.
I thought it would be cool to make a chatbot that can at least make meaningful deductions from what it knows.

Here is how I imagine the knowledge base:
Each word has a node that lists it. There may be more than one meaning to a word and each node is a lemma for similar words (be, is, was).
A node is defined simply by a number. Each node has a list of properties, contexts and other stuff associated with this concept.

For example:
353: Sky
Is A: location(52)
Property: far(32) Above(532) ground(153)
Property: light(3510) blue(1225)

1952: car
Is A: vehicle
Property: has(5) 4 wheels(519)
Passive: be(2) driven(5021)
context: if driven(5021) can(51) go(312)
context: if goes(312) consumes(3582) oil(12884)

Something like that. Naturally it would take years to make a such a list for all English words, so I'll just add the most frequent ones. My chatbot, if I complete it, will take user's input, assign the most probable nodes for words with many meanings based on context, look at the input like a logical problem, then make deductions based on what it knows. While it thinks it will generate it's own persistent knowledge base, and store it separately from the primal one I make. But only if it deems the information important - after a while it could easily generate loads of useless data (like, if a box it to the right of the barrel, nothing happens.) so some information it would forget after a while.

Anyway, I would like to see if anyone has done anything like this, because as I remember, there have been a few chatbots made by wirerers. I've actually heard one speak on HP's server and it was hilarious

[Solece]

Sounds like a neat idea Nikita, I've tried making an AI that uses a database of words with their own definitions, it's easy to get a list of words, the hard part is what to do with those words.

[Lyinginbedmon]

I've been considering the possiblity of a half pattern recognition half quasi-neural network "brain" for a while, haven't put anything substantial together in code yet though, trying to solve the data storage issues.

[BenBob]

Nikita, your concept there sounds like a bit like a 'Is a - Has a' setup, "A dog is a mammal", "A dog has a tail", "A player has a physgun", "A physgun is a weapon", etc. etc.