Artificial Intelligence/Life thread [!OT]

**mrwiseman** · September 24th, 2003, 10:16 AM

I thought some might want to talk about this. Post your thoughts, comments, ideas, ethics, whatever.

----

As you may know from the "what games are you currently playing" thread, I've been recently working on Darwinian Artificial Life (just for fun) consisting of a virtual machine which runs processes which have mutated their code from the mother (the first process). The VM works and I've been running it with 1024 processes, each one with 4 KB code. I'm still a long way from getting any interesting results, but it might work (and if it doesn't, I don't care, look at the fun I had [img]smile.gif[/img] ). I need to upgrade the VM to add more addressing modes, to tune up the evolution mechanism, and most of all, to design a way of rewarding processes who do something "good" with reproduction, while limiting others' capabilities (they overpopulate they system with their trash code and it quickly fills up the process limit. And allowing more processes would only make matters worse, because of the geometric progression they may have if they loop a FORK).

Mine is not the only project of this kind; I've seen others with Turing machines. The problem is it's so hard to do anything with a Turing machine, and it takes so many cycles, I don't think it'd be the best approach, so I went for a more complex Von Neumann machine which I suppose will give better results in the mid and long term.

**mrwiseman** · September 24th, 2003, 07:45 PM

sort of a doubt. how does work that VM and how does it "control" the processes?

<font size="2" face="Verdana,Tahoma">Every process state is defined in a structure which has registers, memory and flags. The VM may either execute them step-by-step or to the end. In another module, the operating system handles an arbitrarily long process queue and calls the VM in step-by-step mode. It's in charge of process reproduction with or without mutation (the VM reports to the OS that the process wants to reproduce). The OS also prints stats and will dump processes, save and load states, etc. Also, the VM and OS I/O is handled by a separate module, so I could improve or port it easily.

can any of them appear like the original "VM" and become a "god" of that "Universe"

<font size="2" face="Verdana,Tahoma">Kadomony 0.01 (the VM) is still not Turing-complete, it can't emulate itself. But this is the very first version of it, I plan on upgrading the machine with more memory and addressing modes.

**melfice** · September 25th, 2003, 03:08 AM

Interesting, are you trying them to "learn" for themselves huh?

for example, you could add a "satisfaction" variable, which increases as they do "good" things [img]smile.gif[/img]

sort of a doubt. how does work that VM and how does it "control" the processes?

(can any of them appear like the original "VM" and become a "god" of that "Universe" [img]tongue.gif[/img] )

<font color="#a62a2a" size="1">[ September 24, 2003 12:22 PM: Message edited by: Melfice ]</font>

**mrwiseman** · September 25th, 2003, 05:23 PM

A VM function is passed a process structure. The step-by-step function does as follows:

1. Check for wait states - if the process is waiting, it returns, doing nothing

2. Load next instruction from code memory (as pointed to by the Instruction Pointer register)

3. Fetch instruction (preprocessing)

4. Load operands. Currently, all I need to do here is to get a new random number for RD (Random register) if it's detected as an operand for the instruction. (This VM has "free" random numbers via the RD register, I thought it'd be a nice tool for my pets. =) )

5. Detect opcode and execute instruction. It does what the instruction says, using registers and memory as necessary. For example, like this:

<BLOCKQUOTE><font size=2 face="Verdana,Tahoma">code:</font><table border="0" width="90%" bgcolor="#333333" cellspacing="1" cellpadding="0"><tr><td width="100%"><table border="0" width="100%" cellspacing="0" cellpadding="2" bgcolor="#FF9900"><tr><td width="100%" bgcolor="#E0E0E0"><pre>case 0x7C00: //JAXBEf reg, off - Jump if AX below or equal to reg (fwd)
p->IP += p->reg[arg1] >= p->AX ? arg2 : 1;</pre></td></tr></table></td></tr></table></BLOCKQUOTE>

6. Increment IP (unless a jump instruction has been executed). Note that IP now points to the next instruction

7. Check for valid addresses: if a pointer register is out of bounds, bai-bai to the process (Kadomony 0.01 does not support exception handling)

Once the 7 stages are done, the step-by-step function returns a value with some flags (like if the process died or if it requested reproduction). A process dies when the maximum number of cycles allowed to that process is reached (this avoids infinite loops), when any pointer register gets out of bounds, or when it executes the DIE or the CESAR instructions. A process requests reproduction to the OS (which may be either cloning or mutation, depends of a random factor) via the FORK or the CESAR instructions. (CESAR = FORK + DIE.)

The Operating System (System HAWWA) does this:

1. Create the mother processes (named HAWWA). A mother process just stays there and eventually reproduces and dies, hoping for their children be mutated into something useful.
2. Run N steps from the next process in queue (N is an arbitrary constant)
3. If the process requested reproduction and the number of processes is below the limit, decide whether to clone (exact copy) or mutate (vary randomly) the process, and do it
4. If the process requested death, it's removed from the que and freed
5. Report some stuff and deal with the UI (if any)
6. If number of processes > 0, back to 2

The child processes inherit either an exact copy of their parent (clone) or a variation (mutation). Currently, I haven't introduced any special goals other than doing anything useful or amusing. I have to think of a Good/Bad system and a reward system, and maybe design a hostile medium they'll have to adapt to or they are killed in x cycles.

**melfice** · September 26th, 2003, 03:45 AM

Mm... so the "beings" (the processes) are controlled by the VM via Step-by-step execution? how it is done that? (here you know how to create separated processes not how do you execute them step by step...)

And another thing. What do the "son" processes do? it is, what is their basic behavior, besides executing their code. it is, in what consist their task?

**melfice** · September 29th, 2003, 09:37 AM

I think that an hostile environment would be useful, and i think that relatively easy to implement, though, it would create "Fighting creatures" And most of the AI i know are only "perfect fighters"

For example, a "helping process" would be good. or even a mix of them. process which are innate "fighters" and other processes which are good for helping other processes or even themselves. (I.e., they give more life to other porcesses and stuff like that)

though the reward system is definitely the best you could give them. though...

a question...

how creatures decided to start having these kind of interests, it is, how did we learned to see what was good for us and what not? and how did it evolved to this from simple aminoacids?... just a question that i thought the other day...

**mrwiseman** · October 5th, 2003, 03:12 AM

That's the line that separates false intelligence (AI) from true intelligence (ours, and true AI, not done yet).

A big difference of the current AIs and true AI/our intelligence, is consciousness. We are self-conscious, while our current AIs aren't conscious of themselves.

However, I think I found the biggest difference in Nietzsche's philosophy: The Will to Power. The will to power is the will to create, to expand, to cover everything, and to control the Universe. Living creatures have an inherent will to power within themselves, which is what fuels reproduction, creation, improvement, creativity, originality, and most intelligent things. Quoting Nietzsche, <u>life is Will to Power in itself</u>.

Our current AIs lack creativity, the will to create, to innovate, to invent, to expand for the sake of it, and to have fun for the sake of fun. Once we can create an algorithm capable of this, we will have our first true AI. And there's a very important issue here I'll discuss later.

Moreover, I think will to power and consciousness are closely related. If we can create one, it'll be by creating the other as well.

It seems hard to think of a system with such capabilities. However, keep in mind that if we did a complete Physics simulator (up to atoms and ions simulation), we could introduce our own brain structure on it, and we would have a machine running some way of will to power and self-consciousness. After all, our brain too is like a machine, only it's bioelectric, and uses different materials and paradigms.

As for the issue I mentioned, here is it:

Should true AIs have the right to live?

What do you think? I think true AIs, if ever developed, should have the right to live, and to learn, and even that their "dead" bodies should be treated respectfully as any other living creature. I will develop the reasons behind this later.

**Denial701** · April 17th, 2023, 04:22 AM

You are right, AI is now increasingly being used in software and in general to improve the management of business tasks. I have read about AI in wealth management which can now help you manage your company better through automation and better wealth management.

**ak0n** · August 4th, 2023, 11:16 AM

Hi all, I recently came across the fact that AI created content is now unlisted and often penalized on various Google platforms, do you know what can be done about this?

**piokil** · August 4th, 2023, 11:17 AM

Hi, I know a thing or two about this! If you need some serious stealthy AI text for articles or copywriting, try undetectable ai paraphraser. I'm telling you that their reframers and AI stealth features are next level, making my AI emails read very naturally, not as if some robot wrote them. Highly recommended. I'm sure it should help you!