Yeah, if we make creature movement completely dependent on the NNs, a lot of “stupid creature” NNs would be expected to emerge and be very visually present for the players
It is very difficult to simulate something similar to learning behavior or thinking because of the technological limitation.
A human brain is converted into computer memory, it is 2.5 million gigabytes:
And if we do an estimated calculation, of a worm alone, the weight of its brain is something like ~0.3 gigabit (a very rough calculation), and it is a simple creature with 302 neurons and connections.
Imagine a creature as complex as a fish can reach much more.
I will give you an example of something that is most similar to this scenario: the game “Black and White 2” you can teach a creature anything you want. The weight according to the Guinness World Records can reach a weight of one gigabit (the game was born in 2005, a gigabit for a home game is not much when the average C drive at that time was about 32 GB. I remember it well).
Just imagine this complexity. I don’t think even with the best programmers we are able to simulate how a mouse thinks in Thrive. We will need to go for the classics of a behavior diagram and a classic pronunciation tree to avoid going down this hole.
Edit: if one creater that weight 1 GB its not enough, imagine an Ecosystem!
2 Likes
Yeah, I was wondering if simulating like two dozen brains with the openworm brain complexity and simulating the game at the same time was really feasible…
1 Like
Look we are not into stimulating real neurons and the
302 neuron in worst case scenario will take 302^2 X8 byte adding to 0.75MB then even 100 copies could be handled easily,
However a truly complex NN with different type of Neurons is impossible to scale or stimulate
( This a video about someone trying to make a truly complex brain structure using EA and the number of neurons is realtivily little ( way less then 300)) and you could easily se that it took too much time)
2 Likes
Using a moneygreedy studio to create brains doesn’t seem like a good idea…
1 Like
Haha, EA is Evolution Algorithm
1 Like
Well, in that video there were quite a bunch creatures being simulated at once. In Thrive not that many would you expect to be simulated at one moment.
1 Like
You don’t understand the essence of EA , that process of stimulation is needed to make a good brain structure and the problem is not about the stimulation but about creating a creature that have a good brain through the stimulation our problem is the opposite it’s to create a good stimulation through the brain, but this goes to show the hardness of making a realistic brain structure
1 Like
Yeah, and all that needs to work even on relatively weak computers…
1 Like
No, one said we will use this method but it is a method nonetheless, what we are going to use is way simpler version of this and it will use “gradient decend” to train rather than an EA which is way faster and ressource economic
1 Like
Did you talk about “gradient descent” yet?
1 Like
All the models and calculations I made where using 'gradient decent ’ ( if you have watched the video, you’ll clearly see that there is no notion of layers in his NN model which is the main characteristics of a " gradient decent" approche
1 Like
When do you think this can start getting coded in?
1 Like
Once every thing is clear as far as the concept have been totally done and ( you could use a basic algorithm to control the IAs at first) then once everything is stable in the late multicellular stage, we could lance this as a mode where creatures ( learn and adept to their environment) as a feature in the last stages of development
1 Like
Oh I see. Multicellular as a whole is supposed to be finished by late 2026 or early 2027. Are you sure you’ll learn enough coding to code in the NN stuff by then?
1 Like
I won’t make promises about a thing I didn’t start doing but will certainly be able to say something after 1 year, if I gained the some skills or not yet , but the foundation is already there ( I know some coding and I am going to become a software engineer after 3 years , so if I got enough time during those Three, I’ll be working on documentation of the code base since I think this is the most important thing right now)
1 Like
Yeah. We also have the whole aware stage development era before the evolution stages are “complete”.
1 Like
I don’t know, but we do what we can for now , I mean this hole idea is not necessary and even could be a mode or a debatable feature at best , so there is no rush( I thought of making it a mode since the delay of stimulation and NNs training might impact the game experience so it might be a mode for those willing to pay the price ( time and maybe a good pc depending on the actual implementation) .
1 Like
can’t help but read this and be reminded why the brain is the most complicated organ of the human body that we understand the least, the Brain as a computer doesn’t Operate on Binary, Dopamine, Serotonin, Acetylcholine, Glutamate, all neurotransmitters that mean different things not just 1 and 0. NASA computers would have a hard time understanding a single human brain, it’s also part of why Biological computers have so much potential they could be at a similar level to Quantum Computing.
1 Like
Awakening will have “small evolutionary changes.” Awakening Stage - Thrive Developer Wiki
2 Likes