I’ve been searching the asset store for that elusive Prediction AI asset without success. Are there any related ready made packages of sorts out there that anyone may be aware of suitable for use with Unity?
Prediction of what? You are going to have to be a little more specific here.
AIs tend to be very domain specific. Most of the time they are custom built per game.
So far the general AI is still just a dream. Although according to the latest @Arowx thread, it is just around the corner.
1 Like
LOL I though you would get the fact that I am fascinated by what deep learning AI can do for game developers. Don’t confuse Deep Learning AI (DLAI) automation with human level (or >human) general AI. DLAI’s are complex and amazing pattern recognizers. Your brain apparently manages to run about 300 of these at once.
When they start hooking up a few hundred or thousand DLAI’s together then things will get interesting.
FUN FACT: DLAI’s for translation build up language pattern networks that can be quickly trained to new languages.
@GB51 Do you mean an AI system that does prediction, like Bayesian probability theory?
As most AI systems tend to do some kind of search over possible moves then try and pick the best one.
1 Like
@BoredMormon I’m probably expecting to find something a little too generic I guess, something that will at the same time address the needs of everyone in general but of no one in particular! lol ! I’m a little lazy when it comes to implementing my own AI solutions, especially in regards to predicting or perhaps anticipating the next player move and then having an NPC respond to such by blocking doorways, exits, placing obstacles, attacking, etc. Maybe Deep AI will someday provide an easier way of doing this, perhaps landing itself in some form or other on the asset store!
__@Arowx __Can’t wait to see those 100 core computers arrive on store shelves soon (I wouldn’t hold my breath doing that though)! lol ! Bayesian probability seems to describe my situation quite well; I’m a little confused, however, as to how I could effectively, let alone properly, implement the required in-game behavior without having to simultaneously learn and apply the necessary theory! I guess, for now and until an off-the-shelf solution is available, I’ll have to resort to the good-old-fashioned way of doing things by dusting off a few cobwebs from some long neglected brain cells and then applying a little bit of grey matter muscle power to that effect! lol !
If you want something done, but don’t want to invest a lot of time into it, you can always hire a freelancer.
I happen to be one and I’m kinda into AI, but you can always find others on the job seeking section of this forum.
I’m currently on a very tight budget, but I’ll seriously take collaboration on a non-commercial project (in the proper forum of course) some serious consideration once I’ve got a workable, half decent prototype set up. Behavior Designer does appear to provide some sort of prediction-like solution by way of what’s called Utility Theory; I’ll definitely look into that. AI Designer Pro also looks somewhat promising. But, thanks for the offer, nevertheless; once I can get the bulk of this AI stuff done, I’ll certainly get in touch!
…even though it is unknown around which corner or how far away that corner is.
I’m think there’s no such thing. An AI asset at best provides you some building blocks (like decision trees), and then you use them to create your own solution.
Speaking of AIs, RAIN framework is available for free, so you could give it a whirl. But it won’t provide “predictive AI” for you → you’ll need to build this out of building blocks provided.
I’ve discovered that nGrams are algorithms used to produce predictive behavior based on the probability that some previously learned behavior will likely be repeated. This is as close to what I’m looking for as it gets, so far. NGrams receive some appreciable treatment in a book for Unity on artificial intelligence programming from Packt Publishing but I won’t give its name here since forum rules apply restrictions on advertising. The RAIN framework is a very comprehensive set of tools for the implementation of behavior trees, but you are right, it simply falls short of anything close to implementing predictive behavior suitable to my needs. I am grateful, nevertheless!
LOL It’s not the core count it’s the Network Count, and they would need way more cores than 100, Neural networks tend to run best on GPU’s that have thousands of ‘cores’ or on dedicated Neural chips.
It’s estimated that the average human brain has about 100 billion synapses that work in parallel and are estimated to run around 300 pattern recognition systems that are you and your knowledge, behaviours and understanding of the world.
Whoa!
That…just…literally blows my mind! Then again, maybe it’s a good thing artificial machines won’t be replacing us anytime soon! lol! Thanks for the insight!
Number of neurons in human is around 10^11 (one hundred billion).
Number of synapses is between 10^15 and 5*10^15 (100 to 500 trillion).
Also… there were predictions that by 2050 we’ll have a machine capable of emulating neural activity of entire humanity at once.|
Still, those predictions may be a bit off.
By 2050! Hope I’m dead before then, lol! 