I just had an idea, loosely related to your mental models a AI Terminator requires. Putting that here, just because it's the latest blog topic.

I was thinking about the biggest problem for me as one man game dev: Speech. I don't like that i can not do this myself. I can do code, art, even music, but i can not do voice acting.
Thus i often think about how to avoid a need for speech. Which could be just text, or a setting where characters can not talk, etc.

And actually i was thinking about visual story telling using images / movies.
Imagine, you as the player, have killed a guard.
Then you sneak along, and after a while you see two other guards. And one guard says to the other: ‘Watch out, there's an intruder here. He killed one of our colleagues!’
To do this without speech but only visually, i could show a speech bubble over the talking guard, like in comics. And inside that bubble, i could show an animation of the players avatar killing a guard.

Ofc. i'm not totally serious with that, because technically this would be super expensive. We need to run a simulation of the player killing the guard, so we eventually need to stream now distant locations, calculate lighting for that, render it, etc. All the stuff. It's a game inside the game. Paying some voice actors would be little effort in comparison.

But - if we need that functionality for mental models of AI anyway, maybe that's indeed a feature future games should have?
Maybe such expensive mental simulations could have many applications, not just AI?

Could be an interesting idea, or maybe it's just bullshit. That's all i have to say today, hihi :D

It's a game inside the game.

Well since what you`re saying is speech related you could also imagine not just a game that is totally dependent/an exact reflection of the initial game but rather two distinct games evolving side by side, they are somehow related to each other and yet different, in the sense that each game has it`s own logic and it`s own inertia. There could be a striking resemblance between some of the characters of the two games but technically speaking the two games are separate.

imagine not just a game that is totally dependent/an exact reflection of the initial game but rather two distinct games evolving side by side, they are somehow related to each other and yet different,

Yes. The difference is important. For example, the guard telling the murder event could depict a brutal and reckless player, while in reality the player acted just out of self defense. And if the second guard tells the story he heard to another third guard, the player becomes even more of a monster, and so on. That's an interesting story telling mechanic, as we could follow how imagination of NPCs tends to exaggerate like in real life.

Another application would be if the player could use the same imaginary simulation sandbox to test some actions before actually executing them. The imaginary world would be incomplete, and contain only the stuff the player has already seen. He can not enter a room he does not know about yet, and a guard he has not seen yet would be missing. This motivates to explore. The visualization could be abstract or hazy, like in a dream. CP 2077 did something similar with it's Brain Dance recordings to get some detective game mechanics. But that was more like editing a movie and no real interactive gameplay.

It may be an interesting idea, but there also is a risk of making games even worse. The more advanced and realistic our technology becomes, the more we loose the ability to spur the players imagination itself. Imo that's one of the key problems of modern games. When it started, we got an Atari game with great box art but blocky abstract visuals. The goal was clear: We want the game to look like the box art. Now we have that, but without noticing, we lost the imaginary process on the players which turned pixel blocks into awesome worlds just in their mind. Modern games completely fail at spurring the players imagination, which is a major reason why they are so boring. So if we give the player a imaginary simulation tool, that's like saying: ‘You are too dumb to imagine, so i'll give you a helping hand to do so.’

Contrary, if we hide our abstract world representations we use for AI (a simple grid, navmesh, boxes and capsules, etc.) from the player, then we give him at least the option to imagine how those systems work, so he can optimize his actions based on imaginary assumptions and observations. That's not bad currently.

However, the idea of imaginary simulation, mental models, etc. remains very interesting. I think static navmesh or simple grids are no longer enough. We want worlds with rich options, so it's about dynamics, interactive objects, etc. And we can not model intelligent NPC actions within such environment using such restricted data structures and algorithms.
On the other hand idk how the ability to do imaginary simulations helps us with intelligent behavior at all. Probably we have to stick at ‘player is the only intelligent character in a game’ for a long time.
And even past that, it's not yet clear how intelligent NPCs would give us fun games. I still have most fun with Super Mario, where enemies are dumb as fuck and thus totally predictable. We need to preserve or replace some things, so games remain games.

Now we have that, but without noticing, we lost the imaginary process on the players which turned pixel blocks into awesome worlds just in their mind.

I think the graphics in a game need to be just explicit enough for us to tell what the object we are looking at are.

Imo if there is too much detail/too much realism your brain goes into ‘watching TV’ mode, a 'relaxed' state of the brain where you are absorbed by the detail. It`s a state where you`re all into the image and have limited ability to respond (usually all you can do is just dumb aiming and press the fire button). A classic RTS/FPS even one with 3d graphics (but not too detailed) allows you to be ‘half here half there’, which means that you don`t loose your ability to think.

I think we like video games not just because visually speaking they are an exact replica of reality but because in a video game we also see the relationships that exist between real life objects. Like a video game stimulates your brain into thinking you`re assisting a real life scenario of a particular type while being excused from having to suffer the nasty real life implications of that scenario.

A classic RTS/FPS even one with 3d graphics (but not too detailed) allows you to be ‘half here half there’, which means that you don`t loose your ability to think.

The advantage of low detail artstyle often is a clear design to easily distinct floors, walls and enemies, for example.
But you can still have this with high detail, if you design with that in mind. You could still have pretty flat / uniform colored walls, where the detail is only used to add some small scale structure and subtle color modulation. And you can still have enemies standing out by giving them red / blue clothes, plus a lot of detail like folds, pockets, simulated hair, or whatever.
The resulting artstyle would not be realistic but detailed. I don't know many examples, maybe Halo. But you're right - most games just go for realistic assets, which often is confusing and does not even look good.

The other gfx topic is lighting. But here we can be sure that better lighting only has advantages and no downsides, because it improves depth perception.

Personally i'm convinced we should still aim for improvements on graphics. It's a never ending source of showing progress over older games, which is good.
But if game devs want this so badly they take the bait from HW industry to establish 100tf GPUs at 800 watts for 1000 bucks as a requirement, then thy'll just shoot their own feet.
Looking at current raytracing or UE5 hypes, the results simply don't justify the high HW requirements. It feels we have left a sane sweetspot some years ago, and currently devs do not even notice the conflict of interest rising between software and hardware industry.

This, together with the older problem of stagnating innovations on gameplay and genres, will give us a next video game crash.
I say this since quite some time, and now i read the games industry has a loss of 15% on in just one year. The Titanic starts to sink.

Maybe the fun is over, and video games will die out, just like rock'n'roll did.
Or, maybe we'll see a new generation of devs, focusing more on games than on billion dollar budgets, hollywood cutscenes, trumpeting silly hypes, or pretending competence on social issues.

Either of those options is better than the current state ; )

The advantage of low detail artstyle often is a clear design to easily distinct floors, walls and enemies, for example.

Starcraft 2 is a good example for making things hard to distinguish.

But you can still have this with high detail, if you design with that in mind

to some degree yes.

to my mind high resolution means more information in the texture. If the texture is just one color there is no way to tell if it`s hi res or low res. Some thing happens if the texture is a transition between two colors.

Starcraft 2 is a good example for making things hard to distinguish.

Agree. The classical mistake of having too much variance in high frequencies, so the whole image becomes just noise to our perception.

to my mind high resolution means more information in the texture. If the texture is just one color there is no way to tell if it`s hi res or low res. Some thing happens if the texture is a transition between two colors.

The related science and terminology is signal processing (beside art, which is too subjective).

Example: we have a texture of a simple gradient. Although there are subtle changes between neighboring texels, a low res texture is still good enough to capture the gradient. Just as with a single color.
This is because the gradient is low frequency content.

Now lets add dirt over our gradient. To show small particles of dirt, we need a high res texture, because the dirt is high frequency content.

But the dirt causes only low variance to the gradient - it darkens the colors at some spots a bit, that's all. Due to that low variance, we can still use a low res texture and won't miss so much.

Now, instead dirt, lets place hundreds of Super Mario, Peach and Toad sprites over our gradient. This causes high variance (contrast) at high frequencies, and we get ugly, confusing noise.
We need high res texture, so we can still see that's Super Marios, but the result looks just as shitty as with using a low res texture. To our perception, the result is just noise no matter what.

Usually, what we want is high variance in low frequencies (walls vs. floors with different materials and lighting), and low variance at high frequencies (dirt, or interesting details of a rock surface).
If we do this right, using higher resolution geometry and textures gives us only upsides, no downsides.
Contrary, using low resolution, we see a lot of issues. E.g. approximating round objects with hard edges, or seeing blocky texels instead fine grained details of dirt.

The company recently released a video showing one of its robots being taught to walk by humans. The source resumes planet shares reviews about online writer. The robot has been trained to follow a human around a course in an indoor area, keeping up with him as he walks up stairs and around obstacles.