Hacker News Reader

Comment by kbenson | Hacker News Reader

In his presentation, he observed that the AI reminded players of the first Half Life game, which came out seven years earlier. “It seems that we haven’t made much progress in seven years. There has to be more we can do with game AI.”

That's what came to mind reading this. I remember the AI and Half-Life, and while it sounds like they did better in this game (which I haven't played), it also sounds like an iterative step, and after seven years, that's not a lot. I happen to own this game on Steam from some humble Bundle pack or something, so I think I might have to break it out and have a play...

Edit: Sheesh. 17GB install for a 2006 game. That's steep.

View on HN

There was a article years ago about the Deus Ex team trying to find tune their AI. They found that a dumber AI actually worked better for player enjoyment. Players would frequently think the AI cheated when they got flanked and hasn't seen the enemy move to their flank. The human players just ended up doing low fun strategies like using hall ways or other bottle necks to wait for the AI to attack. Will try to find the article later.

Edit: So apparently it was not from Deus Ex, but from someone anonymous. Still interesting read: http://askagamedev.tumblr.com/post/76972636953/game-developm...

Oh really? This is exactly what Mass Effect 3 did with the cover buster mechanic, throwing in some randomness. Combated the bottlenecks by map design and giving AI grenades, gave cover buster and charging enemies distinctive appearance. Made for good gameplay.

This is because the Steam game actually includes the base game and both expansions (even though they are shown as separate items in Library, Steam downloads the files for them).

Playing against a bot programmed to fight intelligently and efficiently just wouldn't be fun for long. A friend and I tried playing against the Reaper bot in Quake years ago. It mopped the floor with us. Which in terms of my skill probably doesn't mean much but my friend was really good at Quake. The bot didn't let up, didn't pause, memorized weapon locations, move at a constant run, and never missed. It would even jump, 180, and beam you with a rocket. That made for fun novelty one evening but no one wants that for multiple levels.

There's a difference between a bot that's been programmed to act intelligently and a bot with superhuman abilities and reflexes. You can have the first part without the second.

League of Legends is a really good example of this. They've done a pretty good job of gimping their bots considering that simple player-made bots can be competitive at the highest levels of ranked play.

>player-made bots can be competitive at the highest levels of ranked play

Any source for this? I refuse to believe it.

Many of the useful abilities in League of Legends are 'skill shots' - manually aimed, with travel time and wind-up animation, and are thus missable or dodgeable

With superhuman reflexes, a bot can be nearly immune to skillshots in a 1-on-1 scenario. A bot can also always hit skillshots, up to its accuracy of prediction of your movement, and your reflexes in dodging.

In theory, a team of humans might demonstrate better teamwork, communication, and adaptation than a team of 5 AI. However, given how much of the game depends on mechanical 1v1 conflicts, I'd expect a simple bot to beat even the most expert human players.

I bet that a game AI model of a human mind's executive function would actually be embarrassingly simple. Something like a 7 element ring buffer of tasks and their parameters, with interrupts. When the bot tries to remember or do more than 7 things, it just starts forgetting stuff, which it might notice again later. Just like me.

Agreed. This is why single player game AI can be made fun with simple FSMs and logic models.

The reaper bot was a good start, but it wasn't until the Frogbot[1] was released that Quake bots really came of age.

[1] http://www.quakewiki.net/archives/botarea51/main/qbot7429.ht...

Side note: I regularly played against the Frogbot's author (who used the handle Frog) at LANs. He was reknowned for being the most analytical and "bot-like" player of our scene.

F.E.A.R.s bots are by no means intelligent - they're really dumb and predictable, at least once you've played through a couple times - you know that they're going to seek another angle when they get blocked, and there's only a couple options for them to do that -

but it's fun when you're playing enemies who are seeking another angle, even a simple one. Compared to the other games on the market. F.E.A.R. remains the best FPS I've played.

>my friend was really good at Quake

no, he wasnt:( he was just better than you. Reaper on default skill 1 was an ok challenge for average players. Skill 3 was a competitive against good players, but still nothing special against deathmatch veteran spending days in gaming cafes or internet servers.

For reference best Quake bot, Omicron(dev went to work for ID software and wrote Quake3 AI etc) on highest skill level vs actually good dm player https://www.youtube.com/watch?v=hEeDXPkEnME

also https://youarenotsosmart.com/2014/11/10/yanss-podcast-036-wh...

Don't play perseus mandate, stop after you've beaten extraction point.

Just don't. It's just boring and uninspired and... ugh.

HL2 AI was definitely worse.

HL2 AI was still pretty amazing

I miss how good HL2 was... Alas, the third won't happen

Imagine the effect HL3 would have on HTC Vive sales

I want to hit headcrabs with a crowbar so bad

As someone developing for VR I'm not sure I'd want to go on a weekend long "I'm gonna play through all of HL3" binge.

Even simple VR games can make you feel like you're moving when you're standing still or back you into a corner with enemies. HL3 might screw you up a little. HL2 did a very good job of matching the scene to the plot and that made for a more immersive experience. Add in VR and modern graphics and that's a level of realize that would probably be too much for a few people.

https://youtu.be/e0WqAmuSXEQ

I played through HL2 again recently and it was fun but I found it to be over designed. HL2 (and Portal!) try way too hard to make the player feel clever. You'll find some nook or cranny or hidden path that always turns out to be only way to go.

I used to rag on the guns in the game too; they are not fun to use, but looking back I think that was a deliberate strategy to encourage the player to use the gravity gun more.

HL1 certainly had a much more organic feel to it. Like the abstract bases and hell of DOOM vs the uncanny valley cities of Duke Nukem 3D, the missile base of the original HL, which has no basis in reality in terms of its design, feels more authentic and believable than City 17.

Most of the battles in HL2 are either forgettable or boring (SHOOT A ROCKET AT THE THING!), but I must say I really enjoyed the one near the end of the game where you fight against soldiers on multiple levels as you climb your way up some kind of warehouse: https://youtu.be/91uWeqArrok?t=3h38m19s

I'd argue that combine soldiers were closer to objects as part of physics puzzles than intelligent opponents.

Reply on news.ycombinator.com