People are talking about agents cloning software, including games, to the point where Anthropic is attempting to build in guardrails so that Claude Code won't make clones of already existing software.
By the time this post is released, things will have changed, so just keep an eye on the news.
But I'm here to talk about something that will become a greater and greater likelihood: agents playing games for you.
Maybe we could have laughed it off before where you are easily able to identify bots in-game, but I think that it's not going to be a laughing matter much longer. It's only a small leap from an agent such as Muse going out and ordering takeout for you, obtaining tickets to a Taylor Swift concert, and texting a potential date to healing in a raid.
 |
| Imagine losing a Call of Duty match to this thing and having it talk about having marital relations with your mom. From Meta via The Wall Street Journal. |
And if you thought Blizzard had issues before with keeping people engaged in the leveling process, look out...
***
Okay, that pretty much begs the question: why are you playing a game at all if you're letting an agent do it for you? It's pretty much the same thing as paying a person to level your toon so that you can take over when you're at max level. Or if you feel like your physical skills aren't as good as they used to be, let the agent go into the raid and get the gear for you.
No fuss, no muss.
Or, in a PvP game such as EVE Online, some wealthy individual attempting to create a ton of agent-driven clones to augment a fleet action. (Aside for EVE players: given that EVE is famously an "all's-fair-in-love-and-war" kind of game, is that a violation of the TOS? Or would that be applauded as a "solution to a problem" in-game? And if agents get good enough at playing, would Fenris even recognize that agents are running accounts?)
The last time I thought about this, there was the guy who created a WoW private server with 1800 bots, but that was prior to the release of Muse. The existence of a Muse demonstrates that you can have an agent do so much more than was possible only months ago, and it's only a matter of time before agents are directly adapted to gaming.
I wonder if we're going to be seeing the eventual endgame of some of these multiplayer games if agents get good enough at behaving like real human players that basically we'll see agents playing against agents, with no real humans involved. About a year ago I would have laughed at the idea, since the entire point of playing a game is for you, a human, to enjoy yourself. Now, I'm not so sure. The ease at which a Muse can simply keep you from having to deal with other humans is quite astonishing*, and the temptation is there to say "I don't really want to run Sunken Temple myself, but I need the gear in there. Maybe I'll let Blizzbot (tm) run it for me." I know people who simply despise certain instances or questlines, so letting an agent do it for you will become a viable option really soon. Once that happens, the dam has burst and it becomes easier and easier to let the agent basically run your account for you.
Before you say I've been taken in with all the hype, I haven't. I still think the AI bubble is real, and I still am betting on it bursting sooner rather than later. I also think that a lot of the tech bro community underestimates just how much ire is focused in their direction with data centers and prices being out of control for consumers. When you're wrapped in your own little hype bubble, you don't see what people out in the rest of the country are saying and thinking, and I believe there's a reckoning that will be coming soon. That, however, doesn't preclude me from realizing that these agents are getting "smarter" at an incredible rate, and things that we used to believe were years or decades off might be arrive much sooner than we realize.
*This is me making a value judgement based on the articles I've read about Muse. There's no way in hell I'd let Meta have access to all of my finances and emails and whatnot. You don't even need to have any skeletons in your closet to realize we're talking about huge blackmail potential. I wonder if Zuckerberg is so sold on Muse that he'd let it control all of his finances and personal interactions (such as his private email) without any qualms whatsoever.
The collapse of the bubble won't uninvent the technology and it's just too useful to fade away. What might go is the current free or pretty cheap access the general public has to it but the next logical step is for these technologies to run locally so the whole data center issue and the associated environmental damage goes away. Some of the functions will do that already but it's probably only a matter of time before they all will.
ReplyDeleteAs for having an agent that can play games for me- I can see a lot of upsides in that. The kinds of games I play now are about 75% watching cut scenes or running between NPCs. I'm already watching that like it's TV, just hitting Space each time the voice actor stops talking, so having an agent playing for me would probably improve the experience. The bits I don't enjoy are the fights, particularly the boss fights. If I had a skip button for those I'd use it so having an agent do the fighting for me would be excellent. I'd do all the running around town parts and taking screenshots on my own, though. That's the fun bit as far as gameplay goes. The rest is a slightly awkward way to watch a really good movie.
That actually brings up a question from me: is story extremely important when playing a game? Could you see yourself playing games without much of a theme or are abstract, such as Mahjong, Go, or Chess?
DeleteGiven that CCP/Fenris publishes a ban report monthly now and that bots (or "automated game play" as the company puts it) are one of the bigger complaints by players, I suspect that EVE Online will not embrace AI agents. That does not mean that AI won't be hard to detect.
ReplyDeleteThat said, the reason that bots are a problem in EVE is that a lot of the PvE game play is repetitive and easy to automate, such that AI is probably an overly expensive option when simple algorithms with some timing variables to give the illusion of normal game play are probably still the most efficient path.
As usual, I expect the coming crash to cause the price of using LLMs to rise dramatically. I am sure there will be somebody who will pay to have agents mine in EVE for them, but it will become a pricey option compared to more traditional options.
Even if there isn't a crash, the price of using LLMs will have to rise quite a bit for these new companies to break even. Given that using a lot of these more advanced models already or will soon cost more than the employees they replaced, jacking up the prices will grind the "we'll replace everybody with AI!" movement to a halt.
DeleteIn this case, I think the Chinese models, based on open source, will eventually have the upper hand. Sure, they're not quite as good as the latest OpenAI or Anthropic models, but they won't need to be to get the job done. And being open source, anybody can tinker with them.
In all of the "China is the enemy!" rhetoric about AI, one thing has been lost: there is no way in hell that Xi Jinping will let anything be created that is more powerful than himself. So, you can take the concept of actual superintelligence and throw it out the window as far as China is concerned. If the Chinese do create a superintelligence, it will be purely by accident, unlike what it certainly seems that Elon Musk and Sam Altman are gunning for with their models. As for Trump? He doesn't care. All he sees is dollar signs with all the buildout of data centers and other assorted AI infrastructure. He likely thinks himself as being able to outwit any AI superintelligence that could be created, because the concept of failing is something that is alien to his thinking.