Is the AI any Better in 'The Last of Us Part One'?
What difference, if any, does the updated engine provide?
So let's talk The Last of Us: a video game that gets more re-releases and reimaginings than a Zack Snyder superhero movie. The Last of Us Part I was released on PlayStation 5 in 2022 and I wanted to ask the big question that I'm sure is on everyone’s mind: how much of the game - or more specifically the games' AI - has actually changed in this new version?
Join me as I talk through my observations having played it and discuss what has changed in this new version. The short answer is a lot, but not as substantially as you might think.
Thanks for reading AI and Games! Subscribe for free to receive new posts and support my work.
What is The Last of Us Part One?
Initially released in 2013 for the PlayStation 3, in The Last of Us players take charge of Joel as he guides Ellie through post-apocalyptic America. While it received much critical acclaim, it was released right on the cusp of a change of console hardware. Barely a year passed before the release of The Last of Us Remastered for PlayStation 4: a new version that carried a number of small improvements to gameplay options but was built largely to sport higher frame rates, improved rendering fidelity and the use of the DualShock 4 controller, so you'd be inclined to buy the new console. And that now brings us to The Last of Us Part I on the PS5 in 2022 and PC in 2023: the third release of the same game in 10 years.
The big difference this time around is the game was rebuilt using the updated version of Naughty Dog's game engine, which was used for The Last of Us Part II, released in 2020. By utilising the tools for improved accessibility, controls, gameplay, rendering, audio and of course AI, Naughty Dog seeks to provide the definitive version of the original game.
In an interview in 2022, the creative director Shaun Escayg stated that a key motivation was to allow players "to have the ability to play Part I and Part II continuously without this large gap in technology or visual fidelity". Which y'know, conveniently encourages people to buy PlayStation 5's - now for even more money depending on where you live in the world, provided you can even find one. Also, it's especially convenient with that HBO TV show that just came out too am I right?
Look, corporate motivations aside, it's a valid question to ask whether it's worth shelling out for another version of the same thing. Especially for a game that seems to be getting more and more expensive every time they re-release it just so you can see all the creases in Joel's miserable face.
Though I'm inclined to point out that this is not a simple cut-and-paste job. This requires a ton of work by artists, animators, designers and programmers to reproduce that which you already know well in this new version. It's actually pretty hard to successfully remake something you've made before in a way that feels like it did previously. Naturally, your mileage may vary on whether this merits a purchase, but for the purposes of this piece, I was happy to take the plunge and sacrifice myself to the capitalist gods and then relay my experiences (oh god please support the Patreon so I can justify this stupid expense).
The AI of The Last of Us
So here we go: for this piece I played all of The Last of Us Part I, while also revisiting key sequences in The Last of Us Remastered. And then from that, I'd raise any notable difference. This means I have played an exorbitant amount of The Last of Us in the past few months and good grief I really need a break from this now. I'm hearing clickers in my sleep...
Of course, I have discussed at great length how the AI actually works in The Last of Us series. In episode 52 of AI and Games, I explored in detail the AI of the first game, exploring the inner workings of the human hunters, the infected monsters, and even Ellie herself. But of course, as mentioned the pivot over to the updated engine, means that the AI is now much closer to that seen in The Last of Us Part II.
The AI of the sequel was the subject of the 70th episode of AI and Games. In that episode, we discussed the changes to core stealth and combat, the revamped melee system, the new animation framework and ambient AI systems. I am going to raise some technical elements raised in those episodes throughout this one, so if you want to follow along, I'd recommend you check out both of those previous works if you haven't already.
But it's important to acknowledge that several tools from the sequel don't make the jump to the remake, such as the ability for the player to go prone, and the tall grass that compliments it. The melee combat system that I discussed at length isn't added either. Plus the new enemy archetypes from The Last of Us Part II such as the dogs and the shambler. All of these would fundamentally change how each encounter is built and would require not just the game designers to rebalance each encounter, but also potentially for level designers to have to change or re-orient parts of a level to accommodate. All of these would fundamentally mean changing how the game looks and feels, which would ultimately deviate from the mission set out by the developers to reproduce the original game as authentically as possible.
Now there are some features that immediately improve the visual experience such as the use of motion matching: an AI technology that calculates optimal blend points for animations is used to good effect in both The Last of Us sequel and remake. So characters move around more smoothly and animations come together more effectively. But I'm largely interested in any changes to behaviour, that actually impact the gameplay experience.
I think the most noticeable change is the 'Active NPC' caps being raised. What does that mean? Well, as discussed in my episode on The Last of Us Part II, there is a significant number of AI characters that, while active in a given encounter, are reduced to a lower level of awareness and behavioural response as a means to manage game performance. They're effectively declared inactive.
The reason this used to happen was the CPU and memory resources required for enemies in The Last of Us to hunt the player down. Both clickers and hunters have a myriad of complex systems for detecting player activity, searching for the player and then attacking them (again, go watch the video). This is all quite expensive to process and for the original version of the game, the PlayStation 3 would struggle to provide that kind of processing power. Hence this secret system runs under the hood. Personally, I often suspected some feature like this existed as a level-of-detailing or LOD feature, but it wasn't until an interview in 2022 that my suspicion was confirmed.
So anyways, back to the point: The Last of Us Part I uses the increased active NPC caps that are used in the sequel. And that leads to some notable changes in the intensity of combat. Notably, if you consider some key sequences such as the checkpoint or the courtyard in Pittsburgh, the fights with the Fireflys in the hospital, and even some of the main clicker sequences in the tunnels or Bill's Town, the enemies as a whole are more responsive to local events and more likely to interpret what's going on around them.
Plus, as also discussed in my video on the main channel, the use of the increased awareness state is now employed. This adds two new additions: the enemies don't back down from a heightened state once alerted. This combined with the active NPC count being raised means taking a pure stealth approach is much harder to pull off. And then there is the processing of takedowns of their allies. If you hit someone with the bow and arrow but it's within line of sight of an enemy, or they see a dead body they will now search the area the shot came from or the area around the body more effectively. In the case of the bow and arrow, this is actually a nerf of the AI, given before they pretty much knew exactly where you were if they saw the shot. So it's a nerf that actually retains the realism of the experience.
This is important given it means you can't really de-escalate a conflict once it has started. It's actually a change that influences overall difficulty by making a 100% stealth run much more demanding. You have to ensure not only you're taking down enemies, but that the others don't find the bodies and react to them. Naturally, this is only an issue with the human hunters, I mean the clickers don't strike me as being capable of being upset when their buddy gets shanked.
A Change in Intensity
Interestingly, these two small details, combined with the overall improvements in cover post selection are more than enough to have a meaningful impact. I've noticed some encounters are more challenging or dramatic than previously. Meanwhile, there are others you can get away with playing stealthily more effectively. Critically, stealth against humans is I would argue harder, given they're more likely to catch your mistakes and act on them promptly.
By enabling a more intelligent search behaviour during stealth, it caters to two different sorts of outcomes. Either you're found more readily because the AI is searching more naturally through the space, or conversely, it prevents you from getting caught as easily because they're no longer following a fixed patrol route or search area. It felt easier to evade in some circumstances, but it was also easier to be spotted.
Sure, that sounds like a contradiction, but it meant that the game was enabling a little more variability, as the enemies didn't always feel like they were stuck on fixed patrol routes, and in turn, it felt more like I was in control of the situation. There's nothing more frustrating in stealth games than feeling like you weren't at fault when being spotted by the enemy. Knowing you were clearly in the wrong for taking a particular gamble is fine when it's interpretable, and personally, I didn't always feel like that was the case when playing The Last of Us Remastered. In the remake, it was easier to understand this cause and effect, and better learn from it for the future.
The other big thing was that combat is now a lot more intense. As mentioned in my Last of Us Two episode, another improvement was the ability for the AI to make more intelligent judgements of cover positioning and flanking routes, given the information being assessed was richer than before. When you combine this with the increased active NPC count and the improved awareness, everything feels more intense. It's easier to be caught out, it's easier to be spotted moving between cover, and it's easier for an enemy to take a shot at you. While this doesn't feel too bad in the earlier phases of the game, once you start reaching the likes of Pittsburgh the game really ramps up, and even more so than it did in the previous two versions. Plus the hospital sequence in the closing chapter feels much more exciting than before. In each case I was getting flanked and blindsided more frequently, catching me off guard as one enemy put on the pressure while another snuck around.
There are some elements that are either unaddressed or in some cases, it’s unclear whether any change was made. One of the first that comes to mind, is just how much flexibility the AI has in different encounters. During my research for The Last of Us Part II I learned not just about the active NPC limit I've already discussed, but also that due to technical limitations of the time several of the encounters in the original game are heavily scripted. It's difficult to analyse that when you don't really know where to look, but I suspect that for smaller encounters that probably hasn't changed much. Given again it would violate the ethos of delivering a faithful reproduction of the game.
The other big thing is that the companion AI problems don't appear to be improved for the most part entirely: running around the sequences in Pittsburgh and the Suburbs, you will still have Ellie, Henry and Sam, getting in your way and in some instances during stealth they will still wind up running in front of enemies. Now in my original analysis, I conceded that's an incredibly difficult problem to solve in tight situations without heavy designer authoring, and in fact, The Last of Us Part II still has issues with this with Dina and the like in certain moments. But even so, I doubt the work in The Last of Us Part II could prove as effective in this game given the companion dynamic is quite different between games. In Part I, Joel is acting as Ellie's protector and often she has to stick close to him as you sneak around and has no real agency of her own. This is quite different from The Last Of Us Part II where you spend the bulk of the time with allies who are capable fighters in their own right. Hence the considerations for where to go and how to move are quite different.
But arguably the most fun I had with the AI in the game, was actually from playing with the accessibility options. The Last of Us Part I adopts the same sets of accessibility features as Part II. The tools range from visual modifications such as the high contrast mode, or the ability to slow down time in-game when you need it, and even making it easier to find your next objective, nearby enemies, collectables and other resources courtesy of a ping system. But also, there's a suite of options available for players to manipulate the AI to some extent. It's great to see the developers transpose these features to an older game: enabling players to enjoy the game as they wish. But for me personally, playing with the AI settings was great fun, given it allows you to revisit and study the game in a different light.
The accessibility options allow you to customise how responsive the enemies are to your presence. They take longer to spot you and are arguably less prone to hearing you make loud noises. These are fun features in and of themselves, given they allow you to observe a lot of the patrol, search and attack behaviours up close. Perhaps my favourite feature for this purpose was the ability to go completely invisible to enemies by holding down a button. It really lets you see a lot of the AI in action in a way you wouldn't be able to lest someone mods the PC version of the game when it comes out.
But as we finish up here, I want to stress the value of these tools beyond a professional idiot on YouTube running around testing the senses of enemies. As I said already, if these tools allow you to play the game for the first time, given your circumstances prevented you from enjoying it, then I am so happy for you. I hope you have a great time with this game.
So, if you're a big fan of the series or have never played the original, I would argue that Part I is the definitive telling of Joel and Ellie's story - until they remaster it again a couple of years from now. You can see the impact of the sequel and its AI toolchain on the game in several ways. Though mechanically the game stays as it is, and as such it doesn't give players or the AI access to all of the new features that exist in the sequel. So yeah, your mileage is going to vary based on your interest in the series. But if you've played through say The Last of Us Remastered on PS4, it doesn't change things up enough that it feels like a drastic rework. In fact that kind of sums up the experience for me: The Last of Us Part I plays like how you remember The Last of Us Remastered, which is in many respects a compliment to the hard work of the development team to make it feel similar enough that it ties into those prior experiences and memories of the game you've crafted.
Thanks for reading AI and Games! Subscribe for free to receive new posts and support my work.