Square Enix’s AI tech preview of Portopia is off to a very rough start

The bots are not chatty

Last week, Square Enix announced a remastered version of a classic adventure game, The Portopia Serial Murder Case, that would be a showcase for its AI tech. The tech preview is out now, and well, it’s not going over well.

For context, The Portopia Serial Murder Case is a classic adventure game from the ’80s, authored by Dragon Quest‘s Yuji Horii. This actually marks its first official release in the west, years later, and it’s got a new visual style and layout to go with it.

Much like the original Portopia, this new version is also a text-parser adventure. This means you type commands in, or speak them using Speech-to-Text in the new version, to investigate and explore. The 2023 Portopia uses the framing of you (the boss) ordering around Yasu, a junior detective. He apparently needs a more hands-on approach for murders.

The AI tech preview part of this comes through NLP, or Natural Language Processing. I wrote a bit about this around the announcement of Square Enix’s preview, but basically, it means reading and understanding casual language, then issuing the correct response. In essence, it’s something that many text adventure games have done for a while, from Zork up through Facade.

This new version would use Natural Language Understanding (NLU) to help the detective understand the player’s instructions. Originally, there were even plans to generate natural replies for situations where the system did not have a preordained answer. This was scrapped, though, as Square Enix was concerned about the risk of the AI generating unethical replies.

So, ultimately, it’s a remade version of Portopia with more natural language tools under the hood and Speech-to-Text capability. How’s it being received? Not very well.

Talk to me, Yasu

At the time of this writing, the Square Enix AI Tech Preview of The Portopia Serial Murder Case sits at a Very Negative rating on Steam, with only 10% of users leaving positive reviews. Most of them found frustration with the actual AI partner, describing situations in which Yasu simply gets stuck in response loops.

Not one to simply read the reviews, I decided to venture into the AI adventure myself. Things started off okay. A few text bubbles gave me some context, and I soon met Yasu, my partner on the case. Our relationship was fine, at first. I asked him a few basic questions, asked after local suspects. In a few cases, I had to revise something I said. But as long as I asked the most straightforward questions, in a very robotic manner, I got some info.

Then we went to the crime scene.

Here, I found myself trapped in loops as I asked very basic questions of Yasu. I can understand some difficulty with questions like “was the murder weapon found at the scene,” though that should still be pretty answerable. Others, like “did the victim live alone,” could not seem to garner any information.

Taking notes

Here’s where I hit another major snag. For one, there is a list of important topics: suspects, locations, and objects. But there isn’t really a “journal” or any other record of what’s been said. Pen-and-paper seems like a necessity to remember anything but the most surface-level details.

It also feels like the NLU Visualizer, a tool Square shows in their promo images on Steam, would help with clarifying and narrowing in my inquiries. Problem is, I use a numpad-less, tiny keyboard that does not have the button Square has assigned to it. There’s no key rebinding, either. So aside from assigning a specific system shortcut for the rarely-used “Pause” key, I was out of luck. From the sounds of it, it doesn’t seem like the NLU would have helped a ton anyways.

Yasu doesn’t provide much help either. Every time I shook him like a magic 8-ball, trying to elicit an answer to my inquiries, I’d get frequent “Hmmm” or “We should focus on the task at hand” responses. Putting aside a need for oddly specific phrasing in a tech preview meant to highlight natural language understanding, there’s very little guidance to inform you how to better shape your answers.

The Steam community forums has threads aggregating the right phrasings to get certain pieces of info. But at that point, we’ve circled back around to the original text-parser problem we were trying to address. My attempts at casual language rarely worked, and even specific inquiries about information I knew Yasu had didn’t’ click if I didn’t ask about them in the right way. Eventually, I found that “ask around” is the best way to get a push in the right direction. But that’s something I only would’ve learned from other players.

A new case

On top of that, this just isn’t a very impressive remaster. I’ve cited past projects like Famicom Detective Club that have paid a fairly loving tribute to their original works. By comparison, Portopia feels less lively or captivating. The mystery can never be interesting, because I’m too busy trying to solve my phrasing to get overly invested in the unfolding drama.

Characters don’t have a lot of life to them, backgrounds are fine, and its eerily quiet in a way that doesn’t feel intentionally creepy. At the very least, this is free. And if you’re deadly curious, you can have some fun trying to get Yasu to tell you where he found a knife. As a tech preview, it’s still got some wrinkles to iron out. And as a localized version of a classic, it hasn’t left a strong impression on many people.