A simple look around the games industry shows that frankly, there aren’t a lot of games affecting players very effectively on a range of psychological levels outside the normal for a game except horror games. Many many indie games have attempted to create profound interaction, but frankly they fall on their face by nature of not using the medium of interactivity. Games that attempt to engage players on these levels have no capability to do so except those afforded by film and choose your own adventure books. Standard game play has the potential to be broad and open ended, with many complex interactions and sub-interactions. Games whose game play is primarily based on narrative elements are very limited in comparison.
I find this trend rather odd in some ways, but given current technology, it’s to be expected. In real life, some of the most complex things we do are interact with other people, in real life, online, over the phone, and otherwise. In real life interactions we have tremendous range of expression, between tone of voice, body language, facial expressions, and actual word choice. Games by comparison have very little word choice. The games we call RPGs have very limited roles for us to play. Even the best in the genre have at most 16 different dialog choices for a given question. Most people reading this are probably reeling at the idea of having as many as 16 choices for a single input query because they’re so used to maybe 6 at most. However frankly our interactions are limited not just because there are only 16 choices, but because these are discrete choices at all. Combat situations have options that are rarely so discrete. They have analog information that dialog trees will never have. In combat you are not choosing between pre scripted scenarios, you are creating one yourself. The scenario emerges from your actions and the submodalities of those actions. A dialog tree has no submodalities. There is no “unit of choice” smaller than the discrete options you have selected and the sum of those options compiled.
The big reason the majority of our games have focused on combat is because combat is a topic absolutely ripe with analog information in the sub modalities of actions. The location of an attack, its hitboxes, the time it occurs, the angle the attacker is facing, the velocity of the attacker, the distance of the attacker relative to the target of the attack. All of these have so many permutations between them that it enables tremendous emergent complexity. We have no such thing in dialog, because dialog is so reliant on submodalities we cannot emulate. I do not believe that dialog will ever really be a gameplay function on the same level as functions of combat until we are capable of AI with the ability to parse natural sentence structure and reply adequately. We have a few AI right now who can approximate these types of effects rather convincingly, but it’s pitifully easy to break these, and very rarely are they capable of holding a conversation. Their methods of parsing language are limited to reductionist methods (homing in on keywords relative to others, reducing a sentence to its barest level of complexity) and Bayesian filtering in order to produce semi-coherent results. In short, they have no natural understanding of sentence structure, they are only capable of recognizing and replying to trends in sentence formation.
In short, current AI technology cannot understand language well enough to facilitate natural conversations.
Video games are in many ways incapable of precisely modeling real world analogs. A vast number of our real world sports have a huge emphasis on precise body motions, but copying that in a gameplay context is frequently close to impossible. Upcoming VR technology seems to have a huge chance to change this, and in a few limited cases, like wii sports, already has, but we still have many games which require abstractions to represent things like the strength of a swing or the English on a ball. Conversation is similar and yet we see those things rarely because developers know that would never be remotely convincing. In terms of conveyance of understanding for conversations, abstractions like a bluff meter make absolutely no diagetic sense. They are not relatable, they are therefore not easily understood in any intuitive sense, and make poor gameplay mechanics. It gets even worse in that dialog responding to such bluff meter would be selected in a binary manner same as all other dialogue. Mates cut flute maturation Steve (auto correct fucked this sentence up, left it because it’s funny). Computers cannot formulate natural sentence structure in response to an analog input of any kind.
Beyond this, on an emotional level, I don’t think we’ll be able to relate to AIs until they pass the Turing test on a more specific level of interaction. Current Turing tests tend towards broad rambling conversations that don’t have specific directions really. What if an AI is to play a specific role? Not just “human”? (This section is to be expanded on)
We speak in one of the most complex languages on the planet, with the most words and arbitrary rules, yet in (video) games we have none of that complexity. People ask us why we haven’t had a “Citizen Kane” of video games yet. Games are the medium of interactivity, yet our dialog sequences, no matter how complex, are still practically devoid of it.
Games are limited in the types and styles of emotions they can convey effectively because they have a requirement to focus on interactivity. As they engage more thoroughly on comedic, romantic, or other emotional levels, they lose interactivity, because we cannot craft those in an interactive format without the use of live human beings. All that our controllers afford us is broad open actions. There are many things we still cannot capture, like the finesse of fencing. We are forced to reduce these things so they can fit the limited range of what our controllers can input. A controller has no analog to dialog so until it does we are stuck with shallow conversations. I am no AI researcher, so I cannot even begin to provide any sort of recommendation for how to craft more believable AI, but what I can say is that it must have a legitimate understanding of sentence structure beyond mere Bayesian filtering, or it has absolutely no chance of interacting with humans in a gameplay context in any meaningful way.
Personally, my thought is, if you want to have a game about dialogue, a game about emotionality, put people into it. Play with other REAL people and figure out a dynamic that has the players play off each other in a way that can be quantified and fit into the rules structure. Stop waiting for an AI development that’s unlikely to ever happen and make some friends.