You start Halo 4 from your last saved checkpoint, sit in your Warthog and speak into the headset’s mic,
“Cortana, call for backup from Infinity if more than 3 brutes enter the 50m perimeter around me”
and Cortana obliges “You got it, Chief”
There is a race for adding compelling use cases to make AI useful enough for daily use. Some of the well known and established Commercial AI assistants are :
- Google’s Assistant
- Apple’s Siri
- Amazon’s Alexa
- Microsoft’s Cortana
While these assistants have pervaded a lot of devices and use cases, they have not however entered an area fully where the users are the early adopters of anything remotely tech: gaming.
You might point out that OpenAI is the answer to the above. I don’t mean AI as bots or competitors in technically complex games, but as the digital assistants that they already are.
Now take this one step further, allow game developers to build on top of the assistant actions and library that will allow them to handle even more complex requirements relevant for their game. Requirements that change from game to game however complex or simple.
To make this seamless for players, the AI core needs to exist on the console itself and actions should be shipped with the game. All commands and actions will be handled offline powered by the >4 Teraflops machine sitting below your TV.
Where graphics improvements are now getting commoditised, supremely immersive games will be the only winners in the years to come. Immersion at this scale with AI can thus lead to different endings basis each and every action and choice by the player.
However, multiple endings is nothing new to the gaming world and has been represented fairly over the years. GTA 5, Far Cry and Dishonored to name a few.
But why have finite number of endings?
Why ship the game with an ending at all?
Every creative creation has a story. The better the story the more it succeeds. Humans are drawn to compelling stories and will always be.
Some of the best games of our generation were critically acclaimed because of their compelling story. However, in an era where unique content is the king of virality, millions of gamers see the same ending cinematic while finishing Spiderman.
It is time gaming evolved from if-then-else scenarios to non-linear and branching story lines because it is the only creative artefact that can do this right now.
Not movies, not TV-series, not novels but games.
If the game does not ship with an ending, it means it needs to adapt to the user’s choices and build further scenarios and quests itself.
It will require a general intelligence AI to fabricate the story on the go and Google / Amazon / Microsoft will be the vendors that the game will connect with online for processing all the decision variables being received from the player.
Another challenge will also arise, how do you ship a game with voice overs and cutscenes when you don’t know how it will end for a specific user. Generic audio bytes and cutscenes will kill the immersion and shipping with each possible ending would be operationally unrealistic.
Developers will now face the hard task of allowing the game to programmatically creating speech and cutscenes for every unplanned branch.
Clear tools to achieve this is recording common and expected words from voice actors and storing them for use in unplanned branches.
In a similar manner, expected and broadly traversed branches should be pre-computed and should sit on game’s online server.
Here is a mock process for realizing story branching in real time:
- Player has played the game for 3 hours and has traversed the base storyline
- At first decision node (that can affect the ending) player takes Option B out of 3 options
- Option B branch is now selected and handled locally by the game data
- At 2nd decision node, player takes Option C and thus lands in an unplanned story branch
- Local game communicates unplanned branch variables to server which contains the master game logic
- Local game makes player wait while waiting for server’s response by sending waiting communication to the player “You will be notified by Captain of the next mission after he receives orders from HQ” (think Call of Duty style)
- Server runs through computations against game constraints, generates world responses, new quests, character dialogues and scenarios
- Server sends response to local game at player’s end. Local game renders cinematics and uses text to speech for dialogues and sets new missions for player
- Repeat steps 4 to 9 till game constraints are breached which signals to the server that an ending needs to be computed for the player
- Server takes sum of player’s choices and computes ending character dialogues, narration, and repeats Step 8
With sufficient number of decision nodes, player’s decisions will lead to unique endings which will in turn lead to creation of undreamt storylines.
This is what might happen:
- Game developers will store easter eggs in some endings just to peak the curiosity of players
- Players will replay multiple decision nodes just to see what happens at the end
- Few endings will be so rare that they will be uploaded on Youtube and Instagram and will go viral!
- Creators on Youtube and Twitch will have a field day recording decision guides for other players
At the core of this gaming utopia, is a general artificial intelligence that does the hard work of computing alternative story paths every time a new branch is created by a player.
The current AI powerhouses like DeepMind and OpenAI are sufficiently capable of computing alternative storylines. Google Duplex’s current text to speech capabilities already sounds more human than most. Nvidia’s GauGAN can turn basic visual input into photorealistic paintings and hopefully cutscenes soon. The hardware, atleast gaming consoles like PS4 and Xbox, house tons of processing power already to render input from the above.
We just need to press Start.