Listen to this story
Westworld – an HBO series based on a Western-themed amusement park populated by robots – was a concept that left many craving to experience it. The series showed automatons living in colonies, where they interacted with each other in natural language and performed tasks such as farming, policing, looting, etc just like the humans.
Back in October 2016, when it was released, Westworld was something that seemed super sci-fi. However, the lines between truth and fiction are blurring faster than we anticipated, and seven years later, a world like that is no longer just a fragment of someone’s imagination on TV.
Now, researchers have successfully crafted an environment where generative agents create a believable simulation of human behaviour. They have developed a novel sandbox environment inspired by the popular game The Sims, which we believe could pave the way for more advanced simulations of human behaviour, akin to the Westworld.
The sandbox engine uses generative agents to populate the environment with twenty-five interactive agents, who can be communicated with using natural language. These ‘agents’ are capable of interacting with each other and the world through their actions, generating natural language statements at each time step to describe what they’re doing.
The researchers evaluated the generative agents and found that they produce realistic individual and social behaviours. For example, starting with just a single user-specified notion that one agent wants to throw a Valentine’s Day party, the agents autonomously spread invitations to the party, make new acquaintances, ask each other out on dates, and coordinate their arrival time. The researchers demonstrated that the components of the agent architecture – observation, planning, and reflection – each played a critical role in the believability of agent behaviour.
The movements of agents in the simulation are controlled by both the generative agent architecture and the sandbox game engine. When the model dictates that an agent needs to move to a particular location, the system computes a walking path to the destination in the Smallville environment and the agent starts moving accordingly.
Just like in Westworld, users can enter the sandbox world of Smallville as agents and operate within it. The user-controlled agent can either be an existing character in Smallville or a new visitor with no prior history. The inhabitants of Smallville perceive and interact with the user-controlled agent just like any other agent in the simulation. They acknowledge its presence, initiate interactions, and form opinions about its behaviour based on its previous action.
Westworld is closer than you think
Although the theoretical similarities between the aforementioned model and Westworld exist, we still have not yet achieved the level of sophistication in robotics as depicted in the series. However, there is an area in which we can potentially replicate the essence of Westworld – the Metaverse. In this realm, gaming companies are already developing models that can generate 3D renderings of human avatars in real time. Users can interact with these virtual agents, offering a level of immersion that was previously unimaginable.
INSTA is one such cutting-edge model that is capable of reconstructing a highly detailed and photo-realistic animatable 3D neural head avatar within just 10 minutes. The technology is made possible using a single RGB camera to record a short monocular video, which is then used to instantaneously optimise a deformable neural radiance field. This radiance field is embedded in a multi-resolution grid, based on a 3D face model that guides the deformations.
The resulting avatar is incredibly life-like, with the ability to be viewed from a variety of novel perspectives and animated in real time. The efficiency and accuracy of INSTA represents a significant leap forward in the avatar creation technology, offering a level of detail and interactivity that was previously thought impossible.
Companies like Meta have already made significant progress in the field of VRs and today they’re a ‘ready’ product in the market, although on the costlier side. Moreover, now it’s easier than ever to ‘feel’ people in virtual reality, thanks to Haptic feedback, a technology that provides simulated physical feedback, such as the vibration of the keyboard in a smartphone. When used with virtual reality devices, Haptic feedback allows the user to touch and feel things in the virtual environment that are, obviously, not present in the real world.
Danny Parks, VP of technology at Trigger XR, says, “There are several systems that generate ultrasonic fields to simulate touch, and others that use gloves or vibration vests to provide mechanical feedback. It will probably be some time before we see something that fundamentally changes the way we feel things in virtual environments.”
Now, there are several products available in the market which can draw you entirely into the metaverse. For instance, there is a full-body suit, known as Teslasuit, that simulates the forces on your body. Not just that, there are now VR attachments that can emit scent, water, and heat etc.
However, while all these developments are exciting, the issue of affordability still remains. Oculus VR, for instance, costs about $300, Haptic feedback gloves can cost you about $5400 a pair, which falls beyond the purchasing power of most people in developing countries. The Teslasuit, which is a much-needed suit for Westworld to work in the metaverse, costs a whopping $12,000.
So, while the development in the field is a welcome step towards building our own Westworld, the cost issue threatens to almost halt the development. With companies like Meta laying off its VR team, it’s highly likely that the field is not profitable for the companies, and we might not be seeing an affordable Westworld anytime soon.