The digital life plan in the wandering earth has been launched? DeepMind creates fruit flies in computers, netizens: Can it create humans?

"Question Tu Hengyu, understand Tu Hengyu, and become Tu Hengyu."

In "The Wandering Earth 2", Andy Lau plays Tu Hengyu Hengyu is an impressive character. In order to let his daughter who died in a car accident have a "complete life", he ignored the human world's ban on the "Digital Life Project" and has been secretly working alone to improve the structure of digital life, and finally decided to openly violate the rules and upload his daughter's data to quantum computers, and was later imprisoned for it.

## The digital life map Yaya in the movie "The Wandering Earth 2".

After the movie was released, the topic of "digital life" has been discussed a lot.

Recently, this topic has been brought up again. The reason is that many people who have lost relatives and friends are trying to use AI technology to "resurrect" the deceased and create a series of images containing the deceased. virtual video. A comment "Question Tu Hengyu, understand Tu Hengyu, and become Tu Hengyu" has been liked tens of thousands of times.

However, most of these videos are not interactive. According to current technology, even if the interactive function is set up, the person on the screen will look fake because it violates some physical intuitions. After all, understanding the physical world and moving and thinking according to the laws of the physical world is still an unsolved problem. Even the recently popular Sora video generation model is considered to be unable to do this.

What's more, a "digital life" that can truly resemble the real thing is considered to have independent consciousness. There is no consensus in the scientific community on what consciousness is and how it is constructed. Therefore, true “digital life” is actually still far away from us.

#However, artificial intelligence scientists and neuroscientists are joining forces to work in this direction.

Neuroscience Artificial Intelligence: Create a realistic fruit fly in the computer

This is a virtual fruit fly jointly developed by Google DeepMind and Janelia Research Park in the United States (a neuroscience research institution established by the Howard Hughes Medical Institute). It can walk and fly like a real fruit fly.

^{This video shows that the virtual fruit fly reproduces the flight motion of a real fruit fly (spontaneous turning), performing at a speed of 2 cm/s Walk command, turn left and right at the same time. The model also mimics the walking trajectory of a real fruit fly, including walking at different speeds, turning and briefly stopping.}

It is also the most realistic fruit fly simulation to date, combining new anatomically accurate models, fast physics simulators and fruit fly-based simulations. Artificial neural network trained on fly behavior to mimic the movements of real fruit flies.

#In addition to walking and flying on complex trajectories, virtual fruit flies can also control and guide flight with their eyes.

“You take real fruit fly data—how they fly, how they walk—train the network to mimic those movements, and then let’s train this The network controls the flies and tells them how to move," said Roman Vaxenburg, a machine learning researcher in Janelia Turaga's lab who led the project. "It's like a small brain that controls the fly's movements," he said.

The new model is the team’s first iteration of the virtual fruit fly, which they plan to make more realistic by utilizing additional anatomical and sensory features as well as real neural networks. It is also the first in a series of realistic animal models they hope to achieve. They and other researchers can now leverage this common open source framework to develop these models.

These models can help scientists more fully understand how the nervous system, the body and the environment work together to control behavior. Researchers have been exploring these questions in the laboratory with real animals for decades, and realistic virtual models will allow scientists to understand how all these components are connected to each other, and factors that cannot be measured in the laboratory - such as How forces exerted on the body during flight affect behavior.

"The simulation of the body can tell you how the instructions of the nervous system are translated into actions and behaviors, and this "how" is related to the shape of the body and how the body interacts with the world "It's all about the physics of the interaction," said Srinivas Turaga, senior scientist on the project and leader of Janelia's group. "All of that is programmed into this physics simulation."

^{This video shows a fruit fly model executing a direct flight command at a fixed altitude at a speed of 30 cm/s, and then the fruit fly model reproduces Flight maneuvers of real fruit flies: avoidance of perceived threats and spontaneous turns. What is shown next is a fruit fly model executing a walking command at a speed of 2 cm/s while turning left and right. Then the fruit fly model simulates the walking trajectory of a real fruit fly, including walking at different speeds, turning and briefly. stop.}

The new model builds on previous work simulating fruit fly behavior, including using simplified fly bodies and manual control systems to simulate flight. "Grand Unified Fly". The recently launched NeuroMechFly uses a realistic body model and a manual control system with a learning component to simulate walking.

In the new study, Janelia researchers and DeepMind scientists, led by senior research scientists Yuval Tassa and Josh Merel, set out to improve the anatomy of a fruit fly model science, biomechanics, physics and behavioral information to create a more realistic fruit fly simulation capable of performing a variety of behaviors. This work is one of several collaborations between Janelia and DeepMind, which leverage their respective expertise in neuroscience and artificial intelligence to jointly solve scientific problems.

#Matthew Botvinick, senior director of research at DeepMind, said: "While it is generally acknowledged that understanding brain function depends on understanding the body and its interactions with other physical objects, computational neuroscience Research rarely seeks to model things at such a global level. While brainstorming with Srini and other members of this team, we realized there was an exciting opportunity to bring all of this together in the context of Drosophila research. The pieces come together."

In the future there will be virtual mice, virtual zebrafish, virtual...

# #The entire fruit fly creation process can be summarized as:

First, Janelia research expert Igor Siwanowicz used a microscope to image various parts of the adult female fruit fly and used a computer The software built an anatomically accurate in vitro virtual model of the fly, including the movement of the fly's joints and appendages.

DeepMind researchers—including Tassa, Merel, and research engineer Guido Novati—converted this virtual model into code and fed it into the MuJoCo simulator, a A fast open source physics simulator designed for robotics and biomechanics. The tool allows researchers to virtually simulate how objects in the real world move and interact.

To support the fruit fly model, the researchers made significant upgrades to the simulator, including adhesion actuators to simulate the forces generated when insect feet grip a surface. The team also asked Novati to design a new hydrodynamic model to describe the forces experienced by fruit flies as they fly through the air. The model can support a variety of aerodynamic behaviors, including wing-spreading, said Tassa, senior author on the project. Here, they used end-to-end reinforcement learning.

                                                                                                                                                                                                                                                          Flight simulation.

# Walking simulation

#                                      Visual guidance flight missions: altitude control and obstacle avoidance.

Tassa also said: "Because the forces acting on the fruit fly are so small, modeling such a small insect is very challenging."

Next, Vaxenburg built an artificial neural network and trained it on real fruit fly behavior by feeding the network recordings made by experts in fruit fly behavior Video message from experts including Janelia senior group leaders Kristin Branson and Michael Reiser, HHMI researcher Gwyneth Card, and Caltech professor Michael Dickinson.

"Our goal is to improve fidelity, and that's achieved by working on two fronts: One is improving the capture of anatomical detail, the structure of the fruit fly," said Vaxenburg. ;The other is the catching behavior, that is, the actions and reactions of the fruit fly."

^{The above picture shows the body structure and degrees of freedom of the fruit fly model. The Drosophila model consists of 67 body parts connected by 66 joints, equivalent to 102 degrees of freedom. The figure shows a sequence in which all degrees of freedom move in a sinusoidal manner.}

This new model is just the beginning. Next, the team hopes to incorporate other parts of the fly's anatomy, such as muscles and tendons, as well as realistic sensory systems into the virtual insect to create a more realistic fly model. They also hope to be able to use real neural networks, such as the Drosophila ventral nerve cord connectome, to power the model.

Now that the research team has demonstrated that they can create these types of realistic virtual models, in the future, they also want to create virtual mice and zebrafish, two creatures that neuroscientists Research extensively. The process they used to create virtual fruit flies is also freely available to researchers around the world, allowing others to create their own realistic models.

Turaga said: "We have shown how to do this, and we can do it again for another organism."

Currently, a paper on this research result has been published on bioRxiv. The authors write in the abstract of the paper, "The animal's body determines how the nervous system generates behavior. Therefore, detailed modeling of the neural control of sensorimotor behavior requires a detailed body model." To this end, they developed a detailed model of the body in the MuJoCo physics engine. An anatomically detailed whole-body biomechanical model of Drosophila melanogaster is provided, which is the virtual Drosophila we introduced earlier.

Paper link: https://www.biorxiv.org/content/10.1101/2024.03.11.584515v1

On social media, this research attracted the attention of many netizens. Some people asked: Will it evolve? Someone else asked: Can they also make a model that simulates a human being?

These netizens even provided new research ideas:

This Everything looks promising.

However, despite these relatively distant associations, this fruit fly research is also very meaningful for current medical care research, and can help from drug discovery to disease prevention. Research in a series of fields such as modeling.

^{Reference link: https://www.janelia.org/news/artificial-intelligence-brings-a-virtual-fly-to- life}

The above is the detailed content of The digital life plan in the wandering earth has been launched? DeepMind creates fruit flies in computers, netizens: Can it create humans?. For more information, please follow other related articles on the PHP Chinese website!