In the days when the latest season of Love,Death & Robots is broadcast, there are full of discussion in theWeChat moments! In addition to once again continue the story that the audience expects, it also once again shows how exaggerated the “brain hole” can be
The aesthetically bursting and full of metaphors, Jibaro, and The Swarm, which perfectly interprets the relationship between individual memory, genes and wisdom, make people constantly question themselves in just ten minutes of exclamation and applause: Why did I not Such a fancy mind as a screenwriter and director? In addition to urging the creative team every day, how can we achieve “mass production” of high-quality ideas?
“Opening the mind”, a long-standing scientific problem
Most of the ideas of Love,Death & Robots can be traced in reality – Jibaro is based on the real history of Europeans’ colonization and plundering of indigenous peoples in South America, and The Swarm is derived from profound thinking about the change of real civilization order . Looking at it this way, Minds are often the migration of the brain’s learning ability based on reality at the level of imagination.
So how does the brain achieve complex creative output?
Through human research in the past, we know that the human brain is a complex system carrying multiple functions. Scientists decompose regions that can achieve different functions into different modules. As the research on brain regions continues to deepen, it is found that the motor module and the visual module communicate with each other in the process from unfamiliar to proficient, so that people can quickly understand the nature of the task and master it [1]. When going from reality to creative output, the brain is based on various fragmented memories, and the brain network related to imagination is activated, and then many brain regions in the brain responsible for different divisions of labor are involved in the process of imagination. For example, in an experiment, volunteers were asked to make up a story based on three words (related or unrelated), and it was found that some seemingly unrelated brain areas were also involved in the process of story creation in the brain [2]. When the functional areas of the brain integrate various seemingly unrelated information and draw inferences from one thing into a new reality, it is amazing!
In addition to the world in the play, Newton was smashed by a falling apple and extended to gravitation. Einstein was inspired by the law of electromagnetic induction to study the theory of relativity. Many great scientific discoveries and creations are based on the human brain. There is nothing to maximize the creative expression of imagination. And each person’s brain area is different in activity, which determines how fancy the brain is to varying degrees.
Before you master the human brain, draw it first
The human brain created by “Hand of Nature” can be said to be the most sophisticated instrument today. It is like a black box that can automatically integrate fragmented memories to achieve 1+1 > 2 or even > 10 creative expression, creation, research A variety of classic works and theories. So back to the question at the beginning, with the development of today’s science, can AI be used to “industrialize” the production of imagination and bring more excellent output?
In order to better simulate the thinking process of the brain, we need to understand the brain more fully, and map the internal structure of the “black box” of the brain, that is, the brain atlas. Brain imaging analysis plays a crucial role. The process of studying the brain is like exploring a vast universe in the dark. If we can crack the code of the human brain, then perhaps the reference answer to the laws of nature is in front of us.
It is precisely because of this that the research on brain science has also become the focus of scientific research in many countries. In September 2021, my country also launched the “China Brain Project” – “Brain Science and Brain-like Science Research”, which involves 59 research projects. Research areas and directions.
Why is such a high investment required? In fact, just as mentioned above, dividing the area of a healthy brain is already challenging. For example, the functional connection of the human brain is affected by various factors, and it may even be collected at different times in the morning and evening. The functions of the performance are also different. This process requires the collection of a large number of MRI and PET imaging results, but it is a bit too long and cumbersome in the face of countless extremely complex neurons, branches, and synapses in the brain. Therefore, it is necessary to perform statistical tests on the parameters, and perform a large number of complex fittings and assumptions, so that relatively reliable results can be obtained.
As the saying goes, “a thousand people have a thousand faces”, everyone’s brain is different, and many studies lack a large number of sample data, only dozens or hundreds of “small data”, such as Hawking’s ALS, which is very “personalized” “Rare diseases, the incidence rate in my country is only 1-2 people/100,000 people, as well as other diseases such as autoimmune encephalitis, cerebral small vessel disease, and hypermemory syndrome. Suffering from these diseases is painful, but the samples of brain imaging data that can be used to analyze patients are severely lacking. If these problems are not solved well, using a computer to completely simulate the working mechanism of the human brain will be a dead letter.
Just like when artists and scientists create scientific research, different functional areas of the brain automatically use fragmented memory to integrate new creative expressions. Can we use small sample data to build AI models to train more universal large sample predictions?
A study by the National University of Singapore and ByteDance’s top journal in the field of neurobiology, Nature Neuroscience, found that training AI models through meta-learning and meta-matching methods can largely solve this problem.
The effect of the meta-learning method has exceeded KRR (a linear regression prediction method)
In this study, scientists used meta-learning, a cutting-edge method in the field of AI, to transfer machine learning models trained on large datasets to small datasets and train equally reliable AI predictive models.
Seems a little complicated, right? But you can understand this process as the human brain’s clever inference in the meta-learning process, the learner needs to be constantly shown hundreds of thousands of tasks, and eventually the learner will learn the knowledge of many tasks. After learning, there are two stages: the first stage focuses on quickly acquiring knowledge from each task; in the second stage (the learner), the information is slowly taken and digested from all tasks. By solving the problem of less data, machines can learn faster from a small amount of data like humans, and artificial intelligence is also more like humans!
When we stand on the springboard of the future,
Where is the way forward?
You may be curious, why is ByteDance related to AI and medical care?
In fact, the main work of the basic research team of intelligent creation participating in the research is to explore cutting-edge machine learning, computer vision and natural language processing technologies to solve challenging problems in the field of artificial intelligence. Many popular filters and special effects on Tiktok are implemented by this team using AI technology. Their technical capabilities are also providing services to external enterprises through the VolcEngine, a cloud service platform under ByteDance.
Original technical directions such as brain atlas, knowledge modeling and machine learning were once the “cold bench” in the field of computer vision and medical image analysis: it was difficult, long cycle and few researchers. Now, with the participation of technology companies, this situation will be changed.
The methods mentioned in this paper have brought new ideas to machine learning and brain science in the whole AI field, that is, machine learning methods in AI Artificial intelligence field will be used to truly solve clinical application problems, especially personal precision medicine and the treatment of rare diseases. Today, rare diseases still face many difficulties such as misdiagnosis and refractory, but if AI is introduced, In the past, those diseases whose sample size was too small to carry out subsequent clinical diagnosis will also have the hope to have a clearer precision medical treatment plan and save more lives.
In addition, theoretically speaking, this technical achievement is also expected to make “mind reading” based on brain imaging a reality. On the premise that large-scale data sets are used as pre training models, AI models that can predict human intentions can be trained. This is also similar to the “brain computer interface” of Elon Musk’s “neuralink”: both hope to “understand” the brain through technical means.
AI itself has a very broad prospect in the field of neuroscience research. When we have a better grasp of small data scenarios, and some brain regions can be predicted, will the complete prediction of other functional regions of the whole brain also be within reach? It seems that there are new ideas and entry points for the subsequent understanding and development of the human brain, and there are better practice cases in the field of AI combined with brain neuroscience in the future. The idea of “storing and retrieving memory” and “superhuman” in science fiction is no longer out of reach.
The human brain is the most precious gift of nature. The warm cells and interwoven nerves make up the universe closest to everyone but furthest away. The sense of mystery is full.
Fortunately, when AI technologies, including meta learning, become more mature and reliable, although it is still a little far away to industrialize the brain, we can also make a leap in exploring, expanding and controlling the brain. In the process of understanding the sun, moon and stars in our minds, we can finally better understand ourselves.
Comments