Remember this wave of artificial intelligence tide when the rise of the world’s AI lab liked to emphasize competition.
The most classic is that in 2016, Google Depepmind Lab let AlphaGo compete against Li Shishi, the human world champion was picked down, causing a panic. In 2017, when Musk Neurallink promoted brain-computer interface technology, the keynote was to redefine human beings and pursue human-machine integration; even OpenAI, a non-profit organization founded by Musk, defeated human players in the “DOTA” game as an AI player. milestone. When some artificial intelligence laboratories release new models, they also like to pursue labels such as “the world’s largest, “breaking record”, and “allowing AI to run 1.8 billion times for me in 1 minute”.
However, if you open the artificial intelligence information today, you will find that the current AI research, although there are competitions and competitions, also has a lot of softness.
The AI creativity of swiping the world is to let the celebrities in the photos “ant ah hey ” together , “AI resurrects the dead relatives and lovers” to make the headlines, epidemics, floods and other critical junctures, there is no shortage of researchers with AI to help Silhouette.
From showing strong to becoming soft, the changes in AI research may not be a microcosm of the progress of AI industrialization and industrial AI to the present.
From niche to popular: more and more soft AI
Today, AI is killing the Quartet and crushing the “artifact” of mankind, making people’s hearts soft.
In a report in the San Francisco Chronicle two days ago, the software that used artificial intelligence technology to create a surreal chatbot reproduced the conversation with the deceased fiancee, using the GPT-3 model, which can achieve amazing accuracy.
Using artificial intelligence to regenerate the dead, similar technology applications actually appeared as early as 2017. The American journalist Flaho recorded all his father’s words and designed a chat program to simulate the conversation with the late father. But at that time, only a set recovery procedure could be used, which was closer to an experimental idea.
The story is that the AI model was trained 8 years after the heroine had passed away. The emergence of the super-large-scale language model GPT-3 allows AI to make personalized creations based on the personality characteristics of the deceased. You can have a conversation like in the picture below. This AI model really seems to have its own ideas. It is curious about the surrounding environment and uses symbols to make expressions. The most mysterious thing is that it seems to have a keen perception of emotions and know how to say the right words at the right moment.
This means that even if you are not a programmer, you can still simulate the thoughts and thoughts of the deceased in the digital world if you fail to keep the corresponding voice data in time before passing away.
Speaking of softness, one has to mention the famous laboratory MIT media lab, whose researchers have recently directly transformed the rigid physical music keyboard into textiles using digital knitting technology. The keyboard made of fabric can be easily folded, rolled up and packed in luggage like a pair of socks or scarf. It can also be worn to expand the occasions for playing music.
It is not easy to realize the “softening” of the physical keyboard. It is necessary to pay attention to the combination of conductivity and fiber. The keys can simultaneously sense changes in touch strength, stretch and pressure, whether they are alone or in combination. This also provides a new interactive and tactile experience for the innovation of traditional devices.
What’s interesting is that in recent years, the popularity of flexible screen smart phones, flexible display screens and other products may also indicate that when the hardware consumer market is eager to seek changes, it is a good choice to go from hard to soft.
Similar experiments are also carried out in the Berkeley BAIR laboratory. Researchers have been trying to challenge the agent to complete soft fabric tasks. By folding clothes to train AI imitation learning, reinforcement learning, self-supervision, visual prediction, depth sensing and other capabilities. To continuously improve the performance of the model.
This is because, compared to hard and fixed objects, the state and appearance of soft fabrics in activities are more unpredictable. Let the agent try to challenge soft objects. In the future, it will help AI to achieve more complex in the real environment. application.
In addition to the sense of touch, the AI that everyone sees seems to be slowly becoming softer.
Recall, AI hot just when most heard name far and near what applications? In 2017, a new algorithm GAN (Generative Adversarial Network) appeared on the market. Then NVIDIA used it to develop a system that uses the celebrity photo database on CelebA-HQ to generate extremely realistic portraits.
At that time, there was a huge controversy, thinking that AI might be too powerful to be used for fraud (this fact does happen).
However, you must have noticed that many popular AI image applications are more emotional nowadays. For example, blogger @大谷Spitzer used AI technology to repair the image data of Beijing’s late Qing Dynasty a hundred years ago, making people dream about history.
Some programmers repaired the photos of their parents when they were young, so that their faces remained clear and dazzling even after passing through time, and they could even make various expressions. Memories will fade eventually, but AI also allows precious memories to last forever.
How did the increasingly soft AI emerge?
Perhaps soft AI is more popular
In a word, the changes in AI in recent years have become closer to people.
It must be admitted that from 2016, when AlphaGo was born, to 2021, there have been few major breakthroughs in basic AI innovations. Deep neural networks still dominate. DeepMind’s AlphaFold and OpenAI’s GPT-3 can be regarded as one or two in the past. During the year, it was one of the few products that made headlines in global technology media. Later, although news such as “domestic model surpasses GPT-3” and “AAphaFold2 again against the sky” are exposed from time to time, most of them are only concerned by industry insiders.
At the same time, the proliferation of various solution providers, individual developers, AutoML technology and development kits has further strengthened the application value of AI.
The process of AI landing is also a process of constant charm. When ordinary citizens waving unmanned vehicle stopped, the slack season with intelligent voice assistant dialogue was enjoying themselves , robots walking in public places, restaurants and other campus hotel, travel brush face become standard airport high-speed rail station …… “aggressive AI former “Naturally, the temperature was cooled.
The first to be cooled is the self-testing of the eager research capabilities of the giants. Of course, it’s not that AI giants don’t pay attention to top conferences and competitions now. It’s just that they are not attracted by the craze (and hot money) as before and after 2016. They love to attract scientists and top talents by scouring the rankings and relying on the laboratory to recruit. The eyes of the public and investors. Today everyone realizes that determining the success of an enterprise’s AI strategy depends on the ability and efficiency of the docking industry, not just the list of laboratories and the number of papers.
At this time, you may also fall into a certain kind of passiveness because of advocating data, such as refreshing the results on certain specific data sets, and “exam-taking” operations that may not achieve the same effect in real usage scenarios.
What was subsequently cooled down were some operations that once looked cool but couldn’t empathize with others and make them understandable. Simply emphasizing technology is likely to trigger public resistance.
Previously, the training videos of Boston Dynamics repeatedly hitting and molesting robots caused a lot of dissatisfaction among the audience, and the public’s empathy even made pictures of fake robots counterattack popular. Waymo, an unmanned vehicle with a very long road test mileage, has also encountered resistance from Phoenix residents before; Tesla , which has repeatedly made troubles, has also caused public and media dissatisfaction in the United States, as well as multiple security investigations from the government.
The imperfect status of AI technology determines that it will inevitably encounter various problems in its landing. If it cannot effectively communicate with the public, perhaps the next cold winter will be waiting far away.
What does a softer AI look like?
Specifically, being soft means reducing the sense of attack.
The strength and crushing that was once regarded as having a sense of power will only keep people away. Throughout the world’s well-known AI research institutes and corporate research institutes, while playing competitions, we must also consider benevolence and ethics. Use AI to protect tropical rainforests, save endangered animals, or help people with various disabilities and participate in epidemic prevention and disaster relief. Technology continuously releases goodwill and wins respect for enterprises.
Softness also means calmness and collaboration.
When AI enters the acceleration stage of industrialization, it needs to cooperate with all walks of life, various suppliers, developers, and listen to the voices of industry experts, data clerks, designers, engineers, etc., in order to make technology truly land in the industry and exert its effectiveness. New trends such as on-demand cloud services, federated learning, and hybrid multi-cloud also emphasize collaboration rather than competition. Better distribution of profits to partners is the key to the AI giants’ defense of the city.
Softness also means enhancing humanity.
Today, it seems that more humane AI innovations are indeed easier to gain the favor and attention of the public. In 2018, the digital agency Isobar put forward the concept of “Augmented Humanity”, believing that enhancing humanity will allow humans to experience and experience the world more deeply in different ways, which is being verified in AI innovation.
For example, during the 2020 new crown epidemic, various AIs used to accelerate vaccine development, assist doctors in diagnosis and treatment, perform hospital disinfection, and check body temperature in public places have received unanimous praise.
The Tokyo Olympics, which everyone has been paying close attention to recently. Earlier, clothing brands focused on algorithmic insights into emotions when designing sportswear, in order to make athletes feel “When I put on this dress, I feel like a superhero.” “.
These examples all illustrate that good smart products and services should be a combination of human nature and science, technology and art.
As stated in the “Tao De Jing”, soft is better than rigid, and the world knows nothing and cannot do it. Isn’t this the direction of Ai today and even in the future-not only to be known to the world, but also to be felt and used by others.
Under the ubiquitous business atmosphere, the AI ”scientific research” of enterprises is naturally even more inseparable from the market. More “human” and softer innovations will surely sprout in more diversified AI.
Posted by:CoinYuppie，Reprinted with attribution to:https://coinyuppie.com/from-niche-to-popular-why-global-ai-innovation-has-become-soft/ Coinyuppie is an open information publishing platform, all information provided is not related to the views and positions of coinyuppie, and does not constitute any investment and financial advice. Users are expected to carefully screen and prevent risks.