Some of these advances in computer vision technology are driven by three main engines: 1. Very powerful computing power, as you can already see, especially with the spread of gpus, which allows us to train very complex algorithms. Big data. For its face recognition system, Google used 8 million people and 200 million photos to train their deep learning model. At this point, no one on earth could have seen so many people in their lifetime to train their brain’s facial recognition algorithms. Our system can swipe an ID card to see if he’s a legitimate holder.
Giiso Information, founded in 2013, is a leading technology provider in the field of “artificial intelligence + information” in China, with top technologies in big data mining, intelligent semantics, knowledge mapping and other fields. At the same time, its research and development products include information robot, editing robot, writing robot and other artificial intelligence products! With its strong technical strength, the company has received angel round investment at the beginning of its establishment, and received pre-A round investment of $5 million from GSR Venture Capital in August 2015.
From the perspective of algorithms, it is the technology of deep learning. Deep learning is not a new invention, but a Renaissance, much like the Renaissance, a replay of past history. Neural network and deep learning are the further expansion of the popular multi-layer neural network in the late 1980s. Its power came into play when it was combined with big data, supercomputing and so forth, leading to today’s technological advances.
In the last two or three years, there have been a lot of AI problems, tasks. Computers are gradually surpassing our human intelligence. This, too, is a historical certainty, and in many ways it has already happened. For example, you just saw us swipe id cards, to determine whether you are such a task. Right now, computers can do that if 10,000 people try to impersonate you, we can be correctly identified 95 percent of the time.
In what areas can AI surpass humans? The algorithms that ai relies on are deep learning methods. The problem with deep learning is what I call “data fertile,” and it’s good data fertile. Good data fertility means we have a lot of data to learn from.
In addition to inductive learning, we also have a type of learning called deductive reasoning, or deductive learning. If you look at Euclidean geometry, for example, this is deduced. For now, deep learning is only suitable for learning from data. It is more suitable for data collection, acquisition and annotation of more convenient fields. For example, computer vision, speech recognition, or there are more and more internet-based fields that make it easier for us to collect data. However, deep learning cannot solve the problem of reasoning.
Another important AI event happened last year, alphaGo beat the Go champion, and deep learning in AlphaGo accounted for 80% of the learning, but there is another technology, called enhanced learning. It is suitable for the field of automatic judgment of right and wrong, but is not suitable for solving the problem of computer vision recognition. Machines cannot judge for themselves, which makes it difficult to enhance learning by accumulating data about what they are doing right and wrong. If it is completely left to the machine, let it enhance itself, the current algorithm will lead to it learning bias, may be misled, learning silly.
Recently, Kaifu Lee has introduced on many occasions that he believes that in 10 years, artificial intelligence will replace many occupations in the world, and 50% of jobs may be replaced, including translators, journalists, assistants, security guards, drivers, sales and so on. Different people have different opinions. Many experts do not agree that all professions can be easily replaced by AI, but there are also many industries that are increasingly threatened by AI, such as security.
This is us in a unit, they now use our face recognition technology to do access control, the whole group has more than 10,000 people, he can open the door for everyone to come to work, attendance (face recognition technology). This system is just starting to work, and we believe that when this system turns all employees into acquaintances, it will be much better than our human security. A good security guard can recognize 1, 2000 people, but it is still difficult for enterprises with tens of thousands of people.
What areas will AI gradually surpass? One is huge space search problems, and the other is retrieval, such as image retrieval, which is a piece of cake for machines, but not so easy for us humans. In addition, the fields that rely on experience and skills, namely the so-called well-informed areas, may gradually be replaced by AI, such as face recognition, object recognition, or automatic driving. This is also a problem of experience, such as medical map reading.
Our artificial intelligence can combine hundreds of top doctors, through the study of these films can be more than many experienced doctors. A lot of customer service q&A is semi-repetitive, or completely repetitive. So it is entirely possible for ai to learn such skills from historical experience.
And the question is, to go beyond human intelligence, do we need to know how the human brain works in order to be able to create algorithms that go beyond human capabilities? In fact, how our human brain works is still a very mysterious thing, and it is also a subject worth studying. The good news is that we don’t need to be brain-like, and if we stick to one brain-like path, we won’t be able to outdo people.
Our current AI can be simply summed up as an algorithm, or a model, plus a method of data, which allows our machines to learn from a lot of data, more data than we humans can see, but it’s better than human representation and classification.
Take Go for example. With the emergence of Alphago, our go experts and players have begun to break through some of the past thinking frames and learn from Alphago. Also played in the past thought not very good chess, but found that some of the chess moves are better.
So, is it human? It’s not a sign that the algorithm is good or bad. For example, face recognition systems, now we have no idea what kind of features the machine has learned through such a lot of learning, can do better than others. This point, has been beyond our understanding of the category.
Where are the opportunities for humanity? Human intelligence, in addition to algorithms, our brains have an algorithm, there is data learning, we have logical reasoning. Our algorithms and models are something we can design ourselves, rather than machines. A very important feature is that our data is collected actively, not passively like current machine learning algorithms, which learn what data you give it.
We humans also have some very interesting features, like our visual intelligence, and sometimes our mistakes are a very important part of our intelligence. In the picture on the left, you can see, is this block the same brightness as this block, or is it different? I don’t think anyone could really tell that the brightness of the two blocks is the same. If you think the two pieces are indeed the same color, I believe there may be something wrong with your brain and you need to see a doctor.
The color of the top block is exactly the same as the color of the top block, but none of us would perceive such a correct result. In fact, we can think of the world as a figment of our own imagination. However, this kind of imagination is difficult to have the current machine, let the machine judge these two questions, it can also be very accurate judgment of these two answers.
We measure AI’s progress rationally, and in many ways we need to pay attention. We’re seeing a lot of progress, but it’s in specific areas, and there’s no sign of a general AI. Perception is changing, but our cognitive abilities haven’t improved much. The so-called perceptual ability is the ability to see, hear and so on. Another point is that the AI in our current stage cannot learn by itself, let alone learn on its own initiative.
This means that the current AI is domain, experience, data dependent, and determined to be domain-specific. Where is GE’s AI army? There are no precise answers, even in academia.
Now is certainly an era of spring and Autumn and warring States, AI applications in various industries will bloom, but the moment of qin unified the country is far from coming, many industries need their own AI engine production capacity.
Giiso information, founded in 2013, is the first domestic high-tech enterprise focusing on the research and development of intelligent information processing technology and the development and operation of core software for writing robots. At the beginning of its establishment, the company received angel round investment, and in August 2015, GSR Venture Capital received $5 million pre-A round of investment.
I also founded a company last year, which we call CiTECH. We have a Chinese and Western name, CIta. We build such a platform and provide such services, providing users and customers from all walks of life with the ability to produce their own AI engines based on private data. We provide engines and enabling capabilities for facial recognition in Huawei phones, including China Mobile, Ping An and some other big customers.
Just to summarize. In the last few years, perceptual advances have driven the whole AI boom. It should be said that traditional industries with AI have a very good opportunity to upgrade, but universal AI is still some way off. So deep learning, in a sense, we think it requires infrastructure. This is also a very important goal for the establishment of Sci-Tech, hoping to move towards a road of AI technology installation, thank you!