Introduction: How to measure technical complexity? By modeling technology as a composite network, a paper proposes a new method of using structural diversity as a quantification of technical complexity. It is based on the improvement of the network diversity score of the complex network measurement, and can calculate the complexity index of different technical networks , And effectively distinguish between complex technology, medium technology and simple technology. The results of the paper prove many of the default facts that are usually related to technological complexity: such as technological complexity grows with time, complex technology requires greater R & D investment and cooperation scale, and high-tech with higher complexity is often more concentrated Space and so on. The title picture is from Visual China, span> This article is from span> WeChat public account: Jizhi Club (ID: swarma_org) span> < / a> , author: ten-dimensional span> p>
p>
p>
Illustration by French artist and writer Albert Robida in the 1892 novel La Vie electrique p>
p>
Technology, this kind of fire that Prometheus deceived from the gods to help mankind in mythology, has now become the dominant force in the development of human civilization. If this is not the case, at least it is the default consensus. Since the Industrial Revolution, modern civilization has liberated the internal forces of matter like desire, stripped energy and information, and extracted, stored, and multiplied. No longer needs magic, humans have sinceTo be able to directly use the power of nature to control nature. p>
p>
The myth disappears and the classical falls. Once the automaton switch is turned on, the modern giant machine begins to take shape. To be or not to be, human beings have no choice, all they can do is to respond to it and how to respond. p>
p>
—— Tool? A medium? The strongest replicon? Or a new subject in the birth? Scholars often have divergent opinions on these essential tendencies, and each has their own opinions, which eventually turns into the deduction of various concepts and the conviction of belief. p>
p>
To resolve conceptual disputes, it is better to do it first, just like the strategies that scientists have adopted for energy, information, life, and consciousness, start with the operation. That is to say, if we can find some indicators of technical measurement, we can continue to approach its essence from the operationally defined way. p>
p>
In this way, we can not only judge the value of a specific technology, but also allow the country or enterprise to make better decisions on whether to invest in certain fields. p>
p>
The question we should really ask is: How do you measure technology? p>
p>
p>
Thesis Title: Using structural diversity to measure the complexity of technologies p>
Thesis address: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0216856 p>
p>
p >
Figure 1: Thesis author Tom Broekel p>
p>
In 2019, a previous paper published in PLoS ONE, “Using structural diversity to measure the complexity of technologies”, proposed a new method of quantification technology: by modeling technology as a component composition network, you can get An indicator of structural diversity that measures the complexity of the technology, that is, the diversity of topologies in these networks and subnetworks. This indicator can not only effectively distinguish between complex technology, medium complex technology and simple technology, but also has been proved by the paper to confirm many of the default facts that are usually related to technical complexity: such as the complexity of technology grows with time, complex technology needs more research and development and Technologies with higher scores on cooperation and structural diversity are also in more concentrated spaces and so on. The author Tom Broekel is a professor at the Business School of Stavanger University in Norway and a member of the Regional and Innovation Economics Center at Bremen University in Germany. His research direction is regional innovation and technological complexity. p>
p>
One, complexity measurement: Zhongli Xunhuaqianbaidu p>
p>
Although it is difficult to question the ontology of technology, most scholars basically agree that technology itself is a hierarchical structure, which comes from a combination of lower-level technologies. For the complex economist Brian Arthur, the most primitive technique is the capture of natural phenomena and their effects [1]. Therefore, the technology is regarded as a kind of complex network, and there is not even excessive restoration. What really needs to be solved is how to effectively measure the network structure. However, this is a difficult problem. p>
p>
Traditionally, similar statistical physics or information theory methods can be used to calculate the entropy complexity of an object or network [2,3] [4]. For example, the following figure shows six typical network structures: (a) star network (b) lattice network (c) tree network (d) small world network (e) fully connected network (f) random network. p>
p>
p>
Figure 2: Six typical networks p>
p>
The information needed to describe the star network, lattice network, and fully connected network is very small. They are all simple networks. (c) The tree network is slightly more complicated, but only requires limited information. The most important thing is the comparison between (d) small-world networks and (f) random networks. Although small-world networks are typical complex networks, if calculated according to information theory methods, you will find (f) information contained in random networks The volume is even higher than the small world network. p>
p>
Therefore, for traditional complexity measurement, whether it is a typical theoretical network or an empirical network such as a biological network, the most difficult problem to solve is how to distinguish a complex network from a random network and an ordered network . That is, an effective complexity measure distribution should be shown in Figure 2A instead of B: p>
p>
p>
Figure 3: Theoretical and practical distribution of effective indicators of complexity measurement p>
p>
In the study of the actual empirical network, it is mainly based on chemical graph theory (Chemical Graph Theory) span> , And the various properties of ecological networks, scientists have sorted out 16 kinds of network complexity metrics: p>
p>
p>
Figure 4: Sixteen typical complex network metrics p>
p>
However, Frank Emmert-Streib and Matthias Dehmer, two bioinformaticians from the United Kingdom and Austria, after verifying and analyzing these 16 complexity indicators [5], found that they can hardly distinguish between complex networks and Random network. p>
p>
p>
Figure 5: The performance of sixteen complex measurement indicators to distinguish ability under numerical test p>
p>
The results are shown in Figure 5, showing the relationship between the probability density of the complexity value (y-axis) and the complexity value (x-axis). Most indicators are mixed with the distribution of ordered, complex and random networks, except that the non-diagonal complexity can be basically differentiated. But even it has a big problem, not only there are two distribution peaks, but also there is considerable overlap in the complex network (purple) and random network (red). p>
p>
h2>
Second, network diversity score: the most beautiful one among complexity and randomness p>
p>
A complex network can be regarded as a high-dimensional object. I am afraid that a single-dimensional index alone cannot effectively resolve and measure complex networks. What about comprehensive multi-dimensional indexes? p>
p>
Thesis title: Exploring Statistical and Population Aspects of Network Complexity p>
p>
Thesis address: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0034523 p>
blockquote>p>
Along this line of thought, the same two bioinformaticians published a paper in 2012: “Exploring statistical and population aspects of network complexity”, which proposed network diversity indicators (Network diversity score, NDS) span> effectively solves this problem. Their methods are as follows: p>
p>
Let G denote a network with vertex set V and edge set E. The number of vertices is n = | V |, and the number of edges is e = | E |. The indicator is calculated based on the following four variables: p>
p>
p>
” p>
where M is the number of modules in network G. The vector m = (m1, m2 …) is the size of the included module, arranged according to the number of nodes of the corresponding module, the size of the i-th module is mi. In order to identify the modules in the network, a method called Walktrap [6] similar to random walk [7, 8] can be used to find the modules. Its advantage is that it can directly estimate the computational complexity O (e * n ^ 2) . p>
p>
The four variables in equations (30)-(33) respectively reflect different attribute information in the network structure: p>
p>
Variable αmodule: information about the density of network modules. Module is the basic expression of the general organization principle of the network,For complex networks, people always want to find more modules than random networks. p> li> ul>
p>
Variable vmodule: similar to CV (Coefficient of variation) span> value, which measures the variability of the network size relative to the average size of the module. The modules of the random network are large, but the variability is low, and the average module size is also low. However, the large variability of the modules of the complex network is high, and the average module size is also high. p> li> ul>
p>
Variable vλ: Similar to vmodule, but its parameter is Laplace matrix (Laplace matrix) span> [9]. p> li> ul>
p>
Variable rmotif: About network motifs (Network motifs) span> growth rate information. It can be observed from the statistics that the ordered network has the highest rmotif value, the complex network is centered, and the random network is the lowest. p> li> ul>
blockquote>p>
The network motif in (33) variable (Network motifs) span> is an important local feature in the network, it is defined Subgraphs or patterns that are repetitive and can be effectively counted can be regarded as the primitives forming a complex network [14]. It was first widely researched and applied in biological networks such as gene networks. It differs from general subgraphs in that the roles between nodes may have asymmetry. As shown in Figure 5 a and b. p>
p>
where Nmotif (3) and Nmotif (4) correspond to those found in network GThe number of phantoms with sizes 3 and 4 also lists all 13 different three-node phantom structures in Figure 5c. Starting from the four nodes, the total number of phantom types is increasing. The total number of motifs (4) is 199, and the five and six nodes are 9364 and 1,530,843 respectively. p>
p>
p>
Figure 6: (a). Ordinary three-node subgraph (b). Three-node phantom of two node roles (c). All 13 kinds of three Node network motif p>
p>
Based on the above four variables, a single network diversity index can be defined for network G (Individual diversity score, iNDS) span>: p>
p>
p>
p>
p>
For Gm that contains a set of networks, that is, G∈Gm, the overall network diversity index is: p>
p>
p>
p>
where gM represents the total number of networks belonging to the same network model, and Pgm is the probability distribution density of the model as a whole. For example, it may correspond to the random network generated by Erdos-Reyni model [11,12], or it may use the preferential attachment algorithm (Preferential attachment algorithm) span> The generated set of all scale-free networks [13,14]. p>
p>
p>
Figure 7: The effect of network diversity indicators under different samples and node numbers. When the number of samples is S = 50, the three networks are completely separated p>
p>
It can be seen that the network diversity index is based on a combination of multiple network structure principles and combines the advantages of multiple complexity measures for evaluation. The researchers also demonstrated that among the four variable combinations, a single density value calculated by equation (35) can optimally separate random, complex, and ordered networks. The effect is shown in Figure 6. The larger the number of network nodes and the total number of samples, the better the separation effect. When the sample S = 5, the effect of the network diversity index has far exceeded the best-performing off-diagonal complexity of the sixteen complexity indexes. p>
p>
h2>
Three, structural diversity indicators: the source code of technology p>
p>
As mentioned above, as a complex network metric, only the network diversity score can consistently separate ordered, complex, and random network structures. p>
p>
Complex networks represent a mixture of ordered and random structures. Compared with ordered networks, it is more heterogeneous in topology. But in contrast, the one with the most heterogeneous topology is the random network. p>
p>
In the study of network diversity indicators, based on scientific numbersThe value experiment generates a large number of random networks, which makes the calculation of random samples of the network very large, so it may ignore the weight of some real network experience evaluation. For example, in a technology network or a knowledge network, unlike a biological network, although a network formed by random combinations exists, in the final analysis, technological development has a cumulative characteristic, and the relevant social processes behind it will always ensure the existence of the system structure. p>
p>
Therefore, in order to measure the technical complexity more effectively, Tom Broekel revised the network diversity index. Because the motifs rmotif (3) and rmotif (4) of size three and four are the highest in the ordered network, medium in the complex network, and the lowest in the random network [15]-however, all randomness in the network diversity index The number of network samples is very large [16]. p>
p>
The specific method is to modify the variable rmotif, and adjust the three and four motifs motif (3) and rmotif (4) to the number of graphs with the number of nodes three and four. That is, the ratio between the network structures that meet the experience is: p>
p>
p>
p>
p>
So the complexity of a single technical network is: p>
p>
p>
p>
Sample random samples of S technical networks and estimate the average iNDS for each sample network to get the value of its network diversity index:
p>p>
p>
p>
As a result, large values indicate random networks (complex technology) span>, and median values indicate complex networks (medium complex technology) span>, and low values indicate an ordered network (simple technology) span>. p>
p>
To use the above structural diversity indicators, Tom Broekel used the REGPAT database (OECD) span> “text-remarks” label = “Remarks”> (2018 version) span>, the database covers patent applications filed with the European Patent Office. Due to the time difference between the priority date and the acquisition of patent information, the analysis was limited to 1980 to 2015 and included information on 3,137,881 patent applications. Technology is defined according to the company’s patent network. The company’s patent network is divided into 9 categories at the highest level and more than 230,300 subcategories at the lowest level. In order to provide a good trade-off between technology decomposition and the number of manageable technologies, researchers use a four-digit company patent network to define 655 different technologies for acting [17,18]. p>
p>
h2>
Four, empirical research on the quantification of technical complexity p>
p>
Analysis of the diverse structure of patents confirms four typical facts of technological complexity: technological complexity grows over time, and more complex technologies involve more Research and development, and require more collaboration, complex technology tends to more concentrated space. span> p>
h3>
Technical complexity grows with time p>
Figure 8 shows the distribution of 655 technical structural diversity from 1980 to 2015. p>
p>
p>
Figure 8: 1980-2015, distribution density of technical structure complexity p>
p>
The horizontal axis represents the complexity value of structural diversity. The minimum value is zero and the maximum value is 14.98. The vertical axis between the two is its probability density, showing a bimodal distribution, with one peak at zero and the maximum density peak in the middle. Zero peaks reflect technologies that have no or too few patents to calculate structural diversity. It can be seen that after excluding zero peaks, no matter what year, the overall appearance is a bell curve with a normal distribution, but as the number of years increases, the left side becomes shorter and the right side becomes longer and longer, which shows the technical complexity Is increasing. p>
p>
p>
Figure 9: 1980-2015, the change of the median box line of the complexity of the technical structure p>
p>
The boxplot in Figure 9 directly shows the median change in the structural diversity of these technologies over time: from less than 5 in 1980 to an average of about 10 in 2010. This is consistent with the trend of technological development in most people’s impressions, and it also proves that the complexity of technology is increasing. p>
p>
Due to the cumulative nature of knowledge and technological development, each generation of new technology isThe predecessors [19-22] developed on the basis of the technical environment established. As the adjacent possible space expands, new technologies are possible. For example, the aircraft engine with digital control system uses more technology than the previous hydraulic mechanical engine [23]. The Microsoft Windows system code has grown from more than 3 million in 3.1 to more than 40 million lines in Vista [24]. p>
p>
But it is worth noting that the variance of structural diversity is still large: not only is the lowest value in 2015 much lower than the median value in 1980, the technology that reached the maximum value in the early 1980s is still greater than the median in recent years The highest value of the number. In other words, although the average complexity of technology is increasing, each era has its own complex technology. Technology growth is not as fast as imagined, as you can see from the figure is almost linear growth. p>
span> p>Complex technology requires greater R & D investment span> p>
span> p>Technical progress is achieved by searching for potential components in the technology space and testing technology combinations to create new knowledge combinations, which usually require trial and error to complete [26] [30]. p>
p>
Compared with simple technologies, complex technologies are based on greater knowledge diversity and less common knowledge combinations, which further increases the difficulty of research and development. Because learning complex knowledge itself requires more resources and greater absorptive capacity [27] [31]. It takes longer development time to translate these characteristics of complex technology into complex products [28] [29]. p>
p>
p>
Figure 10: 1980-2015, the correlation between technical complexity and the number of technical patents p>
p>
Figure 10 shows the correlation between technology complexity and the number of technology patents from 1980 to 2015. The correlation coefficient is very obvious at all time periods and increases from r = 0.45 to r = 0.69. In terms of R & D intensity reflected in the number of patents, the results confirm the positive correlation between technological complexity and R & D intensity. p>
p>
In addition, the increasing correlation coefficient shows that this relationship strengthens over time-that is, reaching a higher degree of complexity is more dependent on R & D investment than in the past. This potential trend reflects the hypothesis of diminishing return on investment: as a result of diminishing returns on R & D investment over time, innovation will be increasingly distributed on more products [32]. p>
h3>
p>
Complex technology needs more cooperation p>
span> p>Another characteristic that complex technologies usually have is that they have greater requirements for R & D collaboration [33,34,22] [36], and work together to solve the problems faced Complicated issues [26]. span>
p>p>
p>
Figure 11: Correlation between the diversity of technological structure and the average number of patents of patents from 1980 to 2015 p>
p>
Figure 11 shows the ranking correlation between the diversity of technical structures and the average number of inventions per patent, which indicates the degree of patent-based teamwork. From these years, the significantly positive correlation coefficient provesThe importance of collaboration in more complex technologies. After reaching about 0.4 in the late 1980s, the correlation coefficient basically fluctuated at this level. After 2000, the correlation coefficient has declined, which may indicate that for software technology, fewer developers may also create complex technologies. P>
p>
Complex technology needs more concentrated space
p>p>
In the study of regional economic geography, it is generally believed that the development of complex technologies requires special skills, expertise, infrastructure or R & D institutions that are rare in other regions [37-39] [26]. p>
p>
In the learning domain area (Learning regions) span>, innovation environment (Innovative milieu) span>, regional innovation system (Regional innovation systems) span> [40-42] [35,34]. p>
p>
In order to explore the relationship between structural diversity and the spatial distribution of technology, the study uses information on the location of patent inventors, and estimates the Gini coefficient for the technology-specific regional distribution of the number of patents (Gini coefficient) span>, also targeting the European NUTS2 region, including 270 regions and 1,557,416 patents. p>
p>
If the inventors are concentrated in a few areas, the coefficient will be close to 1, otherwise the coefficient will converge to 0 when they are evenly distributed in space. Figure 8 shows the correlation between the spatial Gini coefficient of the technology and its structural diversity value. Since technologies with a smaller number of patents may be distributed equally among regions, technologies with at least 1,000 patents are also discounted in red to show their relevance. p>
p>
p>
Figure 12: 1980-2015, the correlation between the diversity of technology and complex technical structures and the regional Gini coefficient p>
p>
When considering all technologies (blue error bars) span>, the correlation has a strong negative significance, indicating that complex technologies are more complex than simple technologies The distribution is more even. However, when focusing on the technology with many patents (red error bar) span>, the situation will change. Due to the small number of technologies with more than 1,000 patents, the correlation has been negligible until the mid-1990s. Since that year, the correlation has fluctuated between positive and insignificant. It is worth noting that this coefficient has been increasing since 2009 and has maintained a significant level since 2012, indicating that larger and more complex technologies are indeed increasingly concentrated in the space. p>
p>
Five, summary: significance for research on technical complexity p>
p>
Once an effective indicator of technical measurement is found, the problems that previously caused confusion or controversy, whether it is a detailed inspection of a certain technology, or a prediction of the overall trend of technology, or an analysis of the impact on the social economy , There is a basis for further discussion. p>
h3>
p>
Analyze the technical potential through structural complexity p>
span> p>The network structure between innovative technology components is highly ordered and does not have much redundancy, especially because of its new knowledge and R & D investment It not only has no node or structure information, but only meaningless redundancy or noise. span> p>
p>
One of the most direct effects of using structural complexity and network diversity indicators is that you can accurately evaluate a technical product,Or does a certain network have truly effective complexity, or is it just a seemingly complex random network or artificial network. p>
p>
For example, for 5G technology, patents or the number of network nodes alone are not sufficient to explain the pros and cons of a certain technical solution. Using network diversity indicators to examine the complexity of its network structure, combined with cost analysis, before investing in or developing a technology, it may be possible to predict in advance whether a technical solution has sufficient innovation and development potential. p>
p>
p>
Figure 13: Overall Architecture of 5G Bearer Network Technology p>
p>
In addition, the mining of network diversity indicators can be extended to any high-dimensional object, such as an artwork or even an industrial ecological network. Through proper network modeling, it can be identified whether these objects really have a complex structure, or whether they are only charged through redundancy, or with the help of random noise, which is unintelligible and meaningless “flicker”. Of course, the analysis also applies to biological species, whether it is a gene regulation network or a brain neural network. p>
p>
h3>
Inspect the impact of technology on social economy p>
span> p>Complex technology tends to diminish economically, that is, for high-tech R & D, it often has a lower input-output ratio. span> p>
span> p>At the national level, studies have pointed out [21,43] that due to complexity and increased development diversity, investment in high-intensity R & D will exceed economic output, and The average income will be affected by factors such as difficulties in learning new skills. This is reflected in the proportionality between the particularly complex high-tech and the Gini coefficient of the region (Figure 11) span>. p>
p>
From this point, we can understand why the United States has been cutting research funding in recent years. NASA has closed some aviation projects to private space companies such as SpaceX and Blue Origin. In addition to engineering, in addition to cost, technical complexity also depends on the management of technical systems [44]. p>
p>
It can be seen that whether it is scientific research, technology or products, the “low-hanging fruit” will always be picked up quickly, and high-complexity innovations need to face the risk of failure and the return on investment. p>
p>
For national or enterprise R & D, the most important thing is to think clearly about where you are and where you want to go. If you want to achieve basic food and clothing, you still need to develop with creativity. If it is the former, in the early stages of development, you can cut off technical projects with too high complexity, give priority to the development of medium-complex technologies, develop business first, and then engage in technological innovation. Lenovo President Liu Chuanzhi once suggested that the path of “trade, industry and technology” should be followed first, followed by the path of “technology, industry and trade”, which is in line with the development path of most early Chinese technology companies. p>
p>
Quantitative research on the overall trend of technology: intelligence and technological singularities p>
p>
As you can see from Figures 7 and 8, although the complexity of the technology has indeed increased over time, it is far from being as fast as expected, and it is only a linear proportional increase overall. This conclusion is similar to the result of a paper published in the journal Nature Human Behaviour in January 2020, which did a quantitative comparative study of the evolutionary rates of nature and culture. p>
p>
And the technology is not always more powerful and complex in the future. At an earlier time, there are similar products that are much higher than the current technical level, such as the Apollo Moon Project, which produced more than 3,000 patents at the time. More than 1,000 were later converted to civilian use, and the total number of collaborators reached an astonishing 300,000 people. The “Saturn 5” it produced is still the most powerful rocket propeller in human history. (compared to SpaceX’s” Falcon “[46]) span>. p>
p>
p>
Figure 14: Apollo ’s “Saturn 5” and SpaceX’s “Falcon” on the moon p>
p>
However, the computing power of Saturn 5’s “hardware programming” operating system is not as good as an ordinary smartphone. p>
p>
But what is interesting is that people argue that the main basis for the accelerated development of technology comes from the improvement of the computing power of the system. As indicated by Moore’s Law, if machine computing power increases exponentially, artificial intelligence singularity will soon be reached in a few decades. This view is like saying that a “fast thinking dog” will also play chess. p>
p>
Calculation speed is only one aspect of intelligence. It does not necessarily mean that the system has higher complexity. p>
p>
In the field of artificial intelligence, with the current mainstream machine learning, especially neural network technology, it is generally believed that the current breakthrough in intelligent system bottlenecks is mainly due to the improvement of computing power and the collection of massive data. p>
p>
p>
Figure 15: Trend schedule of AI surpassing human computing power in Futurist Ray Kurzweil’s “Singularity Approaching” p>
p>
One of the biggest characteristics of man-made technology is that it can improve the system capacity almost infinitely with the proportional input of energy and information, but it does not have higher complexity. For an AI system that does not have higher structural complexity, its intelligence level, although there are different machine intelligence judgment standards-the general intelligence level, is inherently doubtful. p>
p>
If you compare life, it is the number of biological genomes and the complexity of gene regulation networks. From the perspective of the overall evolutionary tree, from mycoplasma and bacteria to birds, mammals and amphibians, biological recoveryMiscellaneousness is the that grows with the size of the genome (as shown in the figure below, plants belong to different evolutionary trees) span>. Obviously, the number of genomes is relatively small, and mycoplasma or bacteria are not, and it is not likely to produce general human-like intelligence. p>
p>
However, the relationship between genome size and the complexity of organisms only grows linearly in prokaryotes. In eukaryotes, we can see that crustaceans (Crustaceans) span> are almost the same order of magnitude as humans, and amphibians reach 10 ^ 10 bp (number of base pairs) span>. This evolutionary anomaly was once referred to as the C-value paradox (C-VALUE PARADOX) span> [47]. p>
p>
p>
Figure 16 : Comparison of the number of genomes of different types of organisms, the unit is the number of base pairs span> p>
p>
On the other hand, this year ’s new research shows that at least the animals of the Houkou animal door (Deuterostomia, including vertebrates, mammals) span> are genetically Overall, it has been “subjecting”: keep new mutations with fewer genes, and discard a lot of useless or even core genes [48]. p>
p>
p>
Figure 17: Starting from two symmetrical animals (Bilateria), the changes in gene gains and losses p>
p>
It can be seen that quantity is not a sufficient condition for complexity, but a necessary condition. Behind the number, what really plays a key role is the adjustment and optimization of the gene regulation network for the environment, that is, gene function. The more complex the organism, the more genetic functions it has, and the more it adapts to the external environment. p>
p>
Whether it is the simple eyes of the protozoan green eyeworm, the Hydra reticular nervous system, or the developed and plastic central nervous system of birds and mammals, and even the learning ability, language ability and cultural inheritance ability of primates (For example, FOXP2 is important for human language ability and MEF2A for children ’s learning plasticity [49]) span>, all in line with this law. p>
p>
p>
Figure 18: The main events of the evolution of life complexity p>
p>
One of the most typical examples is the birth of sexual reproduction. Matt Ridley argued in “The Cooperative Gene” (Chinese translation of “Mendel ’s Demon”) span>, sexual reproduction has brought a powerful power Error correction ability, thus raising the upper limit of complexity that the living body can carry. In addition, there are studies in theoretical biology that prove that sexual reproduction can increase the entropy and increase the accumulation of suitableThe balance between appetite is equivalent to a MWUA algorithm that can achieve the optimal trade-off between diversity and long-term returns [50]. In addition, homologous genes (Hox genes) span> [51] and so on, which greatly increase the biological complexity. p>
p>
It can be seen that organisms constantly optimize the complexity of the structural diversity of their genes under natural selection. To summarize briefly, the complexity of intelligent systems ≈ the complexity of its own structure + the input of environmental energy + the complexity of information interaction, in which the environment and information parts are the ability of organisms to adapt and change the environment, including learning ability, collaboration ability, problem solving and computing ability. After reaching a new complexity bottleneck, it is inevitable that humans will use language, tools and technology. p>
p>
So, from the perspective of its own structural diversity complexity and network diversity indicators, the current neural network is too dependent on the acquired computing power and data, there is “as much manpower as there is intelligence”, and highly dependent on cooperation with human . However, he lacks the background knowledge of the real world and the autonomous ability to interact with the world. So even if it has the ability to surpass and can simulate the calculation of the human brain, it is still just a powerful computing device, probably like an echinoderm with a gene number of 10 ^ 9 bp. p>
p>
Therefore, the Turing Award winner Judea Pearl only advocated that the new breakthrough of artificial intelligence lies in the “cause and effect revolution”, which enables AI to have causal induction and causal reasoning capabilities; on the other hand, neuromorphic chips and brain-like chip hardware research, robots The research of the technology also allows the artificial intelligence system to further greatly improve its own structural complexity. (Recommended reading: span> Where does the arrow of cause and effect point? | Turing Award winner Pearl ’s” Why “ span> a> ) span> p>
p>
At present, the indicators of structural diversity and network diversity can only allow us to calculate the initial complexity of a system, but this problem is still not resolved: that is, how does the complexity of the system itself interact with the external environment, how is the final decision made? A systematicComplexity and intelligence? p>
p>
p>
Picture 19: “Prometheus Bound” on the Caucasus Mountains, an oil painting created by American artist Thomas Cole in 1847 p>
p>
This may include many issues such as environmental perception, multiplication methods, multi-body interaction, social collaboration, language and culture. Although there is no definite answer for the time being, it is still in a more complicated mist, but with the help of various quantitative tools In practice, human intelligent creatures will continue to explore. One day, they will approach to see the essence of life and technology-just like Prometheus who was bound by the fire stolen in ancient Greek mythology. , And the father of all knowledge and technology, his foresight (Prometheus) span> will eventually be liberated by humans. p>
p>
References: span> p>
[1] W. B. Arthur, The Nature of Technology: What It Is and How It Evolves (Penguin, London, 2010) span> p>
[2] Wiener H. Structural determination of paraffin boiling points. Journal of the American Chemical Society. 1947; 69 (1): 17–20. span> p>
https://doi.org/10.1021 / ja01193a005 PMID: 20291038 span> p>
[3] Shannon CE. A mathematical theory of communication. The Bell System Technical Journal. 1948; 27 (July 1928): 379–423. span> p >
https://doi.org/10.1002/j.1538-7305.1948.tb01338.x span> p>
[4] Dehmer M, Barbarini N, Varmuza K, Graber A. A large scale analysis of information-theoretic network complexity measures using chemical structures. PLoS ONE. 2009; 4 ( 12): 20–26. Span> p>
https://doi.org/10.1371/journal.pone.0008057 span> p>
[5] Emmert-Streib F, Dehmer M. Exploring statistical and population aspects of network complexity. PLoS ONE. 2012; 7 (5). https://doi.org /10.1371/journal.pone.0034523 p>
[6] Pons P, LatapyM (2005) Computing communities in large networks using random walks. In: Yolum p, Gu ¨ngo ¨r T, Gu ¨rgen F, O ¨zturan C, editors, Computer and Information Sciences-ISCIS 2005, Springer Berlin / Heidelberg, volume 3733 of Lecture Notes in Computer Science. pp 284–293. span> p>
[7] Van Dongen S (2000) Graph clustering by flow simulation. Ph.D. thesis, Centers for mathematics and computer science (CWI), University of Utrecht. span> p>
[8] Ziv E, Middendorf M, Wiggins CH (2005) Information-theoretic approach to network modularity. Phys Rev E 71: 046117. span> p>
[9] Chung FRK (1997) Spectral Graph Theory. American Mathematical Society. span> p>
[10] Milo R, Shen-Orr S, Itzkovitz S, Kashtan N, Chklovskii D, et al. (2002) Network motifs: simple building blocks of complex networks. Science 298: 824–7. Span> p>
[11] Erdo ¨s P, Re ´nyi A (1959) On random graphs. I. Publicationes Mathematicae 6: 290–297. span> p>
[12] Gilbert EN (1959) Random graphs. Annals of Mathematical Statistics 20: 1141–1144. span> p>
[13] Albert R, Barabasi A (2002) Statistical mechanics of complex networks. Rev of Modern Physics 74: 47. span> p>
[14] Baraba ´si AL, Albert R (1999) Emergence of scaling in random networks. Science 206: 509–512. span> p>
[15] Emmert-Streib F, Dehmer M. Exploring statistical and population aspects of network complexity. PLoS ONE. 2012; 7 (5). span> p >
https://doi.org/10.1371/journal.pone.0034523 span> p>
[16] Milo R, Stumpf MP, Stark J, Milo R, Shen-Orr S, Itzkovitz S, et al. Network motifs: Simple building blocks of complex networks. Science . 2002; 298 (5594): 824–827. Span> p>
https://doi.org/10.1126/science.298.5594.824PMID: 12399590 span> p>
[17] Schmoch U, Laville F, Patel P, Frietsch R. Linking technology areas to industrial sectors. Final Report to the European Commission, DG Research, Karlsruhe, Paris, Brighton. 2003;. span> p>
[18] Breschi S, Lenzi C. Net City: How co-invention networks shape inventive productivity in US cities. KITeS Seminarpapers. 2011; p. 1–32. span> p>
[19] Nelson RR, Winter SG. The Schumpeterian tradeoff revisited. American Economic Review. 1982; 72 (1): 114–132. span> p>
[20] Howitt P. Steady endogenous growth with population and R. & D. inputs growing. Journal of Political Economy. 1999; 107 (4): 715–730. https : //doi.org/10.1086/250076 span> p>
[21] Aunger R. Types of technology. Technological Forecasting and Social Change. 2010; 77 (5): 762–782.https: //doi.org/10.1016/ j.techfore.2010.01.008 span> p>
[22] Hidalgo CA. Why information grows: The evolution of order, from atoms to economies. New York: Basic Books; 2015. span> p>
[23] Prencipe A. Breadth and depth of technological capabilities in CoPS: the case of the aircraft engine con- trol system. Research Policy. 2000; 29: 895–911. https://doi.org/10.1016/S0048-7333 (00) 00111-6 span> p>
[24] Wikipedia. Source lines of code; 2017.Available from: https://en.wikipedia.org/wiki/Source_lines_of_ codes. span> p >
[25] Fai F, Von Tunzelmann N. Industry-specific competencies and converging technological systems: Evi- dence from patents. Structural Change and Economic Dynamics. 2001; 12 (2) : 141–170. Span> p>
https://doi.org/10.1016/S0954-349X (00) 00035-7 span> p>
[26] Carbonell P, Rodriguez AI. Designing teams for speedy product development: The moderating effect of technological complexity. Journal of Business Research. 2006; 59 (2): 225– 232. Span> p>
https://doi.org/10.1016/ j.jbusres.2005.08.002 span> p>
[27] Cohen WM, Levinthal DA. Absorptive capacity: a new perspective on learning and innovation. Adminis- trative Science Quarterly. 1990; 35 (1): 128–152. https://doi.org/10.2307/2393553 span> p>
[28] Griffin A. The effect of project and process characteristics on product development cycle Ttme. Journal of Marketing Research. 1997; 34 (1): 24–35. https: //doi.org/10.2307/3152062 p>
[29] Singh K. The impact of technological complexity and interfirm cooperation on business survival. Acad- emy of Management Journal. 1997; 40 (2): 339–367.https : //doi.org/10.5465/256886 span> p>
[30] Hargadon A. How breakthroughs happen—The surprising truth about how companies innovate. Bos- ton: Harvard Businesss School Press; 2003. span> p>
[31] Pintea M, Thompson P. Technological complexity and economic growth. Review of Economic Dynam-ics. 2007; 10 (2): 276–293. span> p>
https://doi.org/10.1016/j.red.2006.12.001 span> p>
[32] Madsen JB. Are there diminishing returns to R & D? Economics Letters. 2007; 95 (2): 161–166. https: // doi.org/10.1016/ j.econlet.2006.09.009 span> p>
[33] Hidalgo A, Hausmann R. The building blocks of economic complexity. PNAS. 2009; 106 (26): 10570–10575. span> p>
https://doi.org/10.1073/pnas.0900943106 PMID: 19549871 span> p>
[34] Balland PA, Rigby D. The geography of complex knowledge. Economic Geography. 2017; 93 (1): 1–23. span> p>
https://doi.org/10.1080/00130095.2016.1205947 span> p>
[35] Sorenson O. Social networks, informational complexity and industrial geography. In: Fornahl D, Zellner C, Audretsch DB, editors. The role of labour mobility and informal networks for knowledge transfer. Boston: Springer Science + Business Media Inc .; 2005. p. 79–96. span> p>
[36] Pavitt K. Technologies, products and organization in the innovating firm: what Adam Smith tells us and Joseph Schumpeter doesn’t. Industrial and Corporate Change. 1998; 7: 433–452. Https://doi.org/10.1093/icc/7.3.433 span> p>
[37] Jaffe AB. Characterizing the “technological position” of firms, with application to quantifying technologi- cal opportunity and research spillovers. Research Policy. 1989; 18 (2): 87–97. Span> p>
https://doi.org/10.1016/0048-7333 (89) 90007-3 span> p>
[38] Audretsch DB, Feldman M. R & D spillovers and the geography of innovation and production. American Economic Review. 1996; 86 (4): 253–273. span > p>
[39] Almeida P. Knowledge sourcing by foreign multinationals: Patent citation analysis in the U.S. semicon- ductor industry. Strategic Management Journal. 1996; 17 (S2): 155–165. https://doi.org/10.1002/smj.4250171113 span> p>
[40] Florida R. Toward the learning region. Futures. 1995; 27: 527–536. https://doi.org/10.1016/0016-3287 (95) 00021-N span> p>
[41] Camagni R. Local “milieu”, uncertainty and innovation networks: towards a new dynamic theory of eco- nomic space. In: Camagni R, editor. Innovation Networks: Spatial Perspectives. Belhaven Stress. Lon- don, UK and New York, USA; 1991. p. 121–142. Span> p>
[42] Cooke P. Regional innovation systems: Competitive regulation in the new Europe. GeoForum. 1992; 23: 356–382. https://doi.org/10.1016/ 0016-7185 (92) 90048-9 span> p>
[43] Kim BW. Economic growth: Education vs. research. Journal of Global Economics. 2015; 03 (04). span> p>
[44] Mcnerney J, Farmer JD, Redner S, Trancik JE. Role of design complexity in technology improvement. PNAS. 2011; 108 (22): 9008–9013. span> p>
https://doi.org/10.1073/pnas.1017298108PMID: 21576499 span> p>
[45] Ben Lambert, Georgios Kontonatsios, Matthias Mauch, Theodore Kokkoris, Matthew Jockers, Sophia Ananiadou & Armand M. Leroi. The pace of modern culture. 2020.https: / /www.nature.com/articles/s41562-019-0802-4 p>
[46] David Szondy. Falcon Heavy vs. the classic Saturn V. 2018. https://newatlas.com/falcon-heavy-saturn-v/53090/ span> p>
[47] Gordon P. Moore, The C-Value Paradox BioScience, Volume 34, Issue 7, 1 August 1984, Pages 425–429. https://doi.org /10.2307/1309631 p>
[48] Cristina Guijarro-Clarke, Peter W. H. Holland & Jordi Paps. Widespread patterns of gene loss in the evolution of the animal kingdom. 4, 519–523 (2020). Https://www.nature.com / articles / s41559-020-1129-2 span> p>
[49] Liu X, Somel M, Tang L, Yan Z, Jiang X, Guo S, Yuan Y, He L, Oleksiak A, Zhang Y, Li N, Hu Y, Chen W, Qiu Z, Pääbo S, Khaitovich P. Extension of cortical synaptic development distinguishes humans from chimpanzees and macaques. 6. Genome Res. 2012 Apr; 22 (4): 611-22. Span> p >
[50] Erick Chastain, Adi Livnat, Christos Papadimitriou, Umesh Vazirani, Algorithms, games, and evolution, Proceedings of the National Academy of Sciences Jul 2014, 111 (29) 10620 -10623; DOI: 10.1073 / pnas.1406556111 span> p>
[51] CHAPTER 20, Evolution 3rd Edition, MARK RIDLEY, 2004 span> p>
p>
This article is from span> WeChat public account: Jizhi Club (ID: swarma_org) span> , Author: Three ten span> p>