Since the beginning of the 21st century, with the development of computer technology and the rapid iteration of hardware, human society has successively entered the Internet era, the big data era, and the artificial intelligence era. Especially after entering the second decade of the 21st century, with the resurgence of machine learning, tech giants have made artificial intelligence a strategic pivot. According to Deloitte Consulting, it is estimated that the global artificial intelligence market will exceed $6 trillion by 2025, with a compound annual growth rate of 30% from 2017 to 2025. Meanwhile, the Chinese government has also introduced several policies to promote the development of the artificial intelligence industry, elevating it to a strategic level and striving to seize the high ground of artificial intelligence.
Looking back at the history of artificial intelligence, we can see that it has been divided into several branches since its inception, with the two most influential branches being symbolism and connectionism. Symbolism posits that the human brain is essentially a system of symbolic reasoning, and how to summarize and represent known human knowledge in a form understandable by machines for reasoning and computation has always been a focus of symbolism research.In its early days, symbolism was the mainstream school of artificial intelligence, with early expert systems and today’s knowledge graphs being typical representatives of symbolism.
Connectionism, on the other hand, holds that intelligence is the result of signal transmission between neurons in the human brain, and simulating the brain’s structure to build models has become the main method, with early perceptrons and today’s machine learning being the most typical representatives. After the 1980s, constrained by theoretical and technical limitations, connectionism entered a low period; however, in recent years, the resurgence of machine learning has made connectionism the mainstream in the field of artificial intelligence once again.
Knowledge graphs and machine learning, as the two most cutting-edge hotspots of symbolism and connectionism, have attracted significant attention and research, with machine learning being particularly hot. A search on CNKI using the keywords “knowledge graph” and “machine learning” shows that in 2019, there were 28,304 papers related to machine learning, while only 2,624 papers related to knowledge graphs.Why is the current heat of machine learning far greater than that of knowledge graphs? The possible reasons mainly include the following two aspects:
Firstly, the current construction of knowledge graphs is closely related to natural language processing, and using machine learning to solve natural language processing problems is also a very important approach at this stage. Therefore, constructing knowledge graphs using machine learning has become mainstream.
Secondly, machine learning algorithm models and the hardware devices required for them have gradually matured in recent years, while the star effect of AlphaGo has led to an explosive development period for machine learning.Does this mean that machine learning is about to unify the field of artificial intelligence?The answer is no.
1. Machine learning is one of the tools to assist in the construction of knowledge graphs, but it is not the only one.
From the historical development of knowledge graphs, they can be traced back to semantic networks, which are a form of knowledge representation proposed by early symbolism scholars. They later underwent fusion and extension with theories such as ontology and the World Wide Web, proposed by Google in 2012, when machine learning had not yet emerged. Google’s knowledge graph primarily sourced data from the open-source community. At the same time, Google established Schema.org to clearly define the semantic model, allowing users to embed semantic data in their websites using Google’s Schema, thus improving their search results in search engines, and this data can also be easily collected and organized by Google to form the graph.
It can be seen that the construction of knowledge graphs includes processes such as Schema construction and knowledge extraction, while machine learning is merely one method used for knowledge extraction. However, even using machine learning to solve natural language processing problems to assist in knowledge extraction cannot solely rely on a set of algorithm models from machines to solve everything. Designing a model to feed it a large amount of corpus and then output a knowledge graph is certainly impossible. This is because knowledge extraction is a task that faces data directly, and some tedious and dirty work is unavoidable, as Liu Cixin, the author of The Three-Body Problem, said, “The amount of intelligence in artificial intelligence is proportional to the amount of intelligence behind it.”
In the process of constructing knowledge graphs, practitioners should think and design better knowledge representation Schema model structures, while fully utilizing the collaboration between machines and humans, so as to build knowledge graphs better and faster.
2. Machine learning and knowledge graphs are two core elements of artificial intelligence, both of which are indispensable.
Machine learning, especially the algorithms and models of neural network learning, has elevated the perception capabilities of machines to a new level, while rapidly updating hardware has given machines very powerful computing capabilities. The strong perception and computing capabilities brought by machine learning have endowed machines with a smart brain, and in some cases, their intelligence has exceeded human imagination, such as in the fields of Go and pulmonary imaging diagnosis.
However, with the deepening of machine learning research, more and more people have discovered the limitations of machine learning, such as how to improve the robustness and interpretability of learning models, which have always been hot issues. Machine learning gives artificial intelligence systems a smart brain, but just like students studying, some particularly smart students may make mistakes on very simple problems if they do not seriously and systematically listen to the teacher’s lectures; sometimes, even if they calculate the correct answer, other classmates often cannot understand their problem-solving ideas.
To make artificial intelligence a true top student, it must also have a brain that listens carefully; listening carefully allows this student to absorb a large amount of knowledge summarized by the teacher, which can help the brain reason, thus improving the efficiency of the problem-solving process, making it logical and well-founded. To achieve a higher level of artificial intelligence, it is essential to endow machines with a smart brain, giving them extraordinary learning abilities, while continuously imparting knowledge to them to make them more knowledgeable; thus, both machine learning and knowledge graphs are indispensable. In recent years, more and more artificial intelligence scholars have reiterated the importance of symbolism, reminding people not to be blinded by machine learning.
Machine learning and knowledge graphs are both important links on the road to achieving higher-level artificial intelligence, and neither can replace the other.The construction of knowledge graphs also needs to be planned comprehensively from Schema to knowledge extraction, with machine learning as a method to accelerate the efficiency of knowledge extraction. As builders of knowledge graphs, we must have a clearer mind to recognize the relationship between the two, so as to construct truly useful and usable knowledge graphs.
OMAHA’s Medical Knowledge Graph Development History:
August 2019: Medical knowledge graph Schema model released;
September 2019: “Drug Indications” knowledge graph released;
December 2019: “Clinical Pathway Treatment Related Tests” knowledge graph released;
2020: Upcoming release of “Clinical Pathway,” “Disease-Department,” “Disease-Symptom,” “Disease-Testing,” “Disease-Treatment (drugs, surgical operations)” knowledge graphs.
You can log on to the HiTA service platform (hita.omaha.org.cn) to learn more.
Contact Us
HiTA Service: [email protected]
Head of Digital Medicine Knowledge Center, Xu Meilan: [email protected]
Related Articles:
Knowledge Representation and Medical Knowledge Graph
OMAHA “Clinical Pathway-Disease Treatment Related Tests” knowledge graph released!
HiTA Knowledge Graph “Drug-Indications” graph data released!
OMAHA HiTA: Metadata | Terms | Knowledge Graph
For healing, we choose openness and sharing
Click “Read the Original Text” to immediately join us in the movement to return personal health medical data!