Reprint the official account | Knowledge management is in Xiabo
Since the successful launch of ChatGPT, AI systems like large language models (LLMs) have attracted global attention, although LLMs have been around for much longer. These systems now support many scenarios from chatbots and content generation to brainstorming and scripting. However, as these models become more complex, they become more likely to be wrong.
More recently, large language models like ChatGPT have produced inaccurate reports, argued incorrect facts, and described real-world biases in their answers. This has led to increasing concerns about the reliability of large language models.
This article will explore the problem with ChatGPT and other large language models, and discuss how knowledge graphs can help improve them to provide reliable AI-generated output.
What is AI Illusion?
AI hallucinations refer to the limitations of AI systems that force them to produce outputs that sound plausible but are not grounded in reality, or inconsistent with real-world knowledge. In some extreme cases, the output can be completely irrational and nonsensical. For example, a medical institution called Nabla once tested OpenAI's GPT-3 chatbot for medical advice, and the chatbot suggested that a simulated mental patient commit suicide.
These illusions occur when a model generates output based on statistical patterns in the data without fully understanding the underlying meaning or context, leading to nonsensical results. In some cases, they may also arise from training on biased or incomplete data, leading AI systems to make assumptions or conclusions that do not match reality.
AI illusion is a major challenge in developing reliable, trustworthy AI systems, especially in application scenarios where accuracy is critical.
What is a large language model?
Large Language Models LLMs are machine learning algorithms for interpreting, translating and summarizing natural language texts. These models use deep neural networks to learn from a wide range of training data to produce appropriate outputs. Through the self-attention mechanism of the transformer architecture, large-scale language models can associate words in sentences. Relationships between words are captured by evaluating and comparing the attention scores of each word in a text sequence. The training text data for large language models is accumulated from multiple sources, such as books, the open internet, articles, social media, or research papers.
Large language models are used in a wide range of applications such as conversational AI, content creation engines, search engines, customer service agents, and more. They are a powerful innovation aimed at automating and enhancing natural language processing tasks.
What's wrong with large language models?
While we cannot destroy the complexity and fluency of large language models like ChatGPT, the key is not to rely solely on their results, as they have a tendency to hallucinate. A model like ChatGPT may be flawed in supporting factual queries as it cannot provide hard evidence from up-to-date and verifiable information. This can cause the model to hallucinate and generate inaccurate or outdated responses.
According to AI researcher Yann LeCun, "LLMs do not have a clear understanding of the underlying reality described by language, and most human knowledge is non-linguistic." This means that while LLMs can generate syntactically and semantically plausible text, but they lack real-world experience. This makes them difficult to produce accurate output, especially when dealing with complex and delicate subjects that need to be observed in the real world. For example, humans learn to ski by practical trial and error, rather than using language theory.
As such, LLMs remain of limited usefulness in producing precise outputs, and may also adversely affect mission-critical industries such as healthcare, finance, or national security. For example, an LLM that generates factually inaccurate medical information could lead to incorrect medical diagnoses, resulting in loss of human life. Or a hallucinating LLM feeding financial firms with inaccurate legal analysis leading to decisions that could incur significant losses. LLM can also help cybercriminals generate phishing emails to gain unauthorized access to secure military systems, compromising national security.
What is a Knowledge Graph?
The knowledge graph is a graphical representation of knowledge in the form of nodes and edge networks, depicting real-world data entities and their relationships. Knowledge graphs provide context and meaning to structured and unstructured data, making them understandable. The entity information and relationship information of the knowledge graph are stored in the graph database, and the graph database is used as the knowledge base of the knowledge graph.
Knowledge graphs integrate various data sources and mapping relationships across any data store to help organizations retrieve meaningful facts from data and discover new facts through data analysis. Their ability to manage the fluctuating nature of real-world data allows them to adapt to changing data, which makes them an important tool for uncovering hidden insights from data.
How do knowledge graphs and LLMs work together?
Combining knowledge graphs and large-scale language models can provide a more powerful solution to the limitations of language knowledge and can potentially solve the problem of hallucinations, thereby improving the accuracy of query results.
Integrating knowledge graphs with large language models involves incorporating a contextual knowledge base into the model, allowing the model to make logical connections between concepts. This enables large language models to leverage various sources of information, including structured and unstructured data, to generate more accurate and relevant output. Furthermore, it allows the model to reason with a deeper understanding and generate more meaningful text.
For example, a biopharmaceutical company wants to improve its drug discovery. The organization might want to implement an LLM-based chatbot that can intuitively answer inquiries about clinical trials. However, the LLM may not have all the necessary information to provide an accurate answer.
To address this, the company combined its LLM with a knowledge graph engine to create a detailed medical knowledge base that includes structured and unstructured information about drugs and their trials. Therefore, if a user asks about clinical trials of a drug compound, LLM will quickly refer to the contextual knowledge base of the Knowledge Graph to identify and analyze all information related to that compound. This integration can enable companies to extract powerful insights from data and use them for breakthrough drug discovery.
The benefits of the combination of knowledge graph and LLM
Combining knowledge graph and large language model LLM can bring the following benefits, including:
Centralized source of accurate knowledge: By connecting the output of large language models to a knowledge graph engine, data can be centralized in a standardized format, making it more accessible and analysable. Knowledge graphs provide a visual representation of context, providing semantic insights that can be used to accurately answer questions about data, train machine learning models, or power business analytics.
Structured knowledge fusion of data in different formats: Knowledge graphs provide LLM with a structured way to tie concepts together. By merging data into a single, unified view, knowledge graphs can help organize data in an easy-to-understand format that can be used to make better decisions, identify new insights, and gain insights into data more comprehensive understanding.
Increase the potential intellectual value of collected data: By connecting together previously siled and inaccessible data, a knowledge graph engine presents all collected data as a single source of truth that can be analyzed to uncover hidden treasure troves of knowledge. This provides contextual depth to the informative value of large language models, which cannot be obtained by the model alone.
Provides LLM with a real-world human frame of reference: Large-scale language models excel at processing natural language text, but lack a real-world frame of reference. By connecting them to knowledge graphs, LLM can access structured knowledge that reflects patterns in real-world data, leading to a deeper understanding of nonlinguistic knowledge. This can help them generate more accurate and contextually relevant responses.
OpenKG
OpenKG (Chinese Open Knowledge Graph) aims to promote the openness, interconnection and crowdsourcing of knowledge graph data with Chinese as the core, and promote the open source and open source of knowledge graph algorithms, tools and platforms.
Click to read the original text and enter the OpenKG website.