Can AI Replace Traditional Legal Services?

31 Jul.,2025

AI legal article retrieval occasionally encounters errors; legal translation is not standardized and cannot solve the problem of localization; AI does not have human experience; it is powerless in the field of legal psychology and AI technology cannot break through its own original limitations.

 

Can AI completely replace traditional legal services? ——Take DeepSeek as an example

                                 Written by Rocke,Feb.8, 2025

 

Recently, DeepSeek (hereinafter referred to as DS) has emerged, making many people full of expectations that AI can replace traditional legal services. I am a university law professor and a senior lawyer who has been in the industry for more than 30 years. I just want to talk about my views on this issue from the perspective of legal profession.

Can AI completely replace traditional legal services? My answer is: NO.

The reason I say NO is that AI legal article retrieval occasionally encounters errors; legal translation is not standardized and cannot solve the problem of localization; AI does not have human experience; it is powerless in the field of legal psychology and AI technology cannot break through its own original limitations.

1. AI legal article retrieval occasionally encounters errors

My saying that AI legal article retrieval occasionally encounters errors may cause some people to be angry. However, facts speak louder than words, and we can find the problem by doing a few simple tests.

a. The answer is wrong.

Let's first do a test on the legal text search of the Chinese Civil Code, and ask DS a question: "What is the content of Article 890 of the Civil Code of the People's Republic of China?" It gives the answer: "Article 890 of the Civil Code of the People's Republic of China: The custodian shall not use or permit a third party to use the custody, except where the parties have otherwise agreed." We compared the legal text downloaded in advance from the Chinese Laws and Regulations Database (here, we will call it the "correct article") with the answer given by DS, and found that the correct article is "Article 890 The custody contract is established when the custody is delivered, except where the parties have otherwise agreed."

Let's change another legal text for testing and ask a question: "What is the content of Article 41 of the Foreign Investment Law of the People's Republic of China?" DS gave the answer: "Article 41 of the Foreign Investment Law of the People's Republic of China: Any violation of laws and regulations by foreign investors and foreign-invested enterprises shall be investigated and dealt with by the relevant departments in accordance with the law, and shall be included in the credit information system in accordance with relevant national regulations." However, the correct article is "Article 41. If the state has other provisions on the management of foreign investors' investment in banking, securities, insurance and other financial industries in China, or in securities markets, foreign exchange markets and other financial markets, they shall be governed by such provisions."

If the test continues, similar errors are expected to occur.

In theory, legal provisions are very rigorous, and no word or punctuation mark should be wrong, but AI retrieval has such a problem. If a lawyer takes the legal provisions retrieved by AI to appear in court for the client to participate in the lawsuit, or a judge takes the legal provisions retrieved by AI to write a judgment, then it will cause very serious consequences.

b. The answer is incomplete and there are "omissions".

For example, if you ask DS "What does Article 222 of the Chinese Civil Code stipulate?", the answer it gives is "If a party provides false materials to apply for registration and causes damage to others, he shall bear the liability for compensation." In fact, this is only the first paragraph of the law. The second paragraph of the law also says: "If a registration error causes damage to others, the registration agency shall bear the liability for compensation. After the registration agency compensates, it can seek compensation from the person who caused the registration error." This paragraph is not displayed. There are many similar examples, so I won't list them one by one here.

c. Failure to accurately understand the question.

For example, if you ask DS "legal provisions on exemption from prosecution", it lists Article 16 of the Chinese Criminal Procedure Law, Article 37 of the Chinese Criminal Law, etc. However, these provisions are about "no punishment", not about exemption from prosecution. DS obviously confuses the two legal concepts of "prosecution" and "punishment".

2.Non-standard legal translation cannot solve the problem of "localization".

Regarding non-standard translation, we did a test and entered "Translate the following Chinese into English: The policy-based closure of mines means that enterprises can no longer engage in related production and operation activities, which will greatly affect the interests of the enterprises. Therefore, the relevant enterprises are likely to file an administrative reconsideration against the closure decision or file an administrative lawsuit against the reconsideration, demanding the revocation of the closure decision and compensation for economic losses." It gave a translation: "The policy-mandated closure of mines means that enterprises can no longer engage in relevant production and operational activities, which will significantly impact their interests. As a result, the affected enterprises are highly likely to apply for administrative reconsideration of the closure decision or, if dissatisfied with the reconsideration, initiate administrative litigation to demand the revocation of the closure decision and seek compensation for economic losses." This translation, if regarded as a general mechanical translation, does not seem to have much problem. However, it lacks the normative rigor that legal English should have. For example, in a specific administrative litigation case, we can only say that the specific administrative action is "based on policy", so as to explain the basis for this specific action; but we cannot say that the action is " mandated by policy", because policy can only mandate a certain type of action, not a specific action. For another example, in legal English, the word that indicates causality is "Therefore", not the colloquial and weaker "As a result", etc.

In addition, "file an administrative lawsuit against something or somebody" is a fixed sentence pattern, and the sentence pattern of "initiate administrative litigation to demand" is neither standardized nor concise enough.

AI cannot solve the localization problem of legal translation, which is the fundamental reason why it cannot replace manual translation.

What is localized translation? In what aspects is localized translation mainly used?

The general translation provided by AI is not enough for the task of localized translation, because localized translation requires translators to be familiar with the language habits, cultural background and national characteristics of the target market.

In the process of localization, only those that conform to the way of speaking and language habits of local people will be accepted by local people, and only those that are recognized by local consumers can form strong competitiveness, thereby winning the market, occupying the market, and making profits from the market. Of course, only by having a sufficient understanding of the target market can the language be translated flawlessly and localization be completed smoothly.

Translation is not only a conversion at the language level (which is what AI is good at), but also involves considerations of multiple dimensions such as culture, society, and history (which is the shortcoming of AI). Postmodern translation theory believes that translation is a creative interpretation and reconstruction process. Translators can freely practice translation according to their own understanding and purpose. AI obviously cannot reach such a realm and height.

3. AI does not have "human experience"

U.S. Supreme Court Justice Oliver Wendell Holmes once said, "The life of law is not logic, but experience." What is "logic"? One by one legal provisions, one by one legal reasoning, reflects logic. What is "experience"? The historical accumulation of a person doing a kind of thing is experience. Logic is stereotyped, while experience is highly personalized. Due to factors such as age, living environment, cultural customs, and education level, everyone will have different experiences even if they do the same thing. Is experience useful? There can only be one standard, which is to see whether it helps to accomplish or do a thing well.

Here, let me give you another example. You ask DS, "The original document is damaged, there is no other supporting material, and there is only a copy of the IOU. Can I win the case?" The first time you ask it, it answers, "If there is only a copy of the IOU and the other party denies it, the court may not accept the evidence." The second time you ask the same question, it answers, "If there is only a copy of the IOU and the other party denies it, the possibility of winning the case is very low." Such an answer is undoubtedly correct in theory; if you ask the same question to a young lawyer, he will probably answer like this, which is also a standard answer in a sense.

However, a case I personally experienced showed a different result. When my client came to consult me, he only had a copy of the IOU, but not the original. He said he had lent more than 2 million yuan to the other party, and the latter wrote him an IOU, which stated that the loan period was 1 year, but more than 5 years had passed and the other party had not repaid him. He wanted to sue the other party, but he lost the original IOU and only had a not very clear copy. He consulted several lawyers, and they told him that without the original evidence, he could not sue, let alone win the case. Seeing that he was desperate, I told him that he could give it a try, and the lawsuit might go as we wished. As a result, as his attorney, I not only successfully completed the case filing and court hearing procedures, and obtained the winning judgment, but also helped him apply to the court for compulsory execution and helped him recover the more than 2 million yuan. He was overjoyed and asked me how I did it? I told him that based on his description, I inferred that the other party would not hire a lawyer or appear in court himself; based on my previous understanding, I inferred that the judges of the grassroots courts would not carefully scrutinize every piece of evidence. As expected, the other party did not appear in court, which was equivalent to giving up the right to cross-examination; the presiding judge directly adopted the photocopy we provided as the final evidence and made a judgment that we won the case. Of course, the most important point is that the other party did not appeal or raise an objection to the execution.

According to "logic", it is difficult for us to win this case, and the more than 2 million yuan will not be recovered. However, we turned "impossible" into possible, which is the power of "experience".

A few years later, I met the other party by chance. He knows that I am the lawyer who represented the other party in the lawsuit against him. He did not complain as expected, but explained to me that he did not repay the loan on time because he had no money; he did not hire a lawyer because he knew that the case would definitely lose and there was no need to hire a lawyer; he did not appear in court because he was ashamed to see my client. Finally, the last bit of worry about false litigation in my heart disappeared.

4. Helpless in the field of legal psychology

Clients often face pressure and anxiety in legal issues, and the emotional support of lawyers can help clients better cope with these challenges. When communicating with clients, we can understand their emotional state and provide support and comfort. This kind of emotional resonance and identification cannot be simulated by AI.

When handling cases, lawyers need to make moral and ethical judgments, such as how to balance the interests of clients with social fairness and justice. These judgments often involve complex interpersonal relationships and social values, and AI lacks this moral and ethical understanding.

Senior lawyers are usually able to develop personalized legal strategies based on the unique circumstances of specific cases, which requires a deep understanding of the law, client background, and case details. Although AI can analyze data and provide advice, lawyers' experience and intuition are indispensable when developing personalized strategies.

In court, lawyers need to make verbal defenses, using persuasive skills and instant response capabilities to respond to the opponent's arguments. This on-site interaction and flexible adaptability are beyond the reach of AI.

Establishing trust between lawyers and clients, judges, and other legal professionals is crucial to the success of a case. The establishment and maintenance of such interpersonal relationships relies on human social skills and emotional intelligence, which AI cannot do.

There are multiple interests and complex legal terms involved in legal proceedings, and lawyers are able to navigate these complexities and make wise decisions; AI can provide information, but judgments and decisions in complex situations still need to be made by human lawyers.

Our conclusion is that although AI can greatly improve efficiency in legal research, paperwork and information retrieval, lawyers are still indispensable in terms of emotional understanding, ethical judgment, personalized strategies, court defense, interpersonal relationships and complex legal procedures.

5. AI technology cannot break through its own "original limitations"

What is "original limitations"? It refers to the limitations that AI technology is born with and cannot be avoided through learning or optimization.

The basic principle of AI technology is to imitate human intelligent thinking and behavior, and to achieve it through computer programs and algorithms. The working principle of AI technology mainly includes three steps: data collection, data processing and decision generation. Data collection is the foundation of AI technology, and the collected data is not static or unchanged, which means that the time difference between data collection and the generation of the latest data cannot be eliminated. In other words, the lag of data collection cannot be avoided. This lag may even cause the collected data to be outdated or invalid, and the answers generated based on this invalid data will have serious consequences.

Although AI technology can simulate human intelligent behavior, it is essentially an algorithm and model based on mathematics and statistics. The AI ​​system does not have real consciousness, thinking, and emotions. It only realizes specific functions and tasks by calculating and analyzing data. Therefore, the application of AI technology needs to be combined with human expertise and judgment to ensure its correctness and reliability in legal situations.

The basic principles and working principles of AI technology may vary in different application fields, but generally follow the basic process of data collection, data processing, and decision generation. Through continuous data collection, processing, and decision generation, the AI ​​system can gradually learn and optimize its own performance, improve its intelligence level and application capabilities.

However, no matter how AI learns and optimizes, it is impossible to overcome its original limitations, just like a man, no matter how powerful he is, can't lift his body with his own hands.

The above text is just some of my thoughts on the relationship between AI and traditional legal services using DS as an example. If I accidentally offend DS or other AI, I apologize for this. I just think about some problems from a professional perspective.

I stand firmly behind the idea that AI engineers or programmers will be able to discover and solve these problems to the greatest extent, so that DS and other AIs can continue to improve and benefit mankind together.