Transparency, AI, and the Future of Chinese Courts: A Discussion with Professor Benjamin Liebman
Chinese Courts are increasingly adopting artificial intelligence technology.
By: Tim Wang, staff member
Benjamin Liebman is the Robert L. Lieff Professor of Law at Columbia Law School. He is also the director of both the Hong Yen Chang Center for Chinese Legal Studies and The Parker School of Foreign and Comparative Law. Widely regarded as a preeminent scholar of contemporary Chinese law, Professor Liebman studies Chinese court judgments, Chinese tort law, Chinese criminal procedure, and the evolution of China’s courts. He has also recently written about the roles of artificial intelligence and big data in the Chinese legal system, including in a forthcoming article in Volume 59, Issue 3 of the Columbia Journal of Transnational Law titled Automatic Fairness? Artificial Intelligence in the Chinese Courts. I sat down with Professor Liebman to discuss the implications of big data and artificial intelligence for the future of Chinese, as well as other, legal systems.
***
This interview has been edited for clarity and length.
The top-down directive from the Supreme People’s Court to publicize court documents has led to more than 100 million court documents becoming public. Yet China’s court transparency is still relatively incomplete. In your article, you note that this spotty compliance is in part due to the not unfamiliar power joust between China’s central and local actors. What are some of the disincentives that might dissuade local court officials from publishing court documents and opinions online?
First, I think it’s pretty remarkable that they’ve put 100 million documents online. The Supreme People’s Court in particular deserves credit for doing so because this is not what you generally expect an authoritarian legal system to do. Even if it isn’t full transparency, this level of transparency is pretty significant. We should remember that there’s also a big chunk of cases, such as mediation and family law cases, that aren’t supposed to go online. So the set of cases that are not online but are supposed to be online is smaller than the overall number of missing cases. That said, we have found that noncompliance is due to a couple of factors. First, we outlined in our previous piece, Mass Digitization of Chinese Court Decisions: How to Use Text as Data in the Field of Chinese Law, some noncompliance isn’t necessarily for a political purpose or because officials want to hide the ball. Rather, there is bureaucratic sluggishness - digitizing documents is hard, takes time, and requires resources. Publicizing court documents has been a target for evaluating courts but hasn’t been a hard target, and so especially in the early years this wasn’t as important of a goal for local courts as it was for higher-level courts. Now, that said, there’s also pretty good evidence that more sensitive cases, for example, those that involve powerful or influential companies, don’t get posted online. But I don’t know that there are concerted efforts by local courts to push back against the requirement to post things online due to purely local interests. Rather, there’s bureaucratic inertia, and courts may sometimes face pressure from other parties not to put things online.
The CCP endorsed judicial transparency in part because of social and media outrage over wrongly-decided cases. What has been the public response to the public release of millions of court decisions? Has it garnered more legitimacy for the CCP and the courts? And has it helped the media and other civic institutions check the judicial process?
Public opinion is very difficult to measure in China. I don’t think we know an answer or have a way of figuring out an answer. I think the public court documents have mainly been used in the legal community, both by lawyers and the legal tech industry in China. So I’d say it’s been used more in the legal community than the broader community. There certainly was some concern at the time these policies were implemented that the documents could be used to criticize the courts and make them look bad, but I just don’t think we’ve seen that. There isn’t much evidence suggesting that there are downsides to the courts so far.
If you talk to lawyers and judges in China, they will tell you that increased transparency is useful in multiple ways. Part of it is that the judges will be more careful, mistakes will be noticed, and some of the things we might have seen in earlier periods -- where courts are clearly just not following the law -- becomes much harder to do. Transparency does make courts pay more attention. It also can be helpful for protecting the courts because judges facing external interference can now say: “look, this is going online, I can’t help you even if I wanted to.” So the fact that these cases will see sunshine makes it easier for courts to push back against impermissible forms of interference. And, of course, this dovetails with the CCP’s dramatic crackdown on corruption in and interference with the courts that we’ve seen in recent years.
Also, I’ll point out that the policy also accords with the central leadership’s recent emphasis on big data and artificial intelligence. So it’s not just about public support, but also about the courts aligning themselves with the core policy goals of the central Party-state.
Does the adoption of the algorithm-assisted system mean that China is moving towards a de facto precedent-based system? How might an algorithm-assisted system differ from the system of stare decisis in the United States and elsewhere?
You’re right that China is not officially a precedent-based system. But all systems rely on some form of precedent. Even before AI, it was pretty common for courts and lawyers to look to other similar cases informally.
AI is going to be the most useful in the cases where it’s least needed - it can help with high throughput, routine cases by standardizing those cases a bit more. A lot of those cases were already pretty standardized -- for example, judges already know what a lot of the sentences in criminal cases are supposed to be. But there’s still a perception that courts are undermined by inconsistent rulings. I think part of the goal of AI then is to move towards increased standardization of court decisions. And maybe it also helps move cases more quickly through the system. I would call this more standardization than creating precedent. It’s more that courts are trying to ensure that like cases are decided in like fashion, rather than the courts binding themselves to the decisions and reasoning of past judicial opinion.
Some scholars argue that Chinese courts inherently serve political purposes and therefore need the flexibility to depart from legal considerations for political ones. Isn’t an algorithm-assisted system that binds a court in tension with that narrative?
This is exactly right and we discuss this tension in our paper as a central challenge. Courts in China are not just machines that apply the law; instead, they are supposed to consider a broader range of social goals and sometimes political factors as well. Given that, it’s hard to imagine how an algorithm can account for those factors. We’ve seen some shift away from the fairness-based rationales for decisions, including in revisions made in the new civil code, but courts are still supposed to consider the possibility of social unrest or the need to support indigent or weaker parties. So it’s hard to see how algorithms are going to take these considerations into account. The question thus becomes: is AI also going to be used as a tool to push courts away from judging that takes account of factors such as equity and social stability or are algorithms going to have to give way to these types of social considerations in a wide range of cases? It’s too early now to tell which way the central court leadership is leaning on this issue.
Laws can be amended. And in China, there are also shifting political directives that may alter the way certain legal problems are dealt with. For example, the strike-hard campaigns (hard crackdown on crime) in the 1990s were followed by a shift towards the “kill fewer, kill cautiously” policy in 2007. How would an algorithm-assisted system adapt to such legal or policy changes?
This goes to the question of how these AI algorithms are created. My understanding is that there are two different ways of doing it. One is machine learning based on prior decisions, which raises the question of whether the algorithm is baking in cases that might not have been accurate. You would need some human review of the cases that are being relied upon. You can also have people come up with decision trees — which is happening in Shanghai — that try to map out in effect what should happen in each situation. In either case, when you have big policy changes, you’re going to have to figure out a way to translate that into the algorithms. I think what judges would say is that right now and in the near future, they don’t expect to be replaced by algorithms. Rather, algorithms are used as a tool to identify outliers and make sure that decisions are roughly in-line with similar cases. But even so, if there is a policy shift then they will absolutely have to figure out a way to reflect that in the algorithms.
You discuss the proliferation of AI potentially creates inequality because those with more resources are the ones that are able to crunch big data. Do you think this just the inevitable result of judicial systems that already favor those with more resources? To the extent AI exacerbates this inequality further, do you have any proposals for narrowing the gap?
The concerns we raise about equity are about the availability of data generally and who’s using it, which is a common problem around the world. If data can be marshaled to predict outcomes, then absolutely the availability of AI would exacerbate this disparity. I don’t think that’s a reason for restricting the flow of data, but it requires us to think harder about how to ensure equity for litigants. Part of the idea for publicizing these documents was to make them available to ordinary people. But there is an inevitable trend that once there’s a lot of data out there, anyone who has the ability to process the data is going to have an advantage. There’s a lot of litigation consulting that goes on based on data analysis, which allows folks with more resources to have better information.
Another point about equity is that once you have the black box of AI being deployed, it shifts power to the algorithm writer. How do we check if the algorithm is correct? This question isn’t even being asked in China yet. This is a problem in the United States as well, but we also have media and civil society to act as a check to some degree. If you look at some of the work that’s being done on predictive software in the U.S. courts -- for example, on predicting dangerousness for parole determinations -- that work is being done by academics, civil society, and the media. You need somebody to scrutinize and check these algorithms. But China doesn’t have those same voices serving as a check. That makes the inequalities worse because there are fewer autonomous and robust civil society actors that serve as a counterbalance to the shift in power to the corporations writing these algorithms.
You also note in your article that the SPC has recently limited the number of results that an individual can view on the website. Why have they done this?
It’s a little unclear why the SPC is doing this. They say that it is to limit companies from scraping data. Apparently, there is so much scraping going on that it has impaired the ability of ordinary users to use the website. But I think they also view the marketization of this data with some concern. It’s obvious they aren’t completely comfortable with companies marketizing and repackaging the data. Perhaps they are okay with the data being used in individual cases but want greater oversight when the information is used in large datasets.
Unlike China, Other countries’ judicial systems do not have a supervisory arm embedded within the courts to monitor judicial behavior. So is the centralization of power away from the judiciary through AI a problem unique to China? How might executives in other countries, where the separation of powers is institutionalized or a constitutional mandate, encroach upon judicial spheres through using AI?
We’ve seen in the West a dramatic increase in executive power, especially driven by the rise of populist leaders. And that’s at the expense of other branches -- we’ve seen the weakening of the courts and legislature in our own country. That’s one form of centralization of power.
In China, AI is being imposed on the courts by the courts themselves. But one could certainly envision executives in other countries rolling out AI as a way to curtail the courts, perhaps as a way over time to restrict the judiciary’s ability to independently adjudicate cases or just generally weaken the position of courts. We haven’t seen this, but we think there needs to be a conversation about the broad range of potential constraints on judicial authority in different jurisdictions. And as we continue to think about the toolbox that authoritarian or quasi-authoritarian leaders have available to constrain the courts, AI should be a part of the conversation along with some of the other things that have gotten more attention to date.
To conclude, I think that this is an area where China is leading the world. First, there are certain things that can be done with technology in China that can’t be done elsewhere, sometimes because there isn’t that institutional history or sometimes because the institutions have different backgrounds or are weak. We don’t think what happens in China will happen elsewhere necessarily, but we do think that China will have a more rapid rollout of AI in the courts compared to other countries. There may be some important lessons to learn from China’s experience, both regarding what’s possible and also the implications.
Tim Wang is a second-year student at Columbia Law School and a Staff member of the Columbia Journal of Transnational Law. He graduated from Columbia College in 2017.