To discuss the latest developments in artificial intelligence and its applications in the legal industry, we interviewed Ather Gattami, the founder and CEO of Bitynamics, a Stockholm-based AI development company.
– There’s been so much confusion between AI and automation. What would you say is the difference between these?
A: Automation is a general concept where you want things to run in an automated fashion, with as little human intervention as possible. Automation does not necessarily have intelligence that makes a decision. You can, for instance, set your alarm in your cellphone to wake you up at 7:00AM every day, and there is no intelligence behind it. It is purely rule-based.
As regards AI, there are many applications. One of them is trying to mimic a human, e.g. understanding text, recognizing objects and images, having conversations, listening to the speech and transcribing the speech, etc. Another application domain for AI is where it beats human intelligence. In all cases, AI is capable of arriving at a decision.
– Is it fair to say that most AI solutions are based on deep learning? If so, do you think this trend will change?
A: Not sure about that. But what we are witnessing is that deep learning is solving problems that were hard to solve before. It is making progress, especially in such fields as image recognition and natural language processing. The idea of deep learning is not new, it has been around since 1960s. But now we have much more data and computing power to utilize deep learning. It’s been one of the main “cools” of machine learning in recent years. But it is naïve to think that deep learning will solve all the problems.
– Why do you think most legal AI vendors focus on contract review and analysis?
A: There seem to be two reasons:
- They see someone who has done it and then follow-up. For example, in Sweden, one bank started to have a chatbot and other banks followed, although technology is very immature. The tech has been viral. The same applies to AI in the legal profession.
- Many firms are looking into automating “boring” work. You want AI to take care of parts of the work that you don’t think is interesting but is complex and time-consuming.
One should note that the NLP technology for legal work is very complicated from a linguistic standpoint and data point of view. How to you connect the dots between different cases? There’s much of deep analytics work. NLP is still in its infancy, but I believe there are still enough tools where you can actually help lawyers and courts to understand cases better. AI can be helpful in detecting patterns and connect historical cases with the current case. For courts, AI may help to see how different laws were interpreted in different cases and arrive at a more precise interpretation.
– What are the most recent advancements in natural language processing?
A: A combination of Recurrent Neural Networks and Word Vector Embeddings. Text is represented as sequential data where one word may depend on the previous word and where you have output depending on the previous history. One could use a compression technique where a word is mapped to a list of numbers used to describe the relations with the previous words. This way, one may predict the next word depending on the previous word.
– The GDPR requires legal AI providers to be able to explain to users how a decision has been arrived at by their algorithm (the so-called “right to explanation”). What do you think AI vendors should do ensure AI’s reviewability and avoid “black boxes”?
A: You need to make sure that you can see the relationships between different parts of the system and how these are put together. First, you need to project the parameters that are taken into account. There is no generic “one-size-fits-all” solution, and it all depends on the application. You look at a parameter importance and how it affects your output (the decision). You must also have an intelligent system that learns how to “pronounce” these parameters and put them into context. This is one of the hardest tasks in natural language processing. Since today there is no general way to do it, probably you would need to have a custom-based solution for every question you are looking the answer for.
– The recently settled trial between Waymo and Uber has shown that trade secrets will be one of the most frequent bases for corporate litigation. Apart from legal safeguards, is there any technical way to protect trade secrets from being used by a top AI developer who changed the employer?
A: There is no technical way that I know of. It all comes down to trust. The thing is that you need to detect that a person has done something at the previous employer and somehow detect that something similar has been made at the next company, and this thing is extremely hard at so many levels, unless you have complete insider access. That is why companies are so sensitive when top developers are leaving, since IP rights will be transferred to a certain extent, that’s for sure. And that is why companies should focus on retaining talent. At the same time, I think it is fair that a person can bargain with their knowledge in the job market, since it is part of their experience. New employers ask about experience, and you cannot forget the knowledge that you have already gained. But ethical aspects should also be taken into consideration here.
– As a consumer, what legal services you would like to see disrupted by AI?
A: Contract work in general: you need a written agreement in many cases, such as renting or purchasing an apartment, etc. I’d also like to see AI verify the interpretation of laws. We often hear that many laws have been interpreted differently, and AI could help to bring a better alignment in this area. Another area of application would be corporate knowledge management. In a law firm usually there are a lot of email exchanges and interpretations of different cases, and AI could collect and aggregate this knowledge from this previous experience. I’d like to see some kind of a search engine where you see information from emails, meetings, cases – different expertise, who knows what, what did the law firm decide in different cases, all kinds of content that is dynamic and hard for a single person to track. This solution would also help with organizing information left by former employees, where such information would have been hard to preserve over time otherwise.