INTRODUCTION
Using solutions powered by artificial intelligence (AI) to assist with document review is now an accepted practice in the legal community. With the overwhelming amount of data to review for a case or investigation, practitioners continue to look for defensible ways to reduce datasets. The use of analytics such as continuous active learning, concept analysis, clustering and solutions that can detect similar documents are becoming increasingly necessary.
Portable AI models are now the new tool on the scene that can help with not only review for discovery, but also so much more. The use cases are growing, the technology is more powerful than ever.
In recent years there has been a noticeable shift from hesitation around using emerging technologies to embracing them to improve efficiency. To remain innovative, it is imperative to watch tech trends and explore partnerships with providers that can guide where portable models are most beneficial. Even though more organisations want to use new tools, there are grey areas around function and ROI considerations for portable AI models that teams must understand.
THE AI EVOLUTION
Most legal professionals and review teams now accept that they need to use automated tools to get through the documents for a case or investigation. With more data generated daily, manual review is impossible and will not offer the full story in reasonable timeframes or the constraints of budgets.
Advanced technologies powered by AI have offered a mechanism for speedy and accurate reviews that pull out the most important documents to review. It started with technology assisted review (TAR) tools, which automatically categorise and rank documents based on an initial training set. Then came continuous active learning (CAL) which allowed reviewers to review data on a rolling basis and does not need the creation of seed sets. Legal teams then found ways to use AI tools for other purposes such as early case assessment (ECA), contract analytics, and to identify key custodians and search terms to use.
Now, portable models are the next level TAR tool. Portable models are trained on data from prior or related datasets. Prior insights from the portable model are then used to jumpstart new reviews. This drops the need to create and train a new model every time a similar matter or question arises. There are generally two ways that teams are using these tools. Both options offer ways to repurpose past work product across multiple matters. One way is to apply an established pre-trained model to identify language in datasets on repeat topics such as privileged content or insulting behaviour. Innovative service providers in this space are building up libraries of more generalised models to offer to their clients. Another option is a bespoke model, which organisations train to pinpoint issues or answer questions unique to themselves. This is a more customised option that offers client-specific value and be a competitive advantage over others.
COMPELLING BENEFITS
Portable models truly elevate predictive capabilities for review teams. These tools offer a range of benefits and create opportunities to permanently transform processes. Here are four major advantages:
• Speedier reviews: Teams can identify relevant documents earlier in the review process. Portable models can analyse the entire corpus of documents to identify patterns and pinpoint documents responsive to topics of interest. While this benefit is also available with other TAR technology, the models can target specific recurring topics without the need to train a new model for each matter.
• Long term cost savings: Although there is an upfront investment to build custom model libraries or obtain pre-built models the cost savings will materialise over time. These models can narrow datasets resulting in fewer documents for manual review. Portable models can also assist in traditionally costly discovery tasks such as privilege review or ECA.
• Tech layering: Portable models can be layered with other solutions to get more accurate results. There is also the ability to run multiple models in parallel on the same dataset. For example, a litigation team could run models to detect privilege and harassing language simultaneously to pinpoint the hot documents in an employment case. Having this information pre-discovery assists evaluating early settlement opportunities.
• Consistency: Using model libraries results in more consistent work product across matters and legal teams. This is especially true for internally built model libraries, as they will be more customised to the organisation. Options for these bespoke models include fine-tuning pre-built base models to fit the specific environment, building models based on their prior activities, and creating models from scratch.
USE CASES
Due to portable models being such a recent innovation the use cases for portable models is not widespread. Portable AI models can apply in many situations both related to legal processes and other business needs. We’ve highlighted three use cases below:
Legal
There are several instances where portable models can aid legal teams. The most obvious is to cull datasets for review. That being said portable models have broader applications in the eDiscovery process. Portable models can be used for early case assessment, settlement evaluation, aid in coming up with search terms, identify custodians and for litigation risk analysis.
Here are two illustrations of how teams can apply AI models, one in the traditional review sense and the other in a more unique way for litigation analysis.
• Employment: In a sexual harassment case, teams can apply portable models on communication data to pinpoint sexually explicit themes, concepts, and language. The portable models can also detect comments on appearance, bullying, discrimination, harassment, and/or threatening behaviour. This can help parties jumpstart review by identifying key actors and witnesses early on.
• Deposition Preparation: large matters, portable models can help predict possible testimony and case issues. Customised pre-built models can help teams with strategic decisions such as when to settle, who to drop from a case, and who to call as a witness.
Compliance
Portable models aid compliance in several ways, such as filtering irrelevant data during investigations, meeting regulatory deadlines, demonstrating legal compliance, and can even be used for DSAR compliance. Models assign sentiment scores to detect fraudulent behaviour, prioritise evidence hotspots and reduce compliance risk. For instance, a model focused on kickbacks or insider trading can reveal fraudulent patterns by identifying custodians using specific words or phrases.
Another “out of the box” use case for compliance is to monitor employee behaviour. This can be particularly useful in the financial industry where organisations must ensure that employees are acting appropriately and not promising their clients unattainable returns.
Cyber
Incident response plans are crucial for cybersecurity teams as data breaches become more common. These plans map out what to do in case of a breach Including identification and review of exfiltrated data and notification of impacted individuals. In these phases portable models can identify sensitive data, determine where that data exists and who to notify. This enables faster and more efficient response times reducing the impacts of data breaches.
PREDICTED TRENDS
Monitoring market trends is crucial for staying competitive. It’s important to evaluate current and future needs and stay informed about developing technology and how they can be used. This enables organisations to demonstrate ROI to leadership and adjust strategic direction. Two predicted trends in portable models are discussed below.
First, adoption will increase over the next few years. However, it could take longer for organisations not prioritising digital transformation as they may have difficulties harnessing their own data and getting buy-in from leadership. Organisations prioritising transformation will have higher success rates in adopting AI tools as they will have the necessary processes and data in place. These organisations are also likely having partnerships with innovative service providers to leverage off to help them pinpoint specific use cases to maximise ongoing use of AI.
Second, the topic of portable models will make it to the courts. The use of emerging technologies – particularly those using AI has been a trending topic for years. As data volumes get bigger there will be a bigger push to make production more proportional. Portable models among other technology will have a key role to play in making the discovery process more efficient and cost-effective.
Lawyers must not forget they have ethical obligations to understand the benefits and risks of emerging technologies. They should be aware of the basic features of up-and-coming technology, to prove value to internal stakeholders, other parts of the enterprise, and clients. Continuing education and awareness are necessary to fulfill this ethical obligation. Education such as those at legal conferences have started to address this need which is a step in the right direction in increasing awareness and adoption of AI.
CONCLUSION
Portable AI models are the new tech tools to watch. The use cases and maturity will only continue to expand as more organisations become aware of how these models work and what benefits they can offer to legal and other departments. Now is the time to monitor new industry and court developments, evaluate investment opportunities with providers offering pre-built or bespoke models, and discuss potential use cases with leadership teams.
To download the pdf of this paper please click here
