Reveal Ask: Your Search for the Truth

What is Ask?

Reveal, the global provider of a leading AI-powered eDiscovery, review and investigations platform that offers solutions for every stage of the EDRM process, publicly introduced its first Generative AI solution called “Ask” on Jan. 10, 2024.

Powered by Amazon Bedrock (AWS) LLMs, Ask is now an integral part of the Reveal platform and works in harmony with other AI capabilities and features in Reveal to provide an enhanced user experience.

Ask does not require any specialised technical expertise. Ask enables legal teams to simply ask a question about a topic within their workspace and retrieve precise answers, together with citations to specific documents. Thus, legal teams avoid the need to review hundreds of documents to simply get a single answer which saves valuable time and costs. Users may also include additional instructions that direct Ask to format the narrative in a variety of ways, such as using bullet points, making a response more concise, etc. This information can be easily copied and pasted into a separate document.

How does it work?

Ask is activated when you ingest data on a new matter or afterwards. It can be used across the entire workspace or on a subset of documents, e.g. within a particular date range.

Once you've entered a question, Ask retrieves a list of the most semantically similar documents related to your query, regardless of misspellings or synonyms thus addressing problems often posed by traditional keyword searches. It then creates a summary from the top 100 sources (i.e. documents) and provides information on the total number of relevant references found in each relevant document. Ask analyses documents on the snippet level which increases the accuracy of the semantic similarity analysis. Ask is designed for precision rather than recall, it is meant to find the best documents on a topic rather than all the documents on a topic.

The search process typically takes between 20-25 seconds to complete and return with results, regardless of the number of documents involved. It runs at the same speed when it's run on a thousand documents and over 1 million.

The cited documents appear in order of priority and can be added to a search for further filtering using Reveal's varied features and dashboards (such as date ranges, file extensions, entities, custodians, etc.) for the legal team to review.

Like other GenAI tools, Ask is intended to be used on text-based documents only and is not designed to search on images, audio or video files. It can't read metadata, but can read dates within the text of the documents if present. Ask is fully integrated with Reveal’s powerful search engine and allows users to use searches and filters to narrow down the subset of the documents to be searched by Ask.

For example, a question might be “Who ran Company A?”.

Use cases

There are no limits to how or when Ask can be used and creative thinkers can leverage it to explore new ideas and solve complex problems. However, like any AI tool it does not use intuition and asking it to find your "smoking gun" document, for example, simply won't work.

Ask can be used at any (or all) stages of your workflows for disclosure and investigation. For example, you may use Ask to identify a pre-seed set of documents which are then used to quickly build your Active Learning project. Alternatively, you may use Ask to identify potentially privileged documents in a production set by entering privilege factors to avoid any inadvertent disclosure of privileged documents (noting the limit of 100 documents). Ask can also generate a timeline or even provide questions for a witness or custodian interview.

When comparing supervised machine learning and GenAI, George Socha, Reveal's Senior Vice President of Brand Awareness, described it perfectly; supervised learning is asking the tool to “find more like this” and GenAI is asking the tool to “tell me something I don't know”.

He also explained the importance of using other search capabilities and AI tools to get the best results for your case as each feature serves its own purpose. For example, Ask is not a substitute for Active Learning or keyword searching, and vice versa.

Is it secure?

Ask is a Retrieval-Augmented Generation (RAG) AI, which means it’s case-specific and a response will only be provided if a document containing the answer is found in the workspace; if Ask is unable to find responsive content in the documents it is searching, it will tell you that. This ensures the narrative responses generated by Ask are accurate and reliable by using documents specific to an individual matter.

Each workspace is secured to a specific client, so the work product is not shared across other Reveal workspaces or the public internet. The responses generated by Ask are displayed only within the individual Reveal workspace and cannot be reproduced in another Reveal workspace. Ask uses AWS Bedrock as the LLM provider and follows the AWS security practices. To address client requirements that data be located in specific geographical areas, Reveal makes Ask available in a growing number of AWS regions.

Are there data privacy concerns?

Reveal never uses client data to train or update any of their AI capabilities and this applies to Ask as well. AWS Bedrock provides the same guarantees of never using the data sent to Bedrock to train their LLMs.

What are the costs?

Reveal are very upfront about costs and provide two general approaches to pricing. The first is designed for when clients intend to use Ask only for a subset of cases and the second gives clients the ability to use Ask across their whole universe of cases. This gives clients flexibility and allows them to choose the approach that best suits their needs.

How transparent should the parties be with each other?

George Socha explained that as a general proposition you should not have to divulge to the other side that you are using Ask to help you prepare your case – any more than you divulge that you search for keywords or evaluate documents for other patterns.

When using GenAI capabilities to prepare your case, such as determining who to talk to or what to ask about, there is no more obligation to disclose this than there is for other preparatory work typically done by support staff. This information is not commonly shared with the opposing side, and some parties might simply request the documents without needing this detail.

Even when you are using these capabilities to respond to formal disclosure requests, absent a specific agreement or order, you should not have to disclose that you use Ask. There should not be the same disclosure obligation as one that explicitly requires you to divulge your use of TAR.

Conclusion

Ask is an invaluable tool that enhances the user experience for Reveal's clients by efficiently and accurately retrieving information from vast document sets. Its immense potential means that users can come up with new methods in which they choose to use Ask and tailor its application to their specific needs, thus leading to more effective case preparation and strategy planning. Ask is a powerful complement to the existing tools within Reveal, and significantly strengthens the AI capabilities available to legal professionals for handling disclosures and investigations.

Reveal have also confirmed that it is still early days for Ask and they are continuing to make improvements. Furthermore, Reveal are working to build more GenAI capabilities and constantly expanding their toolbox to provide even more comprehensive solutions to their clients.