Marketing

TAR + Advanced AI: The Future is Now

Issue link: https://content.lighthouseglobal.com/i/1412483

Contents of this Issue

Navigation

Page 6 of 12

WHITE PAPER: TAR + ADVANCED AI – THE FUTURE IS NOW | 7 Deep Learning: Deep learning is a subset of machine learning that does not require human intervention to process data. It consists of a neural network with "three or more layers" that simulate the behavior of the human brain and learn from large amounts of data. Because of its multiple layers of neural networks, its predictions are more refined than machine learning. 25 Graphical Modeling: A graph that shows the probalisitic relationships among a set of variables. Within the ediscovery space, graphical modeling is often used to show communication networks within a dataset (see social network analysis below). 26 Limited Memory: Unlike reactive AI, limited memory systems do have the ability to learn from historical data to make decisions. Machine learning and deep learning use limited memory to train by ingesting large volumes of data that is then stored in their memory from a reference model for solving future problems. Machine Learning: Machine learning is a branch of AI, which "focuses on the use of data and algorithms to imitate the way that humans learn, gradually improving its accuracy." 27 In machine learning, algorithms are trained to make classifications or predictions. 28 This is especially important in ediscovery as data volumes continue to grow and diversify. Traditional machine learning requires human intervention to process data. 29 Natural Language Processing (NLP): NLP is a branch of AI computer science that is concerned with giving computers the ability to understand text (and spoken word) in the same way that humans understand it. It combines rule-based modeling of the human language with statistical, machine learning, and deep learning models to process human language and "understand" its full meaning, including the intent and sentiment of the writer (or speaker). 30 Reactive AI: Reactive AI systems are a much older (and somewhat outdated in the non-ediscovery world) system of AI. For example, reactive systems were what IBM's famous "Deep Blue" machine used in the late 1990s. Reactive AI does not have the ability to use previous data and experience to inform present decisions. It can only be used to automatically respond to a limited set of inputs, and thus are not used by machine learning or deep learning. 31 Social Network Analysis (SNA): Visual representation of social networks and communication, using analytic software and graphical modeling. In the legal space, SNA is extremely valuable for case teams to gain a big-picture view of how people represented within the dataset were communicating (who they were communicating with, how often, what times of day, etc.). 32 Supervised Machine Learning: A subcategory of machine learning that uses labeled datasets to train algorithms to classify data or predict outcomes. As data is inputted into the model, machine learning weights that data until the model "stabilizes." 33 Unsupervised Machine Learning: A subcategory of machine learning where algorithms are able to analyze and cluster unlabeled datasets to discover patterns or data groupings without the need for human intervention. 34 TAR vs. AI Common questions around TAR and AI center on their relation: Is TAR, in fact, AI? Does AI in ediscovery mean TAR? The answer is this: The "technology" most commonly behind traditional Technology Assisted Review is supervised machine learning, a subset of AI as previously noted. So yes, the current TAR process uses a subset of AI. However, it's worth noting that the other part of TAR is the process with which the technology is implemented to ultimately classify the responsiveness of documents. The difference between legacy TAR and TAR with advanced AI lies in the additional capabilities opened up by advanced AI subsets outlined to classify data. The supervised machine learning behind traditional TAR implementation uses raw statistics and text analyzation to drive "feature extraction" (i.e., whatever the machine is trying to predict). However, when an ediscovery tool combines that statistical prediction with deep learning and NLP, it can first learn a language model that is then used to make much more accurate classifications – because the language model enables the tool to understand the meaning of words in the context of others. For example, with traditional TAR, the word "train" in the phrases "I am going to

Articles in this issue

view archives of Marketing - TAR + Advanced AI: The Future is Now