Understanding AI is not simply optional for lawyers, it may in fact be mandatory in certain cases.

There
needs to be a professional rethinking of the rules for lawyers to address the use of AI. The American Bar
Association model rules of professional conduct require that lawyers be informed and up to date about
the benefits and risks associated with AI Technology, lawyers should see the outputs delivered by AI.
This may include lawyers’ involvement in testing AI models, to monitor margins of error and assert
monitor control, duty of confidentiality is a paramount especially with respect to the data that is being
given to the companies that are training and testing the AI models, conflict of interest issues around
which some novel issues could also arise. For example if one has the same AI provider that supplies AI
output or recommendations to two opposing parties in the same matter. A rule regarding AI calls in
developers of AI to ensure the products and systems are subject to human oversight and control and to
ensure transparency. The challenge in AI, in any application with this customization to the industry to
the purpose for which AI is being built, needs to be addressed. We need to look at how AI can be
customized for the application and practice of law. There are a couple of components involved in this
matter, the first one is feasibility analysis, checking the technical diligence of AI to assess whether its
possible to build AI to solve a particular legal problem. There needs to be sufficient data to train the AI
model and that data needs to be structured, labelled and organised, then there is the value add analysis
which is the business diligence that is needed to see whether the AI solution had enough value and
whether it assisted the legal and business activity properly. Then there needs to be the measurability
analysis, which means that the AI model needs to be sufficiently accurate. An example of legal diligence
of AI is suppose a tech company or a law firm is building an AI model to detect change of control
provisions in thousands of contracts to assist in M&A Due Diligence. To train the AI model, the company
would feed the system with various data parameters, multiple examples and forms of change of control
provisions, contracts in which it appears, different places in the contract where it appears, the company
would leave some of these contracts out of the training data set and use them to test. When testing, it
would be important to consider not just the false positives but also the false negatives to when we
measure, it would be important to see whether AI missed any contracts with change of control clause.
There are two important points in this matter. First- training data set should be balanced, if an AI model
is being built to predict if a court would find a violation, if the model is being fed with 95% violation
cases and only 5% non violation cases, you are securing a predictive model. So the balanced data set
should not necessarily be 50-50 but should reflect the split in reality. Second- the comparision for
success of the measurement of AI should not necessarily be a coin toss, say, about 50%. So for example
if a court or a judge rules in favour of the plaintiffs 60% of the time, then you could guess that the
decision in favour of plaintiff would have 60 %accuracy, even without using AI model, so AI has to be
measured at least against already existing odds. Now there are increasing number of AI applications in
NLP,natural language processing. There is a spectrum of AI NLP approaches to legal language as well for
the purpose of text solidification, knowledge representation, argument mining, information extraction,
retrieval and so forth. AI has been used to detect potentially unfair contract revisions in contracts like
terms used in online platforms. Machine learning is an AI powered tool named Claudette which is an
abbreviation of clause detector tool. The tool processes and tags clauses for type and then for degrees
of fairness, then the tools is used to classify the provision by using NLP through machine learning. Using
this logic and application, AI NLP can be used for a bunch of other options for example to score
provisions, to detect deviation from standard market clauses, to rank friendliness of the provision to a
certain party of the contract. AI is also currently being used to sift through communications evidence in
investigations of alleged white collar crimes, fcpa violations, Ai is also trained to detect potentially
fraudulant communications not only in emails but also in text messages, so that takes time to train the
AI model to do all of these things. Use of AI to alert the companies and law firms of any regulatory
developments is also being done. AI is also essentially scanning online content and as soon as some
regulatory development occurs in a particular area, it sends alerts. AI can also be taught to identify legal
concepts but without understanding them to identify them and not merely to recognize words or terms
but more than that. This will empower more elaborate and targeted legal research that extend beyond
simple keyword searches. NLP of judgement related texts and extra textual factors are two types of
predictive AI models in judicial analytics which will help in the judicial analytics processes. Some skills
that the future lawyers must possess as they work alongside AI in future are mentioned in the picture
below.