Internet. Email. Wireless. Smartphones. Social media. Technology has changed rapidly in the last few decades. Technology’s evolution has increased exponentially with the rise of Artificial Intelligence (AI) in recent years.
In simple terms, AI refers to computer programs that can perform tasks typically done by a human: decision-making, pattern recognition, and natural language processing. Generative AI programs (ChatGPT, Claude, Gemini) can produce content such as images, code, and text.
Truthfully, AI is not new. The annoying paperclip guy who popped up in your Word document in 1995? That was AI. So were your curated book recommendations from Amazon, music recommendations from iTunes, and your Netflix suggested watch list.
The change came in 2022, when chat-based models emerged putting you in the driver’s seat. Now, a “prompt” instructs the AI on what it is supposed to do – draft a document, code a website, or create an image.
To make the best use of AI in your practice, you should understand what it really is and the potential benefits and risks. With that knowledge, you can choose the right AI model for your firm. A written AI Use Policy will solidify those findings and give everyone in the firm the information they need to use AI the right way.
Understanding AI
AI systems are tools in your practice that can help automate routine office tasks, review contracts for risks, or scan and analyze thousands of records in seconds. Having a general understanding of how the technology works is the key to proper use. This enables attorneys to accurately assess whether an AI program works within their ethical and professional obligations.
Token
Tokens represent the fundamental building blocks of data in AI models. These can be words, syllables, pixels, or other predefined units. In general, tokens consist roughly of 4 characters.
Processing the relationships between tokens is key to AI models learning to predict, generate, and reason. Training datasets can include over a billion or trillion tokens. The speed in which the AI model learns and responds depends on how quickly these tokens are processed.
Tokens fall into different categories. Input tokens refer to the characters processed in your request. Output tokens are those generated in the AI’s response. Reasoning tokens are internal reasoning steps before the output is produced. Cached tokens are reused tokens in a conversation history.
Think of tokens as a way to monitor usage – similar to mobile data. In general, the more tokens used means more thinking, which equals more expensive and longer time to generate.
Prompt
In general, a prompt is the question or instructions given to the AI tool so that it will generate output. Parameters include what you want to include or exclude, tone, output format, and uploaded examples.
A well-written prompt produces the best results. Effective prompts are clear, specific, and targeted. The idea is to fully identify the problem you want to solve.
Large Language Models
Large Language Models (LLMs) train on massive datasets to generate human-like language. LLMs identify relationships between words and phrases in a statistical context. LLMs consider text as tokens, and the placement of the next word in a paragraph is a calculation of the probability of the next token.
LLMs produce text that resembles human speech or writing. Most well-known AI programs fall into this category of AI.
Machine Learning
Machine learning involves AI programs improving performance by learning from experience instead of explicit programming. The goal is to achieve self-learning by training the program on data to detect patterns and improve performance based on that analysis.
In general, the program will calculate probabilities and then recognize patterns and relationships in the data, building a decision model to create output.
Bias and Variance
How an AI model was trained can significantly impact the results it produces. The key metrics to consider when evaluating a model’s reliability are bias and variance.
Bias refers to the tendency of AI to skew results based on the data the model was trained on. This may result from limited training data or the developer’s own bias. However, biased input can lead to biased results.
Variance, in contrast, describes the change in a model’s results when there are changes in data. Models can be sensitive to fluctuations in the training data and typically learn irrelevant details instead of the underlying patterns. These complex models often struggle with new data.
The best results come from a program with low bias and low variance.
Hallucination
AI hallucinations are a tendency to fabricate an answer when it can’t find one. What causes hallucinations? In general, it’s the nature of machine learning. AI learns based on pattern recognition, so outputs are based on the probability of being true.
AI hallucinations often result from bias and variance. Often training data includes both accurate and inaccurate data, as well as social biases. Consider that one of the top sources cited on ChatGPT is Reddit, a social media forum platform, a stark reminder that AI cannot differentiate between fact and opinion.
The Benefits of AI
Even though AI sounds complex and scary, properly reviewed AI can be a great benefit for your firm. AI programs can increase efficiency and streamline processes, allowing you to focus on more complex issues and clients.
Legal Research
AI can search massive databases in a relatively short time. This provides you with extensive case law, statutes, and regulations without hours of manual searching through individual records. AI’s pattern recognition capabilities can identify related issues that could otherwise be missed. AI can also summarize the materials for insights to build your case.
Document Review
Similarly, AI can scan documents and efficiently classify, tag, or summarize them. AI can identify specific clauses or inconsistencies between versions. This is especially helpful when reviewing contracts where risk analysis and regulatory compliance are needed.
Analysis
AI is a predictive tool. That means AI tools excel at assessing outcomes based on historical data. They can also estimate settlement values or identify judicial tendencies on legal issues.
Client Intake
AI tools can automate the intake process by sending forms to clients or scheduling appointments. AI can also improve the client’s experience by explaining what the client can expect during their first visit.
Client Communications
In addition to enhancing the intake process, AI can improve communication and responsiveness. AI’s document-drafting capabilities can transform legalese into language the client can understand. AI programs can also identify follow-up dates and action items from communications, making sure deadlines are added to the calendar.
Risks and Ethical Considerations with AI
While AI is a useful tool and can improve your practice, the abundance of AI tool options is complicated and not at all equal. It is important to recognize the limitations and risks associated with AI.
Knowing what to look for will help you maintain your professional and ethical obligations.
Confidentiality
Many AI tools are continually learning and training from user data and prompts. This means that when choosing an AI tool, you need to verify that your data – most importantly, your clients’ data – will be used only for the specific purpose you provided.
Pay close attention to the user agreement regarding the use of data in this regard. An AI Vendor Checklist is included with this guide.
Accuracy
As described previously, AI can generate false results. There is a database that has identified 1,785 cases with hallucinated content so far – and the list is growing daily.
Lawyers are responsible for any errors in the documents produced. Any AI-produced materials should be thoroughly reviewed and verified before being considered a final draft. In addition, citation verification software can flag citations that are suspicious or unverifiable before filing.
Content Ownership
AI generates output based on training information. Much of this existing data is copyrighted material. Using an AI-produced blog post, website copy, or report could be infringement.
A better practice is to draft your own version and have an AI tool edit it. For example, an AI tool can change a formal tone to a more conversational one or rewrite website text to be more user-friendly.
Informed Consent
Your client should understand the use of AI in their case. The attorney should be transparent about the role of AI in the case and acquire informed consent.
To communicate your firm’s use of AI, a “Use of Artificial Intelligence” clause can be added to the engagement letter or Terms of Representation. Sample language is included with this guide.
Costs
Part of informed consent is understanding the fees associated with the use of AI. Legal fees must be fair and reasonable and appropriately reflect the work completed.
If the use of AI changes the costs of the representation, clients must be informed. This could mean that AI cuts the time a task takes in half, resulting in significant savings for the client on an hourly fee contract. It may also mean that the attorney needs to disclose the fee rate for AI usage if they have a pay-per-token license fee that will be passed on to the client.
Find the Right Tools
One mistake often made is purchasing tools without a clear plan for how they fit into processes. But now that you have a better understanding of AI, you can safely find tools that fit within your practice. Here are some considerations before you choose an AI tool:
Consider a specific scenario and which AI tool could be used to solve that problem, along with the desired outcome. Establish benchmarks to measure effectiveness.
Evaluate how tools integrate with your existing systems. Depending on how you are using them, this may complicate procedures or protect client data.
Review the user agreement for confidentiality and data security risks.
Determine budget impact and future costs.
Develop a strategy for implementation to minimize disruption and resistance.
Look at existing policies and procedures and determine where AI fits or what needs to be adjusted, or if new policies need to be created. Tips for a written AI policy are included.
To assist you with selecting the right software, an AI Vendor Checklist is included with this guide.
Have a Written Policy
To ensure your firm is uniform in its approach to AI, it’s important to have a written AI Use Policy. This gives clear direction on what tools your firm allows and when they can and cannot be used.
Provide a checklist for AI-generated content that requires multiple reviews of sources and citations.
Identify acceptable AI tools and their use cases.
Designate someone to monitor issues and limitations with AI programs and update the team with best practices or issue alerts.
Establish access control parameters so that unauthorized users cannot use AI to view data that they normally would be restricted from seeing.
Ensure that everyone who uses AI has the proper skills and knowledge to use it correctly.
A sample AI Use Policy is included with this guide.
Conclusion
AI presents a new frontier that can make the practice of law easier. The best rule of thumb is to consider AI a powerful assistant, not a replacement for your legal expertise or judgment.
Successfully implementing AI, or any new technology, requires understanding how it best serves your practice, its potential pitfalls, and how to mitigate those ethical risks. An AI Use Policy for your firm will make sure everyone maintains a high standard of care.
