Chapter 4: Exercises

Below are five exercises designed to deepen your understanding of specialized legal AI tools discussed in Chapter 4.

Each exercise is intended for the free version of Claude, where you can use Artifacts (the platform’s feature for displaying tables and other structured data) to compare outputs, perform simple data analyses, and visually organize your findings.

Purpose: To explore and visually compare different AI products (CoCounsel, Spellbook, AI.Law, Alexi) based on key criteria from the chapter.

Instructions:

  1. Initial Prompt to Claude: 

    Create a table comparing CoCounsel, Spellbook, AI.Law, and Alexi.com. Include columns for:  
- Data Security (features)  
- Domain-Specific Training  
- Ethical Compliance Features  
- Primary Use Case(s)  
- Unique Selling Point
Provide a brief explanation in each cell, focusing on how each tool addresses these areas.

  2. Refine the Prompt (if needed):

    • If the generated table is too simplistic, add:
      Please include at least two bullet points in each cell explaining how these tools handle data security and domain expertise. Also, note any disclaimers they provide regarding attorney oversight.

Reflection:
Evaluate the table. Which tool seems strongest in data security? Which one highlights ethical compliance most explicitly? Note any cell where Claude’s summary might be missing details. Would you trust this AI’s output as-is, or do you see areas for further research before relying on it?

Exercise 2: Pilot Testing a Hypothetical AI Tool

Purpose: To practice how you’d start small and test a new AI tool on a limited task, as recommended in the chapter’s “Practice Pointer.”

Instructions:

  1. Scenario Prompt: 

    Imagine your law firm is considering a pilot project with a new legal AI platform for contract review. Construct a short plan describing:  
- Which documents or client matters you’d test the platform on  
- What success metrics you’ll use (e.g., accuracy, time savings)  
- How you’ll handle data security during the pilot

  2. Using Claude’s Artifacts:

    • Ask Claude to format your plan as a table or bulleted list, comparing “Before AI” vs. “With AI.”
    • Example columns might include Task, Old Process (Manual Steps), New Process (AI-Assisted), Expected Efficiency Gains.

Reflection:
Does the plan align with the core evaluation criteria (data security, ethical oversight, domain-specific training)? Would you require any disclaimers or client consent before proceeding with this pilot?

Exercise 3: Exploring a Specialty Use Case—eDiscovery

Purpose: To see how you might apply specialized AI to large-scale document review in litigation or regulatory matters.

Instructions:

  1. Initial Prompt to Claude: 

    Pretend you are an AI eDiscovery platform specialized in predictive coding. Here is a brief description of 1,000 documents in an M&A due diligence:  
- 250 refer to intellectual property concerns  
- 500 relate to routine vendor transactions  
- 250 contain emails with the word ‘confidential’  
Tell me how you would categorize and prioritize these for legal review.

  2. Focus on Visual Output:

    • Ask Claude to display the prioritization in a table or chart, indicating which documents are ‘High Priority,’ ‘Medium Priority,’ or ‘Low Priority.’
    • Prompt: Show me a chart or table that ranks these three document categories by potential risk. Also list possible next steps for an attorney to verify your recommendations.

Reflection:
Check if the AI’s ranking logic aligns with your understanding of risk (e.g., IP concerns and confidentiality mentions might be higher priority). List any reasons you would override or modify the AI’s classification.

Exercise 4: Drafting and Summarizing Clauses with Spellbook-Like Functionality

Purpose: To analyze how an AI tool might handle contract drafting or summarizing key clauses, simulating Spellbook’s workflow.

Instructions:

  1. Scenario: You have a Commercial Lease Agreement that includes a 1-page excerpt covering rent escalation, maintenance obligations, and termination clauses.
  2. Prompt to Claude:
    Summarize the commercial lease excerpt into three concise bullet points, focusing on potential risks for the tenant. Then propose a revised rent escalation clause that’s more tenant-friendly.
  3. Refinement:
    • If you want to add complexity, say: Rewrite the summary to include possible negotiation points. Present them in a table with columns labeled *Original Clause*, *Risk for Tenant*, and *Recommended Revision*.

Reflection:
Are the revised clauses sufficiently detailed? Would you still do a deeper review to ensure compliance with local law or to check if the AI missed any unusual language?

Exercise 5: Ethical Oversight and Bias Analysis

Purpose: To evaluate how attorneys might handle potential bias or misunderstanding in specialized AI outputs.

Instructions:

  1. Initial Prompt:
    Explain how a legal AI tool could perpetuate bias or overlook crucial context when classifying documents related to workplace discrimination claims. Propose steps a human attorney should take to mitigate these risks.
  2. Table Approach:
    • Request that Claude arrange the answer in a table with columns: Potential Bias or Oversight, Example Scenario, and Attorney Action Steps.
  3. Refinement:
    • If the AI’s examples are too broad, ask for specific biases (e.g., certain terms or linguistic patterns it might misread) and concrete mitigation strategies (like random sampling or data audits).

Reflection:
Compare the potential biases with the ethical obligations covered in Chapter 4. How do you, as an attorney, remain the final checkpoint to ensure fairness and legality?