When a law firm evaluates a legal AI tool, the questions are\u00a0almost always\u00a0about the things you can see in a demo. How fast is it? Do the citations look right? Can it draft something usable in under a minute? These are fair questions but not what I would lead with. I want to know if this tool will tell me when it\u00a0doesn\u2019t\u00a0know the answer.\u00a0\u00a0<\/p>\n
An LLM is a brain, and\u00a0almost every\u00a0legal AI tool on the market today is built on top of a general model trained on the open web.\u00a0AI has a long memory and\u00a0an earnest desire to please.\u00a0The moment\u00a0it\u00a0can\u2019t\u00a0find an answer in the material\u00a0you\u2019ve\u00a0provided,\u00a0the model\u00a0performs exactly as trained,\u00a0reaching\u00a0back\u00a0through\u00a0everything else it knows\u00a0in effort\u00a0to help.\u00a0The open web is full of material not rooted in fact, or in other words made up, and AI\u00a0doesn\u2019t\u00a0know the difference.\u00a0\u00a0<\/p>\n
Context is always underestimated\u00a0<\/strong>\u00a0<\/p>\n Most of what you can pull for free online is primary source law.\u00a0Legislation.\u00a0Rules. Official missives from agencies like the IRS.\u00a0Primary source law is\u00a0an\u00a0essential\u00a0component\u00a0of effective\u00a0legal\u00a0AI, but\u00a0taken alone\u00a0it\u2019s\u00a0still like trying to assemble a\u00a01,000 piece\u00a0puzzle without being able to look at the\u00a0image\u00a0on the box for context.\u00a0Secondary source material\u00a0\u2014 the\u00a0treatises, the analysis, the practitioner guidance that explains how a rule\u00a0actually works, why\u00a0it\u2019s\u00a0important\u00a0and what\u00a0it relates to\u00a0\u2014 is that\u00a0bigger picture.\u00a0\u00a0<\/p>\n My go-to example is elections. Search \u201celections\u201d in almost any state and\u00a0you\u2019ll\u00a0get a mountain of provisions, and most of them have nothing to do\u00a0about\u00a0the\u00a0rule\u00a0that prompted your original search.\u00a0Absent\u00a0the larger\u00a0context\u00a0provided by\u00a0authoritative secondary source content, a\u00a0large language\u00a0model has nothing to which it can anchor.\u00a0The result is an answer unmoored\u00a0from any verifiable body of information, such as a\u00a0table of contents,\u00a0drifting\u00a0closer\u00a0and closer\u00a0towards\u00a0potential\u00a0hallucination.\u00a0\u00a0\u00a0\u00a0<\/p>\n Guardrails\u00a0don\u2019t\u00a0stay put\u00a0<\/strong>\u00a0<\/p>\n A new round of testing should\u00a0accompany\u00a0every model update\u00a0and feature release.\u00a0Each\u00a0iteration of\u00a0an LLM\u00a0reasons differently,\u00a0and a guardrail that\u00a0previously\u00a0kept the system anchored to your vetted content\u00a0may\u00a0now have a loophole you never\u00a0anticipated.\u00a0\u00a0<\/p>\n Ask\u00a0the system to build a contract and lay out the\u00a0appropriate\u00a0clauses.\u00a0You can watch to see if the\u00a0eager-to-please\u00a0model will\u00a0out-think your\u00a0guardrails\u00a0in an effort to\u00a0deliver an\u00a0answer.\u00a0\u00a0<\/p>\n An automated test\u00a0will more than likely pass a fault output so long as the\u00a0citations provided appear genuine.\u00a0A\u00a0human\u00a0reviewer needs to\u00a0physically\u00a0click the citation and\u00a0verify\u00a0that it\u00a0does not answer the question\u00a0in order\u00a0to\u00a0catch\u00a0the problem.<\/p>\n Saying \u201cI don\u2019t know\u201d is harder than it sounds\u00a0<\/strong>\u00a0<\/p>\n Nobody likes admitting they\u00a0don\u2019t\u00a0have all the answers.\u00a0Teaching\u00a0a model to say \u201cI don\u2019t know\u201d is\u00a0both a\u00a0matter of training and a statement of purpose. While\u00a0it\u2019s\u00a0true that a\u00a0tool\u00a0which\u00a0sometimes comes back with \u201cI don\u2019t have a source for this\u201d\u00a0may seem less impactful in\u00a0a sales demo, practicing attorneys\u00a0will\u00a0actually\u00a0recognize\u00a0it as solid ground.\u00a0\u00a0<\/p>\n When I practiced law, my greatest fear was\u00a0missing something critical.\u00a0If\u00a0there\u2019s\u00a0no reputable authority that can be cited, that would have been\u00a0genuinely useful for me to know.\u00a0It empowers\u00a0me to begin taking steps in the right direction. On the other hand,\u00a0a\u00a0paragraph\u00a0buoyed by false confidence and\u00a0built from dubious sourcing\u00a0takes much longer to\u00a0untangle.\u00a0\u00a0<\/p>\n Remember that\u00a0transparency is\u00a0a\u00a0vital\u00a0component\u00a0of\u00a0trust.\u00a0Accuracy\u00a0isn\u2019t\u00a0just about pinpointing\u00a0the\u00a0correct\u00a0answer.\u00a0It\u2019s\u00a0also about being able to acknowledge when there is no answer to be found.\u00a0Establishing\u00a0that level of visibility as\u00a0table stakes is essential to\u00a0building a strong future for legal AI.\u00a0<\/p>\n Nicole Stone is Director of AI & Agentic Solutions Product Management at Wolters Kluwer Legal & Regulatory U.S., where she leads product strategy and development for digital legal content and technology solutions. With over 22 years of experience in legal technology and a background as a practicing attorney, she focuses on integrating emerging technologies, including generative AI, into products that serve legal professionals.<\/strong><\/em><\/p>\n \n The post The One Thing You\u00a0Didn\u2019t\u00a0Think To Look For In Legal AI\u00a0<\/a> appeared first on Above the Law<\/a>.<\/p>\n When a law firm evaluates a legal AI tool, the questions are\u00a0almost always\u00a0about the things you can see in a demo. How fast is it? Do the citations look right? Can it draft something usable in under a minute? These are fair questions but not what I would lead with. I want to know if this tool will tell me when it\u00a0doesn\u2019t\u00a0know the answer.\u00a0\u00a0<\/p>\n An LLM is a brain, and\u00a0almost every\u00a0legal AI tool on the market today is built on top of a general model trained on the open web.\u00a0AI has a long memory and\u00a0an earnest desire to please.\u00a0The moment\u00a0it\u00a0can\u2019t\u00a0find an answer in the material\u00a0you\u2019ve\u00a0provided,\u00a0the model\u00a0performs exactly as trained,\u00a0reaching\u00a0back\u00a0through\u00a0everything else it knows\u00a0in effort\u00a0to help.\u00a0The open web is full of material not rooted in fact, or in other words made up, and AI\u00a0doesn\u2019t\u00a0know the difference.\u00a0\u00a0<\/p>\n Context is always underestimated\u00a0<\/strong>\u00a0<\/p>\n Most of what you can pull for free online is primary source law.\u00a0Legislation.\u00a0Rules. Official missives from agencies like the IRS.\u00a0Primary source law is\u00a0an\u00a0essential\u00a0component\u00a0of effective\u00a0legal\u00a0AI, but\u00a0taken alone\u00a0it\u2019s\u00a0still like trying to assemble a\u00a01,000 piece\u00a0puzzle without being able to look at the\u00a0image\u00a0on the box for context.\u00a0Secondary source material\u00a0\u2014 the\u00a0treatises, the analysis, the practitioner guidance that explains how a rule\u00a0actually works, why\u00a0it\u2019s\u00a0important\u00a0and what\u00a0it relates to\u00a0\u2014 is that\u00a0bigger picture.\u00a0\u00a0<\/p>\n My go-to example is elections. Search \u201celections\u201d in almost any state and\u00a0you\u2019ll\u00a0get a mountain of provisions, and most of them have nothing to do\u00a0about\u00a0the\u00a0rule\u00a0that prompted your original search.\u00a0Absent\u00a0the larger\u00a0context\u00a0provided by\u00a0authoritative secondary source content, a\u00a0large language\u00a0model has nothing to which it can anchor.\u00a0The result is an answer unmoored\u00a0from any verifiable body of information, such as a\u00a0table of contents,\u00a0drifting\u00a0closer\u00a0and closer\u00a0towards\u00a0potential\u00a0hallucination.\u00a0\u00a0\u00a0\u00a0<\/p>\n Guardrails\u00a0don\u2019t\u00a0stay put\u00a0<\/strong>\u00a0<\/p>\n A new round of testing should\u00a0accompany\u00a0every model update\u00a0and feature release.\u00a0Each\u00a0iteration of\u00a0an LLM\u00a0reasons differently,\u00a0and a guardrail that\u00a0previously\u00a0kept the system anchored to your vetted content\u00a0may\u00a0now have a loophole you never\u00a0anticipated.\u00a0\u00a0<\/p>\n Ask\u00a0the system to build a contract and lay out the\u00a0appropriate\u00a0clauses.\u00a0You can watch to see if the\u00a0eager-to-please\u00a0model will\u00a0out-think your\u00a0guardrails\u00a0in an effort to\u00a0deliver an\u00a0answer.\u00a0\u00a0<\/p>\n An automated test\u00a0will more than likely pass a fault output so long as the\u00a0citations provided appear genuine.\u00a0A\u00a0human\u00a0reviewer needs to\u00a0physically\u00a0click the citation and\u00a0verify\u00a0that it\u00a0does not answer the question\u00a0in order\u00a0to\u00a0catch\u00a0the problem.<\/p>\n Saying \u201cI don\u2019t know\u201d is harder than it sounds\u00a0<\/strong>\u00a0<\/p>\n Nobody likes admitting they\u00a0don\u2019t\u00a0have all the answers.\u00a0Teaching\u00a0a model to say \u201cI don\u2019t know\u201d is\u00a0both a\u00a0matter of training and a statement of purpose. While\u00a0it\u2019s\u00a0true that a\u00a0tool\u00a0which\u00a0sometimes comes back with \u201cI don\u2019t have a source for this\u201d\u00a0may seem less impactful in\u00a0a sales demo, practicing attorneys\u00a0will\u00a0actually\u00a0recognize\u00a0it as solid ground.\u00a0\u00a0<\/p>\n When I practiced law, my greatest fear was\u00a0missing something critical.\u00a0If\u00a0there\u2019s\u00a0no reputable authority that can be cited, that would have been\u00a0genuinely useful for me to know.\u00a0It empowers\u00a0me to begin taking steps in the right direction. On the other hand,\u00a0a\u00a0paragraph\u00a0buoyed by false confidence and\u00a0built from dubious sourcing\u00a0takes much longer to\u00a0untangle.\u00a0\u00a0<\/p>\n Remember that\u00a0transparency is\u00a0a\u00a0vital\u00a0component\u00a0of\u00a0trust.\u00a0Accuracy\u00a0isn\u2019t\u00a0just about pinpointing\u00a0the\u00a0correct\u00a0answer.\u00a0It\u2019s\u00a0also about being able to acknowledge when there is no answer to be found.\u00a0Establishing\u00a0that level of visibility as\u00a0table stakes is essential to\u00a0building a strong future for legal AI.\u00a0<\/p>\n Nicole Stone is Director of AI & Agentic Solutions Product Management at Wolters Kluwer Legal & Regulatory U.S., where she leads product strategy and development for digital legal content and technology solutions. With over 22 years of experience in legal technology and a background as a practicing attorney, she focuses on integrating emerging technologies, including generative AI, into products that serve legal professionals.<\/strong><\/em><\/p>\n \n The post The One Thing You\u00a0Didn\u2019t\u00a0Think To Look For In Legal AI\u00a0<\/a> appeared first on Above the Law<\/a>.<\/p>\n","protected":false},"excerpt":{"rendered":" When a law firm evaluates a legal AI tool, the questions are\u00a0almost always\u00a0about the things you can see in a demo. How fast is it? Do the citations look right? Can it draft something usable in under a minute? These are fair questions but not what I would lead with. I want to know if […]<\/p>\n","protected":false},"author":3,"featured_media":0,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"_et_pb_use_builder":"","_et_pb_old_content":"","_et_gb_content_width":"","_jetpack_memberships_contains_paid_content":false,"footnotes":""},"categories":[16],"tags":[],"class_list":["post-157697","post","type-post","status-publish","format-standard","hentry","category-above_the_law"],"jetpack_featured_media_url":"","jetpack_sharing_enabled":true,"_links":{"self":[{"href":"https:\/\/xira.com\/p\/wp-json\/wp\/v2\/posts\/157697","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/xira.com\/p\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/xira.com\/p\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/xira.com\/p\/wp-json\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/xira.com\/p\/wp-json\/wp\/v2\/comments?post=157697"}],"version-history":[{"count":0,"href":"https:\/\/xira.com\/p\/wp-json\/wp\/v2\/posts\/157697\/revisions"}],"wp:attachment":[{"href":"https:\/\/xira.com\/p\/wp-json\/wp\/v2\/media?parent=157697"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/xira.com\/p\/wp-json\/wp\/v2\/categories?post=157697"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/xira.com\/p\/wp-json\/wp\/v2\/tags?post=157697"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
\n
\n