The story is all too familiar by now. A lawyer uses ChatGPT to draft a brief, gets back a polished product that looks professional and which appears to cite relevant cases. Only problem? The cases are fabricated or don\u2019t stand for the proposition for which they are cited. The lawyer then gets caught and called out.<\/p>\n
The standard uproar commences. Why didn\u2019t they just read the cases? Everybody knows you have to read every case, every time. Stupid, lazy lawyers.<\/p>\n
But maybe we ought to ask why this keeps happening, especially when the propensity of LLMs to make these kinds of mistakes and hallucinate is well known. As is the need to read cited cases.<\/p>\n
So, What\u2019s Going on Here?<\/strong><\/p>\n Certainly, it can\u2019t be disputed that a lawyer filing a paper with the court needs to read the cases and make sure the cites are correct. But a number of factors work to place pressure on lawyers and legal professionals that can lead and tempt them to not do what they are supposed to do. And these are often ignored by the critics who just say read the cases and there is something immoral about lawyers who don\u2019t. The reality on the ground is not always that simple.<\/p>\n The ability of LLMs to shortcut work is having its impact on expectations and what clients will pay for and, in turn, even what senior partners may demand of associates. And that impact may lead to even greater hallucinations and inaccuracy problems down the road if they go unrecognized.<\/p>\n The Traditional Way of Working<\/strong><\/p>\n As I previously reported, the expected work process for such things as legal research may be changing. That traditional work process was to evaluate the problem and issues, begin to read cases, then read cited cases, then search for some more and related cases. Then refine the search and look other places. Then even read dissents. Review things like the development of the legal theory over time. Understand context and nuance. All of this takes time. It\u2019s tedious.<\/p>\n But with LLMs, this work can be short circuited. An \u201canswer\u201d can be found in seconds. The result? Clients may take a dim view of paying for the old-fashioned workflow when they perceive work can be done in a fraction of the time.<\/p>\n But in doing so, the nuance and context is lost. It becomes easy to miss something that\u2019s not quite right.<\/p>\n Can I Bill For That?<\/strong><\/p>\n Here\u2019s where economics makes things a little complicated. It\u2019s easy to say when the LLM gives you some case cites you still have to read the cases in detail, right? That\u2019s probably true in a perfect world. But most lawyers live in a billable hour world. They live in a world governed by not just the number of hours you bill but what portion of those hours can be collected.<\/p>\n So, where does that leave you as an associate or for that matter, a billing partner? What if the client won\u2019t pay for the kind of studied research that you think should be done? You have billable hour quotas to meet.\u00a0 Your advancement and compensation depend on billables and collections.<\/p>\n If you do the extra work for which the client won\u2019t pay, you have wasted time that could otherwise be spent on billable and collectable time. For partners, their profitability index takes a hit. For associates, what happens when a senior partner implicitly (or perhaps explicitly) says we can\u2019t bill for that, so don\u2019t do it? \u00a0<\/p>\n What Is Your Duty in the Age of AI?<\/strong><\/p>\n And exactly what is the extent of your duty to check citations and other source materials? We have all used string cites to support some clear principles. The standard for summary judgment for example: in pre-AI days, I would routinely cite cases for the standard I pulled from someplace without reading the entire case. Or I might cite a case for a proposition and then look for other cases that stood for the same principle without perhaps reading those cases as carefully as I should have. And now in the time of GenAI, I would be faced with the worry that the work I perhaps should have done might not be billable.<\/p>\n There are also delegation issues. Let\u2019s say I ask an associate or paralegal to run down the cite and they either use GenAI or don\u2019t carefully check the cases. As a partner, do I need to tell them to do work that may be necessary but may not be billable and collectible?<\/p>\n What happens when you are local counsel and are sent a pleading to sign that contains cites and information? Do you have to check the cites even though it\u2019s doubtful you can get paid for it?<\/p>\n These are all tough questions we need to consider as the tools become better, and their use becomes more demanded and expected.<\/p>\n It\u2019s So Easy<\/strong><\/p>\n Add to these pressures the fact that using these tools to avoid tedious, time-consuming tasks is so easy and tempting. Indeed, I suspect most attorney errors thus far have stemmed not from ignorance of the risks, but because the tools usually work well and require no technical expertise. You don\u2019t need IT to help. You don\u2019t need to consult a senior partner. You don\u2019t have to read cases until the wee hours. Easy-peasy. \u00a0And that\u2019s a danger.<\/p>\n What Needs to Happen<\/strong><\/p>\n I\u2019m certainly not saying that we can\u2019t or shouldn\u2019t check citations. But we do need to recognize the potential pressures being put on lawyers by clients, coworkers, and partners. We need to recognize danger areas and be sure we educate everyone, including clients looking to lower legal bills by insisting on GenAI being used without robust checking. Law firms need to make clear what is expected when some of the work can\u2019t be billed or collected but is nevertheless necessary. And we need clear standards from courts and bar associations on what our duties are.<\/p>\n It\u2019s a brave new world. Until we acknowledge these economic realities and adjust our expectations and guidelines to ensure hallucinations aren\u2019t being adopted, we\u2019ll keep seeing more headlines about \u201cstupid, lazy lawyers\u201d while missing the real systemic issues at play.<\/p>\n Stephen Embry is a lawyer, speaker, blogger, and writer. He publishes\u00a0TechLaw Crossroads<\/a>, a blog devoted to the examination of the tension between technology, the law, and the practice of law<\/strong><\/em>.<\/p>\n The post Hallucinations Here, Hallucinations There, Hallucinations Everywhere: Why Do Lawyers Keep Doing It?<\/a> appeared first on Above the Law<\/a>.<\/p>\n The story is all too familiar by now. A lawyer uses ChatGPT to draft a brief, gets back a polished product that looks professional and which appears to cite relevant cases. Only problem? The cases are fabricated or don\u2019t stand for the proposition for which they are cited. The lawyer then gets caught and called out.<\/p>\n The standard uproar commences. Why didn\u2019t they just read the cases? Everybody knows you have to read every case, every time. Stupid, lazy lawyers.<\/p>\n But maybe we ought to ask why this keeps happening, especially when the propensity of LLMs to make these kinds of mistakes and hallucinate is well known. As is the need to read cited cases.<\/p>\n So, What\u2019s Going on Here?<\/strong><\/p>\n Certainly, it can\u2019t be disputed that a lawyer filing a paper with the court needs to read the cases and make sure the cites are correct. But a number of factors work to place pressure on lawyers and legal professionals that can lead and tempt them to not do what they are supposed to do. And these are often ignored by the critics who just say read the cases and there is something immoral about lawyers who don\u2019t. The reality on the ground is not always that simple.<\/p>\n The ability of LLMs to shortcut work is having its impact on expectations and what clients will pay for and, in turn, even what senior partners may demand of associates. And that impact may lead to even greater hallucinations and inaccuracy problems down the road if they go unrecognized.<\/p>\n The Traditional Way of Working<\/strong><\/p>\n As I previously reported, the expected work process for such things as legal research may be changing. That traditional work process was to evaluate the problem and issues, begin to read cases, then read cited cases, then search for some more and related cases. Then refine the search and look other places. Then even read dissents. Review things like the development of the legal theory over time. Understand context and nuance. All of this takes time. It\u2019s tedious.<\/p>\n But with LLMs, this work can be short circuited. An \u201canswer\u201d can be found in seconds. The result? Clients may take a dim view of paying for the old-fashioned workflow when they perceive work can be done in a fraction of the time.<\/p>\n But in doing so, the nuance and context is lost. It becomes easy to miss something that\u2019s not quite right.<\/p>\n Can I Bill For That?<\/strong><\/p>\n Here\u2019s where economics makes things a little complicated. It\u2019s easy to say when the LLM gives you some case cites you still have to read the cases in detail, right? That\u2019s probably true in a perfect world. But most lawyers live in a billable hour world. They live in a world governed by not just the number of hours you bill but what portion of those hours can be collected.<\/p>\n So, where does that leave you as an associate or for that matter, a billing partner? What if the client won\u2019t pay for the kind of studied research that you think should be done? You have billable hour quotas to meet.\u00a0 Your advancement and compensation depend on billables and collections.<\/p>\n If you do the extra work for which the client won\u2019t pay, you have wasted time that could otherwise be spent on billable and collectable time. For partners, their profitability index takes a hit. For associates, what happens when a senior partner implicitly (or perhaps explicitly) says we can\u2019t bill for that, so don\u2019t do it? \u00a0<\/p>\n What Is Your Duty in the Age of AI?<\/strong><\/p>\n And exactly what is the extent of your duty to check citations and other source materials? We have all used string cites to support some clear principles. The standard for summary judgment for example: in pre-AI days, I would routinely cite cases for the standard I pulled from someplace without reading the entire case. Or I might cite a case for a proposition and then look for other cases that stood for the same principle without perhaps reading those cases as carefully as I should have. And now in the time of GenAI, I would be faced with the worry that the work I perhaps should have done might not be billable.<\/p>\n There are also delegation issues. Let\u2019s say I ask an associate or paralegal to run down the cite and they either use GenAI or don\u2019t carefully check the cases. As a partner, do I need to tell them to do work that may be necessary but may not be billable and collectible?<\/p>\n What happens when you are local counsel and are sent a pleading to sign that contains cites and information? Do you have to check the cites even though it\u2019s doubtful you can get paid for it?<\/p>\n These are all tough questions we need to consider as the tools become better, and their use becomes more demanded and expected.<\/p>\n It\u2019s So Easy<\/strong><\/p>\n Add to these pressures the fact that using these tools to avoid tedious, time-consuming tasks is so easy and tempting. Indeed, I suspect most attorney errors thus far have stemmed not from ignorance of the risks, but because the tools usually work well and require no technical expertise. You don\u2019t need IT to help. You don\u2019t need to consult a senior partner. You don\u2019t have to read cases until the wee hours. Easy-peasy. \u00a0And that\u2019s a danger.<\/p>\n What Needs to Happen<\/strong><\/p>\n I\u2019m certainly not saying that we can\u2019t or shouldn\u2019t check citations. But we do need to recognize the potential pressures being put on lawyers by clients, coworkers, and partners. We need to recognize danger areas and be sure we educate everyone, including clients looking to lower legal bills by insisting on GenAI being used without robust checking. Law firms need to make clear what is expected when some of the work can\u2019t be billed or collected but is nevertheless necessary. And we need clear standards from courts and bar associations on what our duties are.<\/p>\n It\u2019s a brave new world. Until we acknowledge these economic realities and adjust our expectations and guidelines to ensure hallucinations aren\u2019t being adopted, we\u2019ll keep seeing more headlines about \u201cstupid, lazy lawyers\u201d while missing the real systemic issues at play.<\/p>\n Stephen Embry is a lawyer, speaker, blogger, and writer. He publishes\u00a0TechLaw Crossroads<\/a>, a blog devoted to the examination of the tension between technology, the law, and the practice of law<\/strong><\/em>.<\/p>\n The post Hallucinations Here, Hallucinations There, Hallucinations Everywhere: Why Do Lawyers Keep Doing It?<\/a> appeared first on Above the Law<\/a>.<\/p>\n","protected":false},"excerpt":{"rendered":" The story is all too familiar by now. A lawyer uses ChatGPT to draft a brief, gets back a polished product that looks professional and which appears to cite relevant cases. Only problem? The cases are fabricated or don\u2019t stand for the proposition for which they are cited. The lawyer then gets caught and called […]<\/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-128893","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\/128893","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=128893"}],"version-history":[{"count":0,"href":"https:\/\/xira.com\/p\/wp-json\/wp\/v2\/posts\/128893\/revisions"}],"wp:attachment":[{"href":"https:\/\/xira.com\/p\/wp-json\/wp\/v2\/media?parent=128893"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/xira.com\/p\/wp-json\/wp\/v2\/categories?post=128893"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/xira.com\/p\/wp-json\/wp\/v2\/tags?post=128893"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
\n
\n