迈向未来：人工智能与律师

How generative AI will help lawyersimprove legal service delivery Report Technology has become essential to the practice oflaw, but the boundaries of technology’s impact in improving legal services delivery continue to expand.More and more lawyers are using artificial intelligence (AI)-basedtools every day. AI is a well-established part of lawyers’ work inlegal research, contract analysis, e-discovery, and predictive As much as those tools have revolutionized certain types oflegal work, another new generation of AI promises an even moresignificant transformation of legal work and better outcomes forclients. The new generation is sometimes called generative AI, andits power is based on the scale enabled by large language models Generative AI: the latest phasein a rapidly evolving technology It has Ceenhard to ignore the interestand exDJteNent PWer neX HeneratJWe AI GPrNs that haWe reDentlZ reaDhed the NarLet 0ne PG thPse HeneratJWe AIprPdVDts $hat(15 RVJDLlZ tVrned JntP a XPrldXJde phenPNenPn It reaDhed aNJllJPn Vsers Jn less than a NPnth Gaster than anZ Pther pPpVlar PnlJne PGGerJnHJnDlVdJnH InstaHraN 4pPtJGZ Pr 'aDeCPPL 5XP NPnths Jn Jt had reaDhed  ChatGPT sprints to one million users Time it took for selected online services toreach one million users Why the interest? Why have ChatGPT and other forms of generative AI received so much interest?First, ChatGPT is freely available for anyone to test out. By simply enteringprompts or commands, it could seemingly address almost any question orexecute any task that was asked of it. Users discovered that it could effortlessly ChatGPT’s simple format has democratized the public’s access to machinelearning tools. It takes instructions and allows users to respond with furtherdirections in plain conversational language. This capability made it easy for Generative AI has entered the collective legal consciousness quickly. Ina recentsurvey conducted by the Thomson Reuters® Institute, ChatGPT and generativeAI awareness is significantly higher among legal professionals, with 91% of And with that awareness comes the realization of potential use cases for usingthis technology in day-to-day work. How does it work? Why do new forms of generative AI seem to be so effective at reproducing Machine learning systems are not “thinking” systems. They are languageprediction systems. Many machine learning tools are designed to predictmissing words in a sequence. When humans review those predictions foraccuracy, the predictions are given a score for accuracy. The machine uses We have known for some time now that, as the data sets that machinelearning systems are trained on have become larger, their predictions andability to provide accurate answers have improved only modestly. The bigbreakthrough with ChatGPT was that when the training data sets reached Generative AI in legal: risks and The legal profession is in a key position relative to generative AI. First, legal workis primarily centered around words, documents, and data. The work consists offinding, analyzing, and creating texts and documents, which are precisely the sort Second, legal is a field where accuracy and precision are essential. There is littleroom for error when legal processes and decisions can impact parties’ important Legal audiences are clearly impressed by the potential power of generative AI, butalso concerned by the dangers, as seen in the recentThomson Reuters Institute While ChatGPT can deliver answers that sound coherent and accurate, this type ofAI is really just a sentence-completion engine. ChatGPT has little intelligence butis very good at drafting language that sounds like plausible responses to a user’sprompt. But it’s not always accurate. There are numerous examples of ChatGPT Those examples are called “hallucinations” and include thenow-famous caseof a New York lawyerwho submitted a brief supporting a motion to a court thatincluded arguments and citations based on the results of a ChatGPT session. Errors like that present an unacceptable level of risk for legal professionals.So how will lawyers effectively use generative AI while maintaining all their A standard methodology to improve generative AI results in specialized fields iscalled Retrieval Augmented Generation (RAG). In systems that use RAG, the user’sprompts or queries do not pass directly through to the LLM; the question runsfirst as a search against a trusted body of content — for example,verified legalcontent from a legal publisheror trusted documents from the user’s organization. The power of generative AI derives from the size and scale of the original data setsthat the AI models are trained on — but in specialized fields like law, it’s critical that What’s driving the adoption ofAI in the legal profession? A ²perGeDt stPrN³ PG GaDtPrs Js pVshJnH laXZers and leHal PrHanJ[atJPns tPXard HreateradPptJPn PG AICased tPPls Jn theJr XPrL 5he NaJn drJWers PG adPptJPn JnDlVde Tasks centered on documents an