效率归自己,账单归客户Efficiency for the Firm, the Bill for the Client
AI 让律师省了时间,但只让 26% 的律师降了费。剩下 74% 的那部分效率,被谁拿走了,报道没写,因为不需要写——它本来就在默认的位置上。
这就是结构暴力的标准形态:没有施暴者,没有可见的事件,只有一条从没被写进合同的分配规则。工具升级了,产出增加了,多出来的那一块自动流向掌握定价权的一方。加尔通的公式在这里算得很清楚:本可以更便宜的法律服务,和实际收的价钱之间,那个差额就是暴力,只不过它穿着一件叫“市场”的衣服。
真正把这件事钉死的是那组对比:企业客户 34% 拿到降价,个人付费客户 23%。同一个州,同一批律师,同一套 AI 工具,价格差出十一个百分点。Rychner 的解释很诚实——企业客户是回头客,自己也在用 AI,有能力为更低的价格谈判。翻译一下:能不能拿到效率红利,取决于你有没有议价能力,而不是取决于效率有没有发生。没有议价能力的人,AI 省下来的时间不会变成他的钱,只会变成对方的利润。
还有一层更妙的。46% 的律师说客户带着错误或误导性的信息上门,41% 说客户拿来的 AI 生成文件需要返工,近一半的律师说客户用了 AI 之后,他们要花更多时间和精力。客户自己动手省下的那点咨询费,最后以返工的形式重新计费回到账单上。AI 在这里不是降低了法律服务的门槛,是把门槛挪了个位置——让客户先免费干活,再付费纠错。
而这一切发生在法律行业。法律是分配权利的行业,谁能拿到什么样的救济,谁的声音能被写进判决,很大程度上取决于谁付得起。一个连降价都要靠议价能力来争取的行业,它对外输出的“公正”,质地可想而知。第四章说的元暴力在这里不需要现身,它不需要——结构自己会跑。
报道里唯一被点名付代价的是那个维多利亚州律师:因为 AI 生成的假判例被剥夺了主律师执业资格,全国第一个。工具出错,责任落在使用工具的人身上,这很公平。但客户返工的那些小时,没有人为它负责,也不会出现在任何一份报告里。
所以下一次有人告诉你 AI 会带来效率红利,先问一句:红利进谁的口袋。答案通常不在技术文档里,在谁有资格谈价这件事上。
AI saved lawyers time. Only 26% of them passed any of it on as lower fees. Where the other 74% went, the report doesn't say, because it doesn't need to — it stayed exactly where it already was.
This is structural violence in its standard form: no perpetrator, no visible event, just a distribution rule that was never written into any contract. The tool improves, output rises, and the surplus flows automatically to whoever holds pricing power. Galtung's formula does the arithmetic: the gap between what legal services could cost and what they do cost is violence, wearing a suit labelled "the market".
The detail that nails it is the comparison: 34% of lawyers serving business clients cut fees, against 23% serving individual paying clients. Same state, same profession, same tools, an eleven-point spread. Rychner's explanation is candid — corporate clients are repeat users, they use AI themselves, they can advocate for lower prices. Translated: whether you capture the efficiency gain depends on whether you have bargaining power, not on whether the efficiency happened. Without that power, the saved hours don't become your money. They become someone else's margin.
There's a second layer, and it's elegant. 46% of lawyers say clients arrive with inaccurate or misleading information; 41% say they bring AI-generated documents that need correcting; nearly half say client AI use increased their own time and effort. The client's DIY saving on advice comes back onto the invoice as rework. AI didn't lower the barrier to legal services — it moved the barrier, so the client does unpaid labour first and pays for the correction after.
And all of this is happening in law. Law is the industry that allocates rights — who gets what remedy, whose account gets written into a judgment, depends heavily on who can pay. An industry where even a fee cut has to be won through bargaining power tells you something about the texture of the justice it exports. Meta-violence doesn't need to show up here. The structure runs on its own.
The only named casualty in the report is the Victorian lawyer stripped of his principal practising certificate over false citations — the first in the country. The tool erred, the user answered for it. Fair enough. But nobody answers for the clients' rework hours, and those hours appear in no report.
So next time someone promises an AI efficiency dividend, ask who cashes it. The answer is rarely in the technical documentation. It's in who gets to negotiate the price.