内容来源：https://www.livemint.com/ai/artificial-intelligence/ai-isn-t-really-stealing-jobs-yet-that-doesn-t-mean-we-re-ready-for-it-11755834528916.html

内容总结：

【评论观点】美国失业保障体系难抵AI潜在冲击，改革迫在眉睫

近期关于"AI窃取就业机会"的讨论持续升温，但政策分析人士指出，当前数据并未支撑这一恐慌性论断——真正令人担忧的是美国失业保障体系尚未做好应对任何重大就业冲击的准备。

就业政策分析师威尔·雷德曼在客座评论中指出，尽管生成式AI尚未导致就业市场显著恶化，但美国失业救济系统存在的系统性缺陷已亮起红灯。数据显示，22-27岁大学毕业生失业率自2010年代起就持续高于总体劳动年龄人口，这一趋势远早于生成式AI技术的普及。

值得注意的是，计算机系统、会计审计、市场研究等需要大学学历的岗位就业率保持增长，金融保险等AI应用程度较高的行业近年来反而增加了应届生招聘。真正需要警惕的是：若AI未来确实引发就业危机，现有体系将难以应对。

当前美国失业救济系统存在两大顽疾：
一是资金拨付机制缺陷。联邦拨付各州失业管理机构的工作经费经通胀调整后较21世纪初缩减约三分之一，且年度拨款额波动无常。更严重的是，由于联邦行政资金池设有严格上限，每年数亿本应用于系统升级、反欺诈建设的资金被转入应急账户。新冠疫情前四年间，近50亿美元管理资金因此遭挪用。

二是资格核查流程低效。各州通常在申请人提交索赔后才向雇主核实离职原因，而雇主回应时限参差不齐甚至不予回复，导致发放延迟和错误支付。劳工部数据显示，四分之三的超额支付无法通过现有程序识别。

专家建议推行两项改革：取消联邦行政资金池上限确保专款专用，以及强制要求企业在员工离职时即时提交分离信息。据测算，后者可减少约20%的支付错误并加速审批流程。

评论最后强调，无论冲击来自AI技术革新、地缘政治危机还是其他经济变故，若不及早改革失业保障体系，未来劳动者在职业过渡期将面临巨大风险。当前最紧迫的任务是未雨绸缪，构建具有韧性的就业安全网。

（本文观点来自尼斯卡南中心就业政策分析师威尔·雷德曼，不代表本媒体立场）

中文翻译：

人工智能尚未真正开始抢走工作岗位，但这并不意味着我们已经做好了准备。

客座评论员威尔·雷德曼撰文指出：即便只是遭遇人工智能引发的轻度就业冲击，现有失业救济体系也难以应对。

作者简介：威尔·雷德曼是美国尼斯卡宁中心的就业政策分析师。

应届大学毕业生面临独特困境——这种独特性却令人担忧。

与往届毕业生不同，22至27岁年龄段的青年失业率目前高于劳动年龄人口总体水平。主流观点自然将此归咎于生成式人工智能的兴起。

这种危言耸论的缺陷在于缺乏数据支撑。至少现阶段，人工智能并未夺走应届生的就业机会——这本该令人宽慰，因为若真出现这种颠覆性局面，美国现行失业保障体系根本无力招架。

早在生成式AI模型问世前的2010年代，应届毕业生失业率就已呈现异常攀升趋势。而过去几年间，那些受AI冲击风险中高的职业领域实际表现反而更为稳健。

最新青年就业数据显示，计算机系统、会计审计、市场研究等需要大学学历的岗位就业率持续增长。金融保险等AI密集型行业近年对应届毕业生的招聘规模也在扩大。

自ChatGPT发布以来，企业AI使用率超10%的行业与使用率不足10%的行业对应届毕业生的招聘数量基本持平。虽然管理咨询、科技咨询等AI超密集型行业确实减少了应届生招聘，但许多低AI应用行业也出现了类似下滑。数据呈现矛盾态势——目前尚未发现AI应用程度与青年就业状况之间存在明确关联。

更紧迫的问题在于：若这种关联真的形成，我们将面临什么？极端预测认为失业率可能升至20%，相对保守（也更现实）的预估显示未来十年美国将有数百万劳动者被替代。

我认为当前体系对任何一种情况——乃至任何重大就业冲击——都缺乏应对准备。

美国失业保障体系存在系统性缺陷。新冠疫情期间，诈骗分子从各州及联邦临时失业救济计划中窃取约1,350亿美元。多年后的今天，大部分州仍未能达到联邦规定的失业金发放时效性与准确性标准。劳动者往往需等待数周才能获得现金补助和再就业服务，而2024年发放的救济金中仍有16%存在差错。这些都应成为警示信号——无论下一场经济动荡源自人工智能、地缘政治冲击还是其他不可预测的经济事件。

有两大改革重点尤为关键：

首先必须修正失业管理资金的筹措方式。自21世纪初以来，经通胀调整后，各州失业机构赖以运作的联邦资金已缩水约三分之一。且拨付各州的具体金额每年无规律变动。这种资金波动使得招聘计划与反欺诈工具投资难以落实。美国劳工部和政府问责署均意识到这个问题，并将资金缩减和"拨款不确定性"认定为困扰各州机构的"历史遗留问题"。

事实上，每年都有数亿美元原定用于福利管理的资金被挪作他用——并非因各州资金充裕，而是由于联邦对项目资金存储的管理基金设定了严格上限。一旦触及上限，超额资金便会转入应急福利账户，导致各州缺乏项目管理的专项资金。

这产生了实质影响：疫情爆发前四年，近50亿美元本可用于系统升级、专家招聘和反欺诈强化的资金被转移。致使各州在应对新冠危机时左支右绌，这种困境很可能重演。

其次需改革失业资格认证流程。发放救济金前，州失业机构需核实申领者的离职原因。但当前机构通常在接到申请后才联系雇主核实——这导致审批延迟和错误发放频发。各州对雇主回复时限规定不一，而雇主往往延迟回复或根本不回应。

将过错完全归咎于雇主有失公允。劳工部估计四分之三的超额支付无法通过现有程序识别，但他们发现改变雇主信息收集方式可减少差错。简单改革——在员工离职时即刻收集离职信息——就能降低约20%的发放错误并加速审批流程。国会应推动将此确立为标准操作程序。

若人工智能真的取代人力，应届毕业生和众多劳动者必将承受阵痛。他们本应相信职业过渡能够平稳实现，但现行体系显然无法提供这种保障。

若不对失业保险体系进行改革，劳动者终将陷入进退维谷的困境。

（本文作者非巴伦周刊编辑人员，所述观点仅代表作者个人立场。欢迎将反馈与评论投稿至ideas@barrons.com）

英文来源：

AI isn’t really stealing jobs yet. That doesn’t mean we’re ready for it.
Unemployment agencies can’t handle even a modest AI-induced employment shock, Will Raderman writes in a guest commentary.
About the author: Will Raderman is an employment policy analyst at the Niskanen Center.
Recent college graduates are unique—and not in a good way.
Unlike preceding generations, graduates aged 22-27 have a higher unemployment rate than the total working-age population. The prevailing hypothesis, of course, blames the rise of generative AI.
The problem with that alarmist narrative is that data doesn’t back it up. AI isn’t stealing new grads’ jobs—yet. That should be a relief, because America’s unemployment system isn’t ready for the disruptions that would come if it did.
Recent graduates’ unemployment rates have been drifting in the wrong direction since the 2010s, long before generative AI models hit the market. And many occupations with moderate to high exposure to AI disruptions are actually faring better over the past few years.
According to recent data for young workers, there has been employment growth in roles typically filled by those with college degrees related to computer systems, accounting and auditing, and market research. AI-intensive sectors like finance and insurance have also seen rising employment of new graduates in recent years.
Since ChatGPT’s release, sectors in which more than 10% of firms report using AI and sectors in which fewer than 10% reporting using AI are hiring relatively the same number of recent grads. It is true that employment for new graduates has fallen in sectors that use AI hyper-intensively, such as management, scientific, and technical consulting services. But plenty of low AI-use sectors have experienced similar hiring declines. So the data is mixed—and there is no clear link yet between higher AI use and worse outcomes for young workers.
A more pressing question is what will happen if and when that link does materialize. Extreme estimates project unemployment would rise to 20%. More modest (and realistic) projections say millions of American workers could be displaced over the next decade.
I don’t believe we are ready for either situation—or any major employment shock for that matter.
The U.S. unemployment system is riddled with systematic issues. During the Covid-19 pandemic, fraudsters siphoned off an estimated $135 billion from state and temporary federal unemployment programs. Years later, the average state still misses federal benchmarks for unemployment benefit payment timeliness and accuracy. Workers are forced to wait too many weeks before getting cash benefits and re-employment services. Meanwhile, 16% of benefit payments contained errors in 2024. These should be flashing warning signs for the next economic upheaval—whether it be driven by AI, a geopolitical shock, or any other unpredictable economic event.
Two fixes stand out.
First, we must correct how unemployment administration is financed. Since the early 2000s, the federal dollars that state unemployment agencies rely on to run their operations have shrunk by about one-third when adjusted for inflation. And the specific amounts provided to states change unpredictably year to year. This seesaw in funding makes it nearly impossible to plan hiring or invest in better fraud detection tools. Both the Department of Labor and Government Accountability Office know this. They have identified declining funding and “funding uncertainties" as a “historical issue" plaguing state agencies.
In fact, hundreds of millions of dollars earmarked for the administration of benefits tend to be repurposed each year—not because states don’t need the money—but as a result of a strict federal cap on the size of the federal administrative fund where the program resources are kept. Once the cap is hit, excess cash is shifted into a separate account for emergency benefits, leaving states without resources intended for program management.
This has real consequences. In the four years before the onset of the pandemic, nearly $5 billion that could have upgraded systems, hired field experts, and improved fraud controls was swept away. As a result, states entered the Covid-19 crisis with one hand tied behind their backs. The same dynamic could play out again.
We must also adjust how unemployment eligibility is verified. Before issuing benefits, state unemployment agencies need to know why a claimant left their job. But agencies generally question claimants’ employers after they file a claim—opening the door to long delays and mistaken payments. Response deadlines vary by state, and employers often answer late or not at all.
It would be a mistake to solely blame employers for these errors. The Department of Labor estimates that three-quarters of overpayments can’t be caught by agency procedures. They have found, however, that changing how employer information is gathered would reduce errors. A simple change—collecting separation information at the moment employees leave—could reduce payment mistakes by about 20% and speed up approvals. Congress should work to make that the standard.
If and when AI displaces workers, recent graduates and many others will feel pain. They should be able to trust that their transition between jobs will be smooth. That isn’t a reliable bet right now.
Without reforms to the unemployment insurance system, workers will be stuck between a rock and a hard place.
Guest commentaries like this one are written by authors outside the Barron’s newsroom. They reflect the perspective and opinions of the authors. Submit feedback and commentary pitches to ideas@barrons.com.