
• 研究人员利用人工智能学习模型分析财报电话录音中的发声特征,首次推出识别首席执行官抑郁症的新措施。研究有助于找出可能导致高管面临心理健康挑战的因素。
首席执行官向投资者介绍公司业绩时,措辞用语可能泄露心理健康问题。本月发表在《会计研究杂志》(Journal of Accounting Research)上的一项研究称,利用人工智能分析首席执行官的讲话录音就能判断是否存在抑郁。
印第安纳大学凯利商学院(Indiana University’s Kelley School of Business)和肯塔基大学(University of Kentucky)的研究人员训练人工智能语言模型分析首席执行官的说话模式,首次提出了鉴别首席执行官抑郁症程度和普遍性的方法。分析内容包括 2010 年到 2021 年标准普尔 500 企业的 14500 多份财报电话数据。
研究证据表明,患有抑郁症的首席执行官在职场上会面临更多挑战。具体来说,首席执行官的心理健康问题与公司面临的更大风险相关,如诉讼或股票收益波动。另有少量证据显示,患有抑郁症的首席执行官获得丰厚薪酬的可能性更高,且薪酬中绩效占比更高。其中年长者和女性的几率更低。
“我们想强调领导层的心理健康问题,以及该现象的普遍性,”印第安纳大学会计学助理教授、研究合著者纳格丝・戈尔尚告诉《财富》杂志。心理健康对高管个人很重要,对组织、员工、投资者和更广泛的经济同样影响深远。
人工智能如何学习识别抑郁症
研究人员一直用语音分析评估阿尔茨海默症和帕金森等慢性疾病,也用类似分析检测抑郁症。
现在研究人员不再盯着以前用于评估健康状况的基础语言要素,例如通常与抑郁症有关的停顿和填充词。而是开始用人工智能捕捉人耳无法察觉的细微模式。
“机器学习模型要复杂得多,”戈尔尚说,“模型主要分析音频文件片段的数值编码,都是人类无法感知的内容。”
戈尔尚从一组接受心理健康评估的非首席执行官样本中收集了语音分析数据,与患者健康问卷等可靠的抑郁症判断工具得出的分数进行交叉对比。她利用数据集训练机器学习模型,该模型可以从首席执行官的演讲中识别出可能暗示抑郁存在的小段数据片段。财报电话会议是收集数据的最佳方式,因为会议中发言时间很长且不间断,而且不会受到手势等视觉交流线索的干扰。
在研究的 14500 多名首席执行官中,机器学习模型分析后认为9500多人患有抑郁。
商业与心理健康之间的关系
借助人工智能的心理健康评估,研究人员已能识别出首席执行官抑郁状况与商业风险之间的关联,不过戈尔尚提醒称二者并无因果关系。
她在研究的公司中发现,如果首席执行官有抑郁症,公司会面临更大风险,包括面临诉讼或股票收益不可预测等。戈尔尚推测,这可能与抑郁症患者对反馈的处理方式有关。抑郁症患者更容易将负面反馈内化,对正面反馈却不太敏感。如果纠结财季表现不如预期,可能引发更多负面的自我对话,进而加重抑郁症状。
研究还发现,尽管证据有限,但首席执行官的心理健康得分与薪酬待遇之间存在关系,例如,患有抑郁症的高管获得的薪酬更高。戈尔尚说,可能因为董事会希望支持或激励陷入困境的高管。
研究人员已在深入研究抑郁症与薪酬、人员流动率和公司业绩等因素之间的潜在因果关系。抑郁症与商业风险之间的关联指明了未来的研究领域,但主要证实了新的人工智能模型可有效衡量抑郁症。戈尔尚断言,未来该类心理健康研究会产生一些无形的影响。
“抑郁症一直有种污名化色彩……我们希望通过研究真正让人们了解抑郁症,尤其是其普遍性,”她说,“希望能开启讨论,帮助高管发现问题,也让公司意识到应该为担任相关岗位的高管提供支持。”
对首席执行官来说,心理健康仍然是污点
高管的心理健康问题并不会在离开办公室后突然消失。根据德勤(Deloitte)对美国、英国、加拿大和澳大利亚3150名员工进行的《2023年工作幸福感调查》(2023 Well-Being at Work Survey),四分之三的高管表示会认真考虑辞职,寻找更有利于身心健康的工作场所。
尽管高管们高度重视心理健康,但由于相关领域的偏见持续存在,鲜少有人公开讨论并寻求帮助以改善心理健康。BusinesSolver 对2万名员工开展的《2024 年职场同理心状况》(2024 State of Workplace Empathy)调查发现,八成首席执行官以及67% 的员工认为患有精神疾病的人很脆弱或会成为负担。
心理健康挑战可能导致企业发生重大变化。Toms是一家有慈善性质的便鞋品牌,2014年其创始人布莱克·迈科斯基将公司50%的股份卖给了贝恩资本(Bain Capital),理由是抑郁和孤独。
“很多明确的意义和目标都没了,” 去年4月,迈科斯基在接受《财富》杂志采访时如是说。(财富中文网)
译者:梁宇
审校:夏林
• 研究人员利用人工智能学习模型分析财报电话录音中的发声特征,首次推出识别首席执行官抑郁症的新措施。研究有助于找出可能导致高管面临心理健康挑战的因素。
首席执行官向投资者介绍公司业绩时,措辞用语可能泄露心理健康问题。本月发表在《会计研究杂志》(Journal of Accounting Research)上的一项研究称,利用人工智能分析首席执行官的讲话录音就能判断是否存在抑郁。
印第安纳大学凯利商学院(Indiana University’s Kelley School of Business)和肯塔基大学(University of Kentucky)的研究人员训练人工智能语言模型分析首席执行官的说话模式,首次提出了鉴别首席执行官抑郁症程度和普遍性的方法。分析内容包括 2010 年到 2021 年标准普尔 500 企业的 14500 多份财报电话数据。
研究证据表明,患有抑郁症的首席执行官在职场上会面临更多挑战。具体来说,首席执行官的心理健康问题与公司面临的更大风险相关,如诉讼或股票收益波动。另有少量证据显示,患有抑郁症的首席执行官获得丰厚薪酬的可能性更高,且薪酬中绩效占比更高。其中年长者和女性的几率更低。
“我们想强调领导层的心理健康问题,以及该现象的普遍性,”印第安纳大学会计学助理教授、研究合著者纳格丝・戈尔尚告诉《财富》杂志。心理健康对高管个人很重要,对组织、员工、投资者和更广泛的经济同样影响深远。
人工智能如何学习识别抑郁症
研究人员一直用语音分析评估阿尔茨海默症和帕金森等慢性疾病,也用类似分析检测抑郁症。
现在研究人员不再盯着以前用于评估健康状况的基础语言要素,例如通常与抑郁症有关的停顿和填充词。而是开始用人工智能捕捉人耳无法察觉的细微模式。
“机器学习模型要复杂得多,”戈尔尚说,“模型主要分析音频文件片段的数值编码,都是人类无法感知的内容。”
戈尔尚从一组接受心理健康评估的非首席执行官样本中收集了语音分析数据,与患者健康问卷等可靠的抑郁症判断工具得出的分数进行交叉对比。她利用数据集训练机器学习模型,该模型可以从首席执行官的演讲中识别出可能暗示抑郁存在的小段数据片段。财报电话会议是收集数据的最佳方式,因为会议中发言时间很长且不间断,而且不会受到手势等视觉交流线索的干扰。
在研究的 14500 多名首席执行官中,机器学习模型分析后认为9500多人患有抑郁。
商业与心理健康之间的关系
借助人工智能的心理健康评估,研究人员已能识别出首席执行官抑郁状况与商业风险之间的关联,不过戈尔尚提醒称二者并无因果关系。
她在研究的公司中发现,如果首席执行官有抑郁症,公司会面临更大风险,包括面临诉讼或股票收益不可预测等。戈尔尚推测,这可能与抑郁症患者对反馈的处理方式有关。抑郁症患者更容易将负面反馈内化,对正面反馈却不太敏感。如果纠结财季表现不如预期,可能引发更多负面的自我对话,进而加重抑郁症状。
研究还发现,尽管证据有限,但首席执行官的心理健康得分与薪酬待遇之间存在关系,例如,患有抑郁症的高管获得的薪酬更高。戈尔尚说,可能因为董事会希望支持或激励陷入困境的高管。
研究人员已在深入研究抑郁症与薪酬、人员流动率和公司业绩等因素之间的潜在因果关系。抑郁症与商业风险之间的关联指明了未来的研究领域,但主要证实了新的人工智能模型可有效衡量抑郁症。戈尔尚断言,未来该类心理健康研究会产生一些无形的影响。
“抑郁症一直有种污名化色彩……我们希望通过研究真正让人们了解抑郁症,尤其是其普遍性,”她说,“希望能开启讨论,帮助高管发现问题,也让公司意识到应该为担任相关岗位的高管提供支持。”
对首席执行官来说,心理健康仍然是污点
高管的心理健康问题并不会在离开办公室后突然消失。根据德勤(Deloitte)对美国、英国、加拿大和澳大利亚3150名员工进行的《2023年工作幸福感调查》(2023 Well-Being at Work Survey),四分之三的高管表示会认真考虑辞职,寻找更有利于身心健康的工作场所。
尽管高管们高度重视心理健康,但由于相关领域的偏见持续存在,鲜少有人公开讨论并寻求帮助以改善心理健康。BusinesSolver 对2万名员工开展的《2024 年职场同理心状况》(2024 State of Workplace Empathy)调查发现,八成首席执行官以及67% 的员工认为患有精神疾病的人很脆弱或会成为负担。
心理健康挑战可能导致企业发生重大变化。Toms是一家有慈善性质的便鞋品牌,2014年其创始人布莱克·迈科斯基将公司50%的股份卖给了贝恩资本(Bain Capital),理由是抑郁和孤独。
“很多明确的意义和目标都没了,” 去年4月,迈科斯基在接受《财富》杂志采访时如是说。(财富中文网)
译者:梁宇
审校:夏林
• Researchers debuted a new measure of identifying CEO depression by using AI learning models to analyze vocal features from earnings call recordings. This research has helped identify factors that may contribute to mental health challenges among executives.
CEOs might be able to give away mental health challenges just by how they talk about their companies’ earnings to investors. A study published this month in the Journal of Accounting Research used artificial intelligence to analyze chief executives’ speech recordings to identify depression.
Researchers from Indiana University’s Kelley School of Business and the University of Kentucky debuted a measure of identifying the severity and prevalence of depression among chief executives by training AI language models to analyze vocal patterns of CEOs. They analyzed data from more than 14,500 earnings calls from S&P 500 companies from 2010 to 2021.
CEOs with depression tend to face additional workplaces challenges, evidence from the study suggested. Specifically, a CEO’s mental health struggles were associated with a company facing greater risks, such as litigation or volatile stock returns. There was also limited evidence showing CEOs with depression were more likely to have larger compensation packages and have a higher percentage of those packages based on performance. They were less likely to be older and women.
“We want to really highlight mental health in leadership roles and how prevalent it is,” Nargess Golshan, assistant professor of accounting at Indiana University and the study’s co-author, told Fortune. “Of course, it is important for the personal health of these executives, but also has far-reaching implications for the organization, the employees, the investors, and the broader economy.”
How AI learns to identify depression
Researchers have long used voice analysis as a tool for assessing chronic illnesses, such as Alzheimers and Parkinson’s disease, and measuring depression through a similar analysis is no exception.
Rather than look at more rudimentary speech components previously used to assess health conditions—such as pauses and uses of filler words, which are associated with depression—researchers are now turning to AI to pick up on patterns too small for the human ear to notice.
“These machine learning models [are] more complicated than that,” Golshan said. “They use numerical embeddings of pieces of the audio file that are not really perceptible by humans.”
Golshan collected vocal analysis data from a sample of non-CEOs who took mental health assessments, cross-referencing that data with scores from reliable tools for determining depression, like the Patient Health Questionnaire. She used that data set to train her machine learning model, which could identify small pieces of data from CEO’s speeches that could indicate depression. Earnings calls are an optimal way to collect data because they feature long, uninterrupted periods of talking and usually aren’t confounded by visual communication cues like hand gestures.
Among more than 14,500 CEOs studied, More than 9,500 were classified as having depression using analysis from the machine learning model.
The relationship between business and mental health
AI-powered mental health assessments have already allowed researchers to identify correlations between CEO depression and business risks, though Golshan warns that no causal connections can be made.
She found among the companies in the study, having a CEO with depression was associated with greater risks to a firm, including facing lawsuits or unpredictable stock returns. Golshan hypothesizes this could have something to do with how individuals with depression interrupt feedback. Those with depression are more likely to deeply internalize negative feedback, but are less sensitive to positive feedback. Dwelling on a worse-than-expected fiscal quarter could cause further negative self-talk, worsening depression symptoms.
The study also found, albeit limited, evidence of a relationship between a CEO’s mental health score and their compensation package, such that executives with depression had larger payouts. This might be a result of a board wanting to support or incentivize a struggling executive, Golshan said.
Researchers are already diving deeper into the potential causal relationships between depression and factors such as compensation, turnover, and company performance. The associations between depression and business risk point to areas of future research, but they mostly validate the new AI model as a useful tool in measuring depression. Golshan asserted there are also intangible impacts of the future of this mental health research.
“Depression has always been attached with the stigma…We hope, with this study, to really bring some light to it, especially how prevalent it is,” she said. “We want to start a conversation and help executives to be aware about it, and also companies, to support their executives in these roles.”
Mental health is still stigmatized for CEOs
Mental health troubles don’t magically dissolve outside the corner office. Three-quarters of the C-suite said they would seriously consider quitting their jobs in order to seek out a workplace that would better support their wellbeing, according to Deloitte’s 2023 Well-Being at Work Survey, which polled 3,150 employees across the United States, UK, Canada, and Australia.
But despite the high prioritization of mental health among executives, discussing and seeking help for improving mental wellbeing has been tamped down by the continued stigma about those struggles. Eight in 10 CEOs and 67% of employees believe someone with a mental illness is weak or burdensome, BusinesSolver’s 2024 State of Workplace Empathy surveying 20,000 employees found.
These mental health challenges can result in material changes for a business. Blake Mycoskie, founder of Toms, the slip-on shoe brand with a philanthropic bent, sold 50% of the company to Bain Capital in 2014, citing depression and loneliness.
“I lost a lot of my clear meaning and purpose,” Mycoskie told Fortune in April.