Deal-Making Reimagined: AI's Investment Banking Revolution

Ai4

AI-assisted summary of Deal-Making Reimagined: AI's Investment Banking Revolution — original source. Check the original for full context.

Source published: 2025-08-28

Detailed notes

## Background Published by Ai4 on YouTube on August 28, 2025, this presentation features Wharton Executive Director Sarah Hammer exploring the intersection of artificial intelligence and investment banking. The video's description outlines how AI complements relationship-driven advisory by automating due diligence, simulating transaction trajectories, and processing complex datasets alongside senior banker judgment.

## TL;DR Drawing on her multidisciplinary career spanning JPMorgan, the U.S. Treasury, state banking regulation, and Wharton, Sarah Hammer analyzes how artificial intelligence is reshaping capital markets advisory. She details how large language models can monitor systemic liquidity stress and accelerate routine banking workflows like presentation drafting, compliance documentation, and financial modeling. However, Hammer draws a firm line at complete automation, emphasizing that raw computational power cannot generate investment conviction or replace fiduciary responsibility. Ultimately, she demonstrates that while AI provides measurable operational productivity gains, successful deal-making still fundamentally depends on human intuition, trust, and relationship equity.

## Atmosphere & style The session has an energetic, authoritative, and engaging tone, blending high-level regulatory war stories with pragmatic market insights. Hammer speaks with rapid, confident delivery, using interactive room polls and candid exchanges about trusting models versus experienced advisors to keep the audience actively engaged. Watching it feels like attending an insider executive briefing that balances quantitative optimism with grounded, real-world skepticism toward full automation.

## Key takeaways - [00:04](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=4s) Sarah Hammer brings a multidisciplinary perspective bridging Wall Street trading at JPMorgan, federal policy at the U.S. Treasury, state bank regulation, and Wharton academia to examine AI's practical role in finance. - [03:08](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=188s) The rapid deposit flight during the March 2023 regional banking crisis inspired Wharton researchers to build domain-specific LLMs to monitor real-time social sentiment and unstructured data for systemic liquidity stress. - [07:02](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=422s) Major industry and macroeconomic benchmarks—including Goldman Sachs' 7% global GDP projection and BloombergGPT testing—demonstrate that AI models systematically outperform human baselines in financial data synthesis. - [07:02](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=422s) A fundamental debate persists between theoretical economists who anticipate total AI automation and market practitioners who argue algorithms cannot replicate authentic human trust and relational intuition. - [10:13](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=613s) Advanced AI analytics equip deal teams to present ironclad empirical evidence to corporate boards and financial regulators when justifying disruptive strategic transactions. - [10:13](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=613s) While AI eliminates human limits on data ingestion, raw computational throughput cannot generate the conviction, lived experience, or institutional memory required to commit investment capital. - [14:41](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=881s) Empirical studies, including research from MIT and a Deloitte capital markets analysis projecting 27% to 35% productivity gains, show AI dramatically elevates operational efficiency in financial documentation and modeling. - [17:10](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=1030s) Even though AI models routinely pass specialized finance benchmarks, clients and institutions remain fundamentally hesitant to delegate fiduciary authority without a trusted human intermediary. - [19:34](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=1174s) During complex economic events and market crises, multi-decade human experience and regulatory judgment provide contextual advisory insights that standalone AI models cannot deliver. - [20:18](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=1218s) Deals are ultimately closed through 'relationship equity' and strategic rapport, positioning AI as an analytical accelerator that augments rather than replaces senior investment banker advisory.

## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=4s) Introduction and Cross-Disciplinary Perspectives

Sarah Hammer opens her address by emphasizing the need to stay on schedule and diving straight into the intersection of investment banking and artificial intelligence. Having served as the Executive Director at the Wharton School, Adjunct Professor at the University of Pennsylvania Carey Law School, and CEO of the Wharton Cypher Accelerator—which focused on scaling artificial intelligence, blockchain, and quantum computing ventures—Hammer brings an extensive background spanning academia, policy, and market practice. Having just recently returned to the private sector, her goal is not merely to outline theoretical AI capabilities, but to examine how the investment banking industry is actively deploying the technology, how global institutions are positioning themselves, and how professionals can collaborate with machines to maximize their commercial potential.

Her viewpoint is shaped by high-level leadership roles across government and Wall Street. Before her tenure at Wharton, Hammer served at the U.S. Department of the Treasury as Deputy Assistant Secretary for Financial Institutions, overseeing the Office of Financial Institutions Policy, the Federal Insurance Office (including the Terrorism Risk Insurance Program), and the Office of Cybersecurity and Compliance Policy. In early 2023, she took a leave of absence to serve as Pennsylvania’s Secretary of Banking and Securities, an appointment by the Governor that placed her on the front lines of the regional banking crisis. Prior to her public service, Hammer accumulated over 25 years of market experience, including trading at JPMorgan, heading portfolio analytics and macroeconomic research, and running an endowment and foundation investment enterprise. Beyond these posts, she co-chairs an international consortium of AI experts studying economic impact and governance, co-chairs the AI Working Group at the Bretton Woods Committee, and participates in the American Law Institute’s Member Consultative Group on Civil Liability for AI.

- **U.S. Treasury oversight**: Directed policy across financial institutions, federal insurance frameworks, terrorism risk insurance, and cyber compliance. - **State regulatory command**: Led Pennsylvania’s Department of Banking and Securities through the volatile March 2023 bank runs. - **Commercial finance roots**: Spent years in capital markets and trading at JPMorgan, institutional asset management, and macro strategy.

---

[03:08](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=188s) Lessons from the 2023 Banking Crisis and Liquidity Modeling

Weaving her regulatory and trading experiences together, Hammer highlights the unique vantage points that inform her understanding of financial technology. In the Treasury Department's cash room, discussions centered on the friction between legacy regulatory frameworks and technological adoption. Later, inside boardroom meetings during the March 2023 banking panic, she witnessed firsthand how instantaneous information flows through social media and digital asset channels altered liquidity velocity. The crisis began with the collapse of Silvergate Bank—a premier institution for digital asset companies that had absorbed assets from FTX—unfolding just weeks before Silicon Valley Bank faced catastrophic deposit outflows.

As panic spread, massive quantities of cash fled distressed institutions toward safer harbors, including regional banks regulated under Hammer’s watch in Pennsylvania. Startups and founders openly broadcasted their capital relocations on social media in real time. This experience inspired Hammer and her research team upon her return to Wharton to construct a specialized large language model designed to monitor real-time social media sentiment and text data to gauge systemic liquidity stress and detect bank-run dynamics before they materialize. While identifying impending crises from unstructured data remains challenging, Hammer views modern natural language processing as a critical breakthrough for financial stability and macro research.

```mermaid flowchart TD accTitle: Banking Stress Detection Pipeline subgraph DataIngestion["1. Unstructured Data Feeds"] SM["Social Media Streams"] Chat["Corporate Communications & Chats"] News["Financial News & Filings"] end

subgraph Analytics["2. AI Processing Engine"] LLM["Domain-Specific LLM"] Sent["Sentiment & Flight-to-Safety Signals"] LLM --> Sent end

subgraph HumanDecision["3. Executive & Regulatory Action"] Reg["Regulatory Intervention"] Treasury["Liquidity & Treasury Safeguards"] Thesis["Investment Thesis Formulation"] end

SM --> LLM Chat --> LLM News --> LLM Sent --> Reg Sent --> Treasury Sent --> Thesis ```

- **Silvergate precursor**: The crypto-focused lender's liquidation preceded SVB by weeks, setting off an initial wave of deposit reallocation. - **Deposit migration**: Startups across tech and AI publicly shared their treasury diversification moves on social media, channeling funds into Pennsylvania-chartered banks. - **Wharton LLM initiative**: Developed an NLP pipeline to harness real-time sentiment data as an early-warning radar for liquidity shocks.

---

[07:02](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=422s) Macroeconomic Benchmarks and the Limits of Machine Intelligence

To frame AI's quantitative promise, Hammer surveys several landmark benchmark studies published across the economic and financial landscape. A widely cited analysis by Goldman Sachs estimated that generative AI could drive a 7% increase in global GDP, a macroeconomic surge rivaling the transformative technological shifts of prior generations. In advisory services, McKinsey projected massive productivity and value generation within investment banking. Sector-specific models have repeatedly demonstrated computational advantages: a 2023 European study established that BloombergGPT outperformed human professionals at synthesizing and analyzing complex financial disclosures, while research from the European Central Bank confirmed that advanced large language models exceed human baselines in processing and interpreting broad economic data.

Despite these impressive benchmarks, Hammer cautions against overstating AI's capabilities. When polling the room on whether AI can perform every human role, she recounts a spirited debate with her colleague Anton Korinek, an economist and AI expert in her consortium. Korinek maintained that AI will inevitably replicate every dimension of human capability. When Hammer pushed back, arguing that machines cannot replicate human connection, genuine trust, or the nuanced emotional perception exchanged between people, Korinek responded: "I'm really sorry, Sarah, but it can do all of that." While Hammer humorously concedes she will not lose sleep over an automated future, she firmly rejects the premise that machine calculation can replace human relational capital.

- **Goldman Sachs forecast**: Anticipates a 7% expansion in global GDP powered by enterprise AI adoption. - **Disclosure processing**: Empirical testing showed BloombergGPT outpacing human analysts in evaluating complex financial statements. - **ECB economic analysis**: European Central Bank research confirmed LLMs systematically outperform humans at digesting macroeconomic indicators. - **The Korinek debate**: Highlights the divide between theoretical economists anticipating total automation and market practitioners defending the primacy of human trust.

---

[10:13](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=613s) Deal Teams, Governance, and the Fate of Junior Analysts

Exploring practical corporate governance, Hammer details how artificial intelligence strengthens communication with executive boards and regulatory bodies without supplanting final accountability. An executive cannot approach a board of directors with a sweeping, disruptive thesis—such as declaring traditional commercial banks unviable due to margin compression, interest-rate volatility, and strict capital requirements—without ironclad data. AI equips deal teams to synthesize sprawling structured and unstructured information into empirical arguments that withstand board scrutiny. Similarly, when approaching regulators regarding bank mergers or structural asset reallocations, arriving with AI-driven compliance analytics establishes immediate credibility.

Turning to career implications, Hammer addresses the pervasive anxiety felt by Wharton students: will AI render entry-level investment banking and venture capital analysts obsolete? An audience member offers their rationale for automation, noting that human cognitive throughput is inherently limited, whereas tireless AI systems can continuously ingest information across disparate sources to extract actionable insights. Hammer agrees that machines excel at parsing data, yet points out that raw calculation cannot generate conviction. An analyst may surface an attractive investment thesis, but seasoned partners rely on lived experience, relational context, and institutional memory to decide whether to risk capital—qualities an automated model cannot replicate.

- **Boardroom persuasion**: Advanced analytics provide the evidentiary baseline required to justify pivot strategies, M&A proposals, or fintech allocations. - **Regulatory alignment**: Empirically rigorous AI assessments streamline approval cycles for complex transactions with supervisory authorities. - **Cognitive bandwidth**: AI eliminates the physical limits of human data ingestion, shifting junior staff roles from routine aggregation toward analytical validation. - **The conviction gap**: Even flawless computational models cannot replace the experiential intuition and personal responsibility required to commit institutional capital.

---

[14:41](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=881s) Operational Augmentation and Measurable Productivity Gains

Turning to day-to-day transaction workflows, Hammer assesses how artificial intelligence transforms standard banking deliverables like pitch decks, investor memos, and presentation materials. When polling the audience on presentation generation, only about 5% report using generative tools for slide decks. Hammer shares that she used generative software to create the visual design and structural layout of her own presentation slides; however, she notes that the conceptual substance, personal anecdotes, and historical lessons from the banking crisis were authored entirely by her. AI functions as an efficient drafting and design partner, but it possesses no inherent narrative perspective.

To measure these efficiency gains, Hammer references foundational academic and industry research. A 2023 study by Shakked Noy and Whitney Zhang (MIT) demonstrated measurable efficiency and quality improvements across professional writing tasks, paralleling broader productivity research from Stanford. Furthermore, a 2023 Deloitte study evaluating the capital markets sector projected that deploying AI within investment banking will generate overall productivity gains between 27% and 35%. Even as algorithms automate routine financial modeling, spread adjustments, and initial deck formulation historically assigned to first- and second-year analysts, the net effect elevates total institutional capacity rather than diminishing the industry's economic output.

- **Presentation drafting**: AI handles formatting, conceptual structure, and aesthetic design, freeing deal makers to focus on strategy and narrative clarity. - **Empirical writing uplift**: Studies by MIT researchers (Noy & Zhang, 2023) confirm marked speed and quality improvements in structured documentation. - **Deloitte capital markets benchmark**: Estimates that artificial intelligence integration will generate a 27% to 35% productivity boost across investment banking operations.

---

[17:10](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=1030s) The Human Edge: Conviction, Intuition, and the Question of Trust

In the concluding segment of this portion, Hammer circles back to the core differentiator in financial advisory: the human edge. She emphasizes that algorithms cannot manufacture trust, express genuine empathy, or exercise human intuition. Touching on state-of-the-art developments, she asks the audience about their experiences with GPT-5 and Deep Research tools. Hammer prefers GPT-5 for enterprise research, noting its accessibility, cost efficiency, and ability to parse complex queries in a single consolidated prompt without cumbersome prompt engineering.

The discussion pivots into a candid exchange with the audience on whether investors should trust algorithmic systems over human advisors. Hammer asks if anyone in the room would entrust their entire investment portfolio to an autonomous AI agent. An attendee responds that they trust AI more than the average human, arguing that human intuition is demonstrably flawed and incorrect at least half the time, while the typical person could not pass the specialized benchmark exams that AI routinely aces. Hammer counters playfully by asking the audience whom they truly trust when capital is on the line:

> "Maybe your close family, your best friend, but beyond that, do you trust me? Give me your money. I'll invest it for you... All right, so if we trust you, we'll give you all our money after this presentation."

This exchange highlights the psychological reality of high-stakes finance: while quantitative models can surpass human processing power, financial transactions ultimately depend on fiduciary trust, personal credibility, and human relationships.

- **Next-generation tooling**: GPT-5 delivers rapid, high-context research synthesis through single-shot prompting, lowering costs compared to specialized deep search alternatives. - **The intuition challenge**: Audience dialogue highlights that while human intuition is frequently biased or inconsistent, human clients still hesitate to delegate fiduciary authority to autonomous machines. - **The relational barrier**: Mathematical competence alone does not bridge the emotional divide required for clients to relinquish control of capital without a trusted human intermediary.

[19:34](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=1174s) Trusting Models vs. Human Experience: The 25-Year Advisory Test

While algorithmic models offer clear advantages, the fundamental tension in modern finance lies in whether clients place greater trust in machines or in people. It is entirely understandable why certain market participants lean toward artificial intelligence: AI operates strictly as a model, whereas human beings are driven by emotion and are inherently prone to making mistakes. However, when evaluating complex transactions where nuance and accountability are paramount, the human touch remains indispensable.

Hammer challenges the audience with a direct comparison based on her own multidisciplinary background. Drawing upon her 25 years of professional experience spanning roles as a financial regulator, academic, business executive, trader, and portfolio manager, she asks listeners to consider a real-world scenario: if someone wanted deep answers about a systemic event like a banking crisis, would they place greater trust in Anthropic's Claude or in a seasoned human expert with decades of frontline market experience?

```mermaid flowchart TD accTitle: AI Processing vs Human Advisory in Deal-Making subgraph Machine["AI Computational Capabilities"] A["High-Speed Data Processing"] B["Synergy & Trajectory Modeling"] C["Autonomous Risk Flagging"] end subgraph Human["Human Capital & Advisory"] D["Multidisciplinary Experience"] E["Relationship Equity"] F["Strategic Judgment & Trust"] end Machine --> G["Optimized Deal Execution"] Human --> G ```

[20:18](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=1218s) The Limits of Autonomous Deal-Making and the Power of Relationship Equity

Despite the extraordinary data-processing speeds of artificial intelligence, deal-making cannot simply be abdicated to autonomous systems. AI excels at crunching complex datasets faster and sometimes better than humans, mapping out strategic corporate synergies, and flagging obscure risks that analysts might otherwise overlook. Furthermore, industry data and surveys conducted across the financial services sector show that investment banks, private equity firms, and venture capital companies are rapidly moving forward with deploying AI and appointing senior data analytics experts to oversee their investment portfolios.

Yet, technical execution overlooks the most critical currency in finance: "relationship equity." While technology can model outcomes, deals ultimately close through human rapport, strategic advisory, and mutual confidence. Hammer illustrates this dynamic through her own presentation: by spending 20 minutes engaging directly with the room, she established tangible relationship equity with the audience that an algorithm cannot replicate, making participants far more inclined to seek out direct counsel on investment banking and deal-making.

- **Rapid AI Adoption Across Capital Markets**: Financial services surveys indicate that investment banks, private equity firms, and venture capital companies are aggressively deploying AI across portfolio management and transaction workflows. - **Top-Level Analytics Talent**: Financial institutions are increasingly hiring dedicated data analytics leaders at the highest executive levels to evaluate portfolios and identify strategic opportunities. - **Complementary Pillars of Modern Banking**: While machine intelligence optimizes analytical precision and risk detection, relationship equity and seasoned judgment remain the deciding factors in strategic client negotiations.

## Notable quotes > [07:02](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=422s) I'm really sorry, Sarah, but it can do all of that. — Anton Korinek

> [07:02](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=422s) Goldman Sachs estimated that generative AI could drive a 7% increase in global GDP, a macroeconomic surge rivaling the transformative technological shifts of prior generations. — Sarah Hammer

> [10:13](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=613s) An analyst may surface an attractive investment thesis, but seasoned partners rely on lived experience, relational context, and institutional memory to decide whether to risk capital—qualities an automated model cannot replicate. — Sarah Hammer

> [14:41](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=881s) AI functions as an efficient drafting and design partner, but it possesses no inherent narrative perspective. — Sarah Hammer

> [14:41](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=881s) Even as algorithms automate routine financial modeling, spread adjustments, and initial deck formulation historically assigned to first- and second-year analysts, the net effect elevates total institutional capacity rather than diminishing the industry's economic output. — Sarah Hammer

> [17:10](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=1030s) Maybe your close family, your best friend, but beyond that, do you trust me? Give me your money. I'll invest it for you... All right, so if we trust you, we'll give you all our money after this presentation. — Sarah Hammer

> [17:10](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=1030s) Mathematical competence alone does not bridge the emotional divide required for clients to relinquish control of capital without a trusted human intermediary. — Sarah Hammer

> [19:34](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=1174s) AI operates strictly as a model, whereas human beings are driven by emotion and are inherently prone to making mistakes. However, when evaluating complex transactions where nuance and accountability are paramount, the human touch remains indispensable. — Sarah Hammer

> [20:18](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=1218s) While technology can model outcomes, deals ultimately close through human rapport, strategic advisory, and mutual confidence. — Sarah Hammer

> [20:18](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=1218s) While machine intelligence optimizes analytical precision and risk detection, relationship equity and seasoned judgment remain the deciding factors in strategic client negotiations. — Sarah Hammer

## Mentioned Sarah Hammer, Anton Korinek, Shakked Noy, Whitney Zhang, Wharton School, University of Pennsylvania Carey Law School, Wharton Cypher Accelerator, U.S. Department of the Treasury, Federal Insurance Office, Terrorism Risk Insurance Program, Office of Cybersecurity and Compliance Policy, Office of Financial Institutions Policy, Pennsylvania Department of Banking and Securities, JPMorgan, Bretton Woods Committee, American Law Institute, Silvergate Bank, FTX, Silicon Valley Bank, Goldman Sachs, McKinsey, BloombergGPT, European Central Bank, MIT, Stanford University, Deloitte, GPT-5, Deep Research, Claude, Anthropic

## Video description Deal-Making Reimagined: AI's Investment Banking Revolution. Investment banking is evolving through the integration of advanced technology with traditional advisory expertise. Modern transactions require analysis of complex data sets that exceed human processing capabilities - from real-time market signals to cross-border regulatory requirements. The conventional banking model can be significantly enhanced by AI systems that analyze global companies continuously, identifying potential transactions through predictive modeling of corporate trajectories and strategic fits. This approach complements relationship-driven deal-making with systematic opportunity identification - recognizing optimal timing and structure for M&A, IPO, and financing transactions. Digital twin simulation enables testing of thousands of potential deal outcomes before approaching companies, while autonomous due diligence systems analyze real-time company data to validate assumptions. This shift enhances established investment banks' ability to serve clients by combining trusted senior banker judgment with sophisticated data analytics and modeling capabilities. The technology enables more precise transaction design and execution, reducing the uncertainty in deal outcomes while maintaining the crucial human elements of negotiation and strategic advisory. This represents a fundamental enhancement in capital markets advisory, suggesting a significant evolution in investment banking practices and a reorganization of the investment banking playing field.

Sarah Hammer, Executive Director at the Wharton School and Adjunct Professor at the University of Pennsylvania Law School; Co-Chair of the International Expert Consortium on AI, The Wharton School

[Source](https://www.youtube.com/watch?v=Xri5WZyPcUY)

中文简报

重构交易：人工智能掀起的投资银行革命结合其在JPMorgan、美国财政部、州银行监管机构以及Wharton跨领域的职业经历，Sarah Hammer深入分析了人工智能如何重塑资本市场咨询业务。她详细阐述了大语言模型如何监测系统性流动性压力，并加速演示文稿起草、合规文档编制和财务建模等日常银行工作流程。然而，Hammer明确界定了完全自动化的局限，强调单纯的计算能力无法产生投资信念，也无法取代受托责任。她最终表明，尽管人工智能带来了可衡量的运营效率提升，但交易的成功从根本上依然取决于人类的直觉、信任以及关系资本。沃顿商学院执行主任兼前美国财政部副助理部长 Sarah Hammer 结合其在摩根大通的交易背景与一线银行监管实践，深入探讨了 AI 在投资银行交易中的实际应用与人机协作前景。 (0:04) 2023年3月区域性银行危机中的快速资金外流，促使沃顿研究团队构建领域专用 LLM，利用社交媒体情绪等非结构化数据作为系统性流动性风险的早期预警系统。 (3:08) 尽管高盛与德勤等机构预测 AI 将显著提振宏观经济并为投行带来 27% 至 35% 的生产力增益，但纯粹的数据处理吞吐量仍无法取代承担投资风险所需的制度记忆、个人经验与决策信念。 (14:41) 虽然华尔街机构正在加速部署 AI 技术并聘用顶级数据分析主管，但完成重大复杂交易的关键依然依托于算法无法替代的人际信任资本（relationship equity）与资深顾问判断。 (20:18)

Reading summary

Drawing on executive experience across Wall Street, the U.S. Treasury, and state banking regulation, Sarah Hammer argues that while artificial intelligence significantly enhances operational efficiency and data processing in investment banking, deal execution ultimately depends on human trust, judgment, and relationship equity. Monitoring Liquidity Velocity and Systemic Shocks (3:08) The March 2023 regional banking crisis demonstrated how modern information channels alter the speed of capital flight. The failure of crypto-focused Silvergate Bank preceded the collapse of Silicon Valley Bank by weeks, triggering rapid deposit reallocation as tech and venture founders broadcast their treasury diversification moves in real time across social media. Regulating institutions during this period highlighted the growing friction between legacy oversight frameworks and digital asset liquidity flows. To address this vulnerability, Hammer's research team at the Wharton School developed a specialized large language model designed to analyze unstructured text and real-time social media sentiment. The natural language processing pipeline aims to provide an early-warning mechanism for systemic liquidity stress and bank-run dynamics, translating chaotic public communications into actionable intelligence for regulators, corporate treasuries, and investment strategists. Productivity Gains and Computational Benchmarks (7:02) Macroeconomic and institutional studies project substantial economic uplift from generative AI integration. Goldman Sachs estimated that generative AI could increase global GDP by 7%, while a 2023 Deloitte capital markets study projected a 27% to 35% productivity gain across investment banking operations. Academic research by Shakked Noy and Whitney Zhang (MIT) in 2023, along with Stanford findings, confirmed measurable speed and quality improvements in professional writing tasks. Sector-specific models also demonstrate clear advantages over human baselines in data digestion. A 2023 European study found BloombergGPT outperformed human analysts in evaluating complex financial disclosures, and European Central Bank research confirmed large language models process macroeconomic indicators more effectively than humans. Across transaction workflows, institutions utilize AI tools to automate routine spread adjustments, pitch deck formatting, and initial financial modeling historically assigned to junior analysts. Governance, Board Credibility, and Analytical Limits (10:13) Within corporate governance, AI equips deal teams to synthesize sprawling structured and unstructured datasets into empirical arguments capable of withstanding scrutiny. When executives present disruptive proposals to a board of directors—such as challenging the viability of traditional commercial banking under margin compression and interest rate volatility—or when navigating regulatory approvals for bank mergers, AI-backed compliance analytics provide an evidentiary baseline that builds credibility. However, machine intelligence cannot resolve the conviction gap in capital allocation. While tireless algorithms overcome human cognitive limits in data ingestion, raw calculation cannot manufacture the conviction required to commit institutional capital. Seasoned investment partners continue to rely on lived experience, contextual nuance, and personal accountability to evaluate risk—capacities that automated models cannot provide. The Decisive Role of Relationship Equity (19:34) Capital markets surveys show investment banks, private equity firms, and venture capital companies actively deploying AI while hiring senior analytics executives to oversee portfolios. Despite these computational advances, a debate persists between theoretical economists who anticipate complete automation of human labor and market practitioners who argue that machines cannot replicate genuine trust, empathy, or emotional perception. While attendees noted human intuition is prone to error and personal bias, institutional clients remain hesitant to delegate fiduciary authority entirely to autonomous agents. Because financial transactions ultimately depend on fiduciary responsibility, personal credibility, and human rapport, technical execution cannot replace the "relationship equity" required to close complex deals. While artificial intelligence continues to expand institutional capacity, model synergies, and detect obscure risks, the industry faces an unresolved tension over client trust: whether market participants will ultimately entrust fiduciary authority to autonomous models or continue to rely on seasoned practitioners with multi-decade track records during high-stakes financial crises.

精炼阅读摘要

前美国财政部官员、沃顿商学院学者萨拉·哈默（Sarah Hammer）指出，尽管人工智能在宏观分析、流动性压力监测与日常投行工作流中展现出显著的生产力飞跃，但复杂的交易达成与资本托付依然建立在人类经验、直觉判断及无可替代的“关系资产”基石之上。宏观预测与算力在金融分析中的实证优势 (7:02) 在量化评估人工智能对宏观经济与金融机构的赋能方面，多项基准研究展现出明确的算力优势。高盛的分析预测，生成式人工智能的深度应用有望推动全球GDP增长7%；麦肯锡的研究亦预计该技术将为投资银行业创造巨大的生产力增量。在专业信息处理基准中，2023年欧洲一项实证研究显示专用模型BloombergGPT在综合与分析复杂财务披露方面的表现超越了人类分析师；欧洲央行的研究同样证实，前沿大语言模型在消化与解读宏观经济指标时已系统性超出人类基线。相较于人类受限的认知吞吐量，机器能够不知疲倦地接入离散数据源并提取洞见，重构了金融研究的信息处理效率。银行业危机流动性建模与董事会治理实证 (3:08) 结合2023年3月主政宾夕法尼亚州银行与证券部应对区域银行业动荡的监管履历，哈默指出即时通讯与数字资产通道显著提升了流动性流失的速度。从加密友好型银行Silvergate的清算到硅谷银行的存款挤兑，初创企业在社交媒体上实时公开资金转移路径，推动资金流向具备牌照的传统区域银行。为应对非结构化数据带来的监测挑战，沃顿商学院研发了专用大语言模型，利用社交情绪数据建立系统性流动性压力的预警雷达。在微观治理层面，面对董事会与监管机构的严格审视，投行团队可借助AI对海量结构化与非结构化数据进行实证整合，为颠覆性的并购方案或战略转型提供能够经受住质询的数据基准。日常业务增效与初级分析师的角色重塑 (14:41) 人工智能正快速融入投行的日常工作流，改变推介材料、投资备忘录等传统交付物的制作流程。麻省理工学院研究人员（Noy & Zhang，2023）与斯坦福大学的学术成果均证实，AI在专业文本撰写任务中带来了显著的速度与质量提升；德勤2023年针对资本市场部门的评估则预测，将人工智能部署于投资银行业务将带来27%至35%的整体生产力增长。尽管算法能够自动化处理财务模型调整、利差测算和幻灯片版式设计，承担过去一二年级分析师的部分繁重工作，但哈默强调AI缺乏内在的叙事视角与个人洞察；初级员工的角色正从基础的信息搜集转向分析结果的实证核验，技术的普及实质上扩大了金融机构的整体承载力。算力无法跨越的信念鸿沟与关系资产 (17:10) 尽管算法模型展现出卓越的计算速度，但涉及高额资金博弈时，技术无法自动生成“投资信念”（conviction）或替代信义信任。哈默明确反驳了经济学者安东·科里内克（Anton Korinek）关于“AI将复制人类所有能力”的理论预判，坚持认为机器无法复刻人类之间真实的情感理解与信任连结。面对复杂交易与系统性危机，资深合伙人敢于承担资本风险依据的是数十年一线经历所积累的制度记忆与直觉。资本市场调查显示，虽然投行、私募与风险投资机构正积极布局AI并任命高层数据分析专家以捕捉协同效应与潜在风险，但促成交易闭环的核心货币仍然是需要面对面沟通建立的“关系资产”（relationship equity），客户最终仍依赖可承担责任的专业人士而非自主运行的算法代理。哈默的论证表明，投资银行领域的AI变革是人机协同与产能跃迁，而非对人类专业判断的单向取代。尽管以GPT-5等为代表的工具大幅降低了复杂调研与单次提示分析的成本，但在道德风险、非理性市场挤兑以及重大利益博弈面前，人类拥有的市场阅历、人际互信与战略定力依然是促成金融交易不可让渡的核心支柱。

Detailed notes — zh

## Background 本内容由 YouTube 频道 Ai4 于 2025 年 8 月 28 日发布。视频主题为《重塑交易撮合：AI 驱动的投资银行革命》，主要由沃顿商学院执行总监兼宾夕法尼亚大学法学院兼职教授 Sarah Hammer 探讨先进技术与传统投行咨询专业知识的融合演进，剖析 AI 如何通过预测建模、数字孪生模拟和自主尽调来赋能交易设计与执行。 ## TL;DR Sarah Hammer 结合自身在财政部、州银行监管层、摩根大通交易台及沃顿商学院的跨界经历，系统拆解了 AI 在投资银行与金融交易中的实际应用。她回顾了 2023 年地区性银行业危机中如何利用 LLM 监测社交媒体非结构化情绪以预警流动性挤兑，并引用高盛、麦肯锡、欧洲央行等权威数据证实 AI 在提升 GDP 和投行效率方面的颠覆性潜力。然而，她强调尽管 AI 极大解放了初级分析师的数据吞吐与机械性劳动，但‘投资信念’、复杂董事会游说以及人与人之间深层次的‘关系资产’与信任依然是不可替代的护城河，未来的胜局属于善用 AI 增强判断力的专业交易家。 ## Atmosphere & style 整场演讲兼具顶层监管者的审慎宏观视角与一线交易员的敏锐实战感，语调沉稳、睿智且富有亲和力。讲者通过抛出‘你敢不敢把资产交给 AI 打理’、‘你更信任 Claude 还是资深专家’等尖锐问题与现场观众高频互动，巧妙营造出思辨与辩论的火花。现场氛围既有对前沿算法算力的理性推崇，又流淌着对人际温度、情感共鸣与直觉价值的坚定捍卫，给观众带来极具启发性的人机协同反思。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=4s) Sarah Hammer 结合美国财政部、宾州银行监管及沃顿商学院的多重实操与学术背景，探讨 AI 在投行领域的实际应用落地与人机协同模式。 - [03:49](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=229s) 2023 年银行业危机中资金在社交媒体发酵后极速外逃，沃顿团队据此开发了基于 LLM 的流动性监测模型，利用非结构化文本情绪指标预测挤兑风险。 - [07:02](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=422s) 高盛预测生成式 AI 将拉动全球 GDP 增长 7%，彭博与欧洲央行的研究亦证实 LLM 在财务披露拆解和宏观经济数据解读上已超越人类分析师。 - [07:02](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=422s) 德勤研究表明 AI 应用可使投行业务收益与效率提升 27% 至 35%，主要通过自动化替代繁琐重复工作并放大承揽产能。 - [09:02](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=542s) 尽管技术乐观派认为 AI 终将替代一切，但人与人之间的信任共鸣、董事会层面的战略说服以及合规监管沟通仍是不可替代的人类护城河。 - [11:44](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=704s) 人类大脑数据吞吐量受限，而 AI 虽能全天候处理海量非结构化数据，却缺乏基于长期行业洞察形成的“投资信念”（Conviction）。 - [17:10](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=1030s) 尽管普通人直觉容易存在认知偏差，但金融交易的底层依然高度依赖深层信任机制，纯客观胜率模型无法完全替代人际信任。 - [20:18](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=1218s) 投行与 PE/VC 机构正在全流程加速部署 AI 工具与数据专家，但交易撮合的核心依然取决于面对面互动沉淀的“关系资产”（Relationship Equity）。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=4s) 讲者背景与跨界经历：从美国财政部、宾州银行监管到沃顿商学院 Sarah Hammer 在开场时准时登台，向现场观众介绍了自己跨越学术界、政府监管与私营部门的多重背景。她曾任沃顿商学院（Wharton School）执行总监、宾夕法尼亚大学法学院（Penn Law）兼职教授，并担任沃顿 Cypher 加速器（聚焦于 AI、区块链与量子计算）的 CEO。在演讲前数周，她刚刚重新回到私营金融领域。她强调，今天不仅要讨论 AI 在投资银行业“能做什么”，更要结合前沿研究与一线实操，探讨全球金融业究竟在“如何实际应用 AI”，以及人类专业人士如何与 AI 协同工作以释放其最大潜力。 Hammer 的职业生涯横跨宏观政策制定与高强度市场一线。在加入沃顿之前，她曾担任美国财政部负责金融机构事务的副助理部长（Deputy Assistant Secretary for Financial Institutions），主管金融机构政策办公室、联邦保险办公室（包括恐怖主义风险保险计划 TRIA）以及网络安全与合规政策办公室。2023 年，她受宾夕法尼亚州州长任命，休假出任宾夕法尼亚州银行与证券厅厅长（Secretary of Banking and Securities），直接在一线带领监管机构应对了 2023 年震惊全美的地区性银行业危机。在此之前，Hammer 拥有超过 25 年的金融市场实战经验，曾在摩根大通（J.P. Morgan）从事交易业务，并长期深耕投资策略、投资组合分析、宏观经济研究，还曾掌管面向大学捐赠基金与慈善基金会的投资业务。在国际治理方面，她联合主持了一个专注于人工智能经济影响、监管政策与工程解决方案的国际专家财团（Wharton AI Consortium），同时担任布雷顿森林委员会（Bretton Woods Committee）AI 工作组联合主席，并参与美国法学会（American Law Institute, ALI）关于 AI 民事责任的咨询工作。 ```mermaid flowchart TD accTitle: Sarah Hammer 职业背景与专业维度 subgraph Regulatory["政府与监管背景"] A1["美国财政部 副助理部长"] A2["宾夕法尼亚州 银行与证券厅长"] end subgraph Academic["学术与前沿研究"] B1["沃顿商学院执行总监 宾大法学院教授"] B2["沃顿 Cypher 加速器 CEO 国际 AI 专家财团联席主席"] end subgraph Industry["金融市场实操"] C1["摩根大通交易员 投资组合与宏观研究"] C2["捐赠基金与基金会 投资业务负责人"] end Regulatory --> D["AI 时代的投行革命与人机协同"] Academic --> D Industry --> D ``` --- ### [03:49](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=229s) 2023年银行业危机复盘：社交媒体情绪与AI流动性监测模型谈到将 AI 应用于银行与交易决策，Hammer 结合自身在财政部现金管理室以及在 2023 年 3 月银行业危机董事会一线处理危机的经历展开了剖析。在硅谷银行（SVB）爆发挤兑并倒闭之前数周，知名加密友好银行 Silvergate（曾接管部分 FTX 资产）已率先破产清算。随后，当 SVB 显露危机迹象时，来自加密社区以及大量初创企业（包括 AI 创企）的数十亿美元流动性开始在极短时间内疯狂外逃并转移。作为当时主管宾夕法尼亚州银行体系的最高官员，Hammer 在后台亲眼目睹了这些资金迅速流入其管辖的宾州银行账户。当时许多创始人与高管在社交媒体（如 X/Twitter）上实时同步自己如何拆分资金、转移存款以避险。这段极具冲击力的监管实战经历促使 Hammer 随后在回到沃顿商学院后，主导团队研发了一款基于大型语言模型（LLM）的流动性监测系统——通过抓取和深度分析社交媒体数据与非结构化文本的情绪指标，预测银行业的流动性压力与潜在挤兑危机。 - **危机传导链条**：Silvergate 破产率先引发警报，数周后 SVB 遭遇流动性危机，促使大量初创公司与 AI 公司利用数字化渠道跨行调拨巨额存款。 - **非结构化数据金矿**：高管与企业在社交网络上公开的资金转移行为，构成了评估系统性流动性压力的实时风向标。 - **沃顿 LLM 流动性模型**：利用大模型解析社交情绪，试图破解金融界苦苦追寻的“危机预警公式”。 --- ### [07:02](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=422s) AI赋能金融的实证数据：GDP拉动、大模型表现与投行增效为了展示 AI 带来的宏观及行业变革，Hammer 引用了数家顶级研究机构与中央银行的权威实证数据，并与现场观众展开了即兴互动：高盛（Goldman Sachs）的研报测算，生成式 AI 的广泛应用将推动全球 GDP 增长达 7%，这一贡献幅度超越了过去数十年间出现的绝大多数单项技术革新。麦肯锡（McKinsey）关于 AI 在投资银行业价值评估的研究同样证实了其带来的生产力飞跃。欧洲在 2023 年的一项专项研究显示，金融专用大模型 BloombergGPT 在解析复杂的财务披露报告（Financial Disclosures）方面已经超越了人类分析师的综合表现。此外，欧洲中央银行（ECB）发布的研究报告指出，大语言模型在宏观经济数据的解读与模式识别能力上优于人类专家。这些前沿研究与 Hammer 团队在沃顿构建的银行危机模型相互印证：在处理海量结构化与非结构化数据、识别行业潜在交易机会方面，AI 展现出了颠覆性的潜力。 | 机构 / 来源 | 核心研究发现 | 对投资银行与宏观经济的影响 | |:--- |:--- |:--- | | **高盛 (Goldman Sachs)** | AI 预计将推动全球 GDP 增长 7% | 带来超越过去数十年传统产业技术周期的宏观增长动力 | | **彭博 (BloombergGPT 欧洲研究, 2023)** | 在财务信息与披露文件解析上超越人类分析师 | 彻底改变投行尽职调查与财务报表拆解的效率基准 | | **欧洲中央银行 (ECB)** | LLM 在经济数据解读与趋势推演上优于人类 | 提供更为精准的宏观周期研判与系统性风险监控 | | **德勤 (Deloitte, 2023)** | AI 应用可使投行业务收益与效率提升 27% 至 35% | 自动化替代繁琐机械工作，放大交易执行与承揽产能 | --- ### [09:02](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=542s) 机器无法替代的核心维度：与经济学家 Anton Korinek 的思想交锋在呈现完 AI 的强大算力后，Hammer 向全场提问：“有谁认为 AI 可以包办一切？”面对台下举手的观众，她分享了自己与著名经济学家、AI 领域权威专家 Anton Korinek（同为沃顿 AI 专家财团成员）的一场风趣而深刻的辩论。Korinek 持彻底的技术乐观主义观点，坚信 AI 终将取代人类能够完成的所有工作。Hammer 曾当面反驳他：“AI 永远无法替代人与人之间的真实关系，无法替代我们建立的信任，也无法替代我们注视彼此时所产生的情感共鸣。”然而 Korinek 看着她微笑着回答：“我很抱歉，Sarah，但 AI 连这一切都能做到。” Hammer 幽默地表示自己并不赞同 Korinek 的判断，并强调至少在现阶段，人类决策与真实情感仍是不可撼动的护城河： 1. **交易团队协同**：AI 负责拉取、清洗并提炼复杂数据，协助交易团队发现盲点，但最终的商业博弈与团队协同需要人类作出决断。 2. **董事会沟通与说服**：向董事会汇报战略方向时（例如根据银行高监管压力、利差压缩、利率波动等数据建议转型 FinTech），AI 可以生成扎实的数据支撑，但承担决策后果与引导董事会达成共识的必须是人类高管。 3. **监管者对话与对齐**：作为前监管官员，Hammer 指出，在向监管层报备复杂并购（M&A）或资本交易时，详尽的 AI 数据模型固然是关键说服工具，但合规信任与政策沟通依然建立在人与人的深层沟通之上。 > "AI 无法替代人与人之间的关系，无法替代我们之间的信任，更无法替代人类在真实互动中所感受到的情绪与直觉。" —— Sarah Hammer --- ### [11:44](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=704s) 投资人视角下的分析师重塑：数据吞吐极限与“投资信念”的本质在沃顿商学院执教期间，Hammer 常被学生问及一个现实而焦虑的问题：“AI 会彻底取代初级分析师（Analysts & Associates）吗？”在现场讨论中，一位观众指出：人类大脑的“数据吞吐量”（Throughput）存在物理极限，而 AI 能够不知疲倦地全天候监测所有数据源并提炼洞察，处理非结构化数据（社交网络发帖、聊天记录、文本资料）的能力远超人类。 Hammer 对此表示赞同，但也指出了 AI 存在无法跨越的认知鸿沟——**“投资信念”（Conviction）**。在投资银行与风险投资（VC）中，分析师需要从数百上千家初创企业或标的公司中筛选线索、构建投资逻辑（Investment Thesis），但 AI 无法凭空产生对某笔交易的坚定信念。即使 AI 呈现出完美的数据模型，如果资深银行家基于自身的行业经历或人际接触察觉到异样，人类依然会推翻机器的判断。关于投行日常中的 PPT 制作、投资备忘录（Deal Memos）与投委会汇报文件撰写（正如 Noy 在 2023 年研究所指出的生产力提升），AI 已经能够高效代劳。Hammer 坦言，自己本次演讲的幻灯片版式与部分思路也是在 AI 辅助下搭建的，但其中的核心逻辑、亲身执政经历与行业故事，完全源自她本人。德勤 2023 年的研究表明，AI 可使投行业务收益提升 27% 到 35%，这证明 AI 消除的是机械性重复劳动，而非投行家的核心判断力。 --- ### [17:10](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=1030s) 人类优势与信任悖论：直觉偏差、GPT-5 与投资决策中的人机博弈在半程的最后环节，Hammer 深入探讨了“人类直觉”与“信任建立”的本质。她提到了最新发布的 GPT-5 与 Deep Research 的对比。现场有观众表示更青睐 GPT-5，Hammer 也深有同感，认为 GPT-5 在交互门槛、单提示词（Single-prompt）响应速度以及性价比上更具优势，能够在更短时间内提供结构化的深入信息。随后，Hammer 抛出了一个直击痛点的问题：“现场有谁敢把自己的全部投资组合交给 AI 打理？” 一位观众当场回应称，自己愿意信任 AI，甚至“比信任普通人类更信任 AI”。该观众阐述理由道，绝大多数普通人的所谓“直觉”至少有一半时间是错误的，而且大多数普通人在专业基准测试中连 AI 能做对的题目都无法解答。 Hammer 顺着这个观点展开了精彩的辩证分析：人们或许会相信 AI 的理性计算胜过平庸的人类，但真正的金融交易依托的是深层信任机制。人们会无条件信任至亲好友，但绝不会在没有建立深度信任的前提下，把毕生积蓄交给台上素未谋面的陌生人。AI 能够提供客观的胜率，但它既没有真实经历带来的顿悟，也无法在瞬息万变、充满博弈的交易谈判中取代人类所构建的信任基石。这一关于数据理性与人类信任边界的讨论，将全场气氛推向高潮。 ### [19:34](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=1174s) 人类经验与 AI 模型的信任博弈：你更信任资深专家还是 Claude？在探讨投行业务的演进时，人与机器之间的信任天平始终是一个微妙的话题。有些人确实表现出更倾向于信任 AI 而非人类的倾向，这种现象完全可以理解——毕竟 AI 是基于数据驱动的纯粹算法模型，而人类拥有复杂的情感，往往更容易受到情绪干扰并犯下错误。在复杂的交易撮合（Deal-Making）场景中，究竟应该彻底倒向数据模型还是依赖人类直觉，似乎成了一个悬而未决的分水岭。从个人视角出发，“人性的温度”（Human touch）依然是交易中不可替代的核心要素。拥有长达 25 年跨领域经验的专业人士——身兼监管者、学者、商业领袖、交易员以及投资组合经理等多重背景，其所积累的直觉与洞察具有独特的权威感。演讲者在此向台下观众抛出了一个引人深思的问题：“如果你现在想就银行业危机提出几个核心问题，你究竟会更信任坐在台前的资深专家，还是去向 Claude 寻求答案？” ```mermaid flowchart TD accTitle: AI 分析能力与人类专业判断的对比 subgraph AI["AI / 模型视角"] A1["纯粹算法与数据驱动"] A2["无情绪波动与认知偏差"] A3["处理速度极快与模式识别"] end subgraph Human["人类专家视角"] B1["跨领域实战履历 (监管/交易/学术)"] B2["人际共情与谈判温度"] B3["复杂危机下的情境研判"] end AI <-->|信任博弈与互补融合| Human ``` --- ### [20:18](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=1218s) 为何无法将交易完全托付给 AI？“关系资产”的不可替代性既然 AI 能够以超越人类的速度和精度处理海量数据、在模型中测算并实现协同效应（Synergies），还能敏锐标记出人类容易遗漏的问题盲点，那么为什么我们不能把并购与交易完全交给 AI 全自动执行？答案在于交易撮合中最容易被忽视、却最关键的维度——“关系资产”（Relationship equity）。 ```mermaid flowchart LR accTitle: 现代投资银行交易架构模型 subgraph Tech["智能技术支撑层"] T1["高算力数据分析"] T2["协同效应预测建模"] T3["实时风险与异常标记"] end subgraph Core["核心交易达成"] C1["高层战略研判"] C2["关系资产 (Relationship Equity)"] C3["信任构建与最终决策"] end Tech -->|赋能与加速| Core ``` 行业调研与来自金融服务业的一手信号显示，投资银行、私募股权机构（PE）和风险投资机构（VC）正全速在投资全流程中部署 AI，甚至纷纷在管理顶层设立首席数据分析专家以深度审查投资组合。然而，无论技术如何演进，交易的本质依然依托于人与人之间通过深入互动建立起来的信任壁垒： - **技术部署加速**：华尔街各大投行、PE 与 VC 机构正迅速落地 AI 工具，并聘请顶尖数据分析专家对资产组合进行多维扫描。 - **关系资产的沉淀**：正如演讲者在现场与观众交流的 20 分钟一样，正是这种面对面的思想碰撞与交流，才积累起远胜于单纯算法输出的“关系资产”，进而促成后续深层次的商业沟通与业务合作。 ## Notable quotes > [09:02](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=542s) AI 永远无法替代人与人之间的真实关系，无法替代我们建立的信任，也无法替代我们注视彼此时所产生的情感共鸣。 — Sarah Hammer > [09:02](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=542s) 我很抱歉，Sarah，但 AI 连这一切都能做到。 — Anton Korinek > [09:02](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=542s) AI 无法替代人与人之间的关系，无法替代我们之间的信任，更无法替代人类在真实互动中所感受到的情绪与直觉。 — Sarah Hammer > [09:02](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=542s) 有谁认为 AI 可以包办一切？ — Sarah Hammer > [11:44](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=704s) AI 会彻底取代初级分析师吗？ — 沃顿商学院学生 > [17:10](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=1030s) 现场有谁敢把自己的全部投资组合交给 AI 打理？ — Sarah Hammer > [17:10](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=1030s) 比信任普通人类更信任 AI，因为绝大多数普通人的所谓直觉至少有一半时间是错误的。 — 现场观众 > [19:34](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=1174s) 如果你现在想就银行业危机提出几个核心问题，你究竟会更信任坐在台前的资深专家，还是去向 Claude 寻求答案？ — Sarah Hammer > [20:18](https://www.youtube.com/watch?v=Xri5WZyPcUY&t=1218s) 交易的本质依然依托于人与人之间通过深入互动建立起来的信任壁垒。 — Sarah Hammer ## Mentioned Sarah Hammer, Anton Korinek, 沃顿商学院 (The Wharton School), 宾夕法尼亚大学法学院 (Penn Law), 美国财政部 (U.S. Department of the Treasury), 宾夕法尼亚州银行与证券厅 (Pennsylvania Department of Banking and Securities), 摩根大通 (J.P. Morgan), 高盛 (Goldman Sachs), 麦肯锡 (McKinsey), 德勤 (Deloitte), 彭博 (Bloomberg / BloombergGPT), 欧洲中央银行 (ECB), 硅谷银行 (SVB), Silvergate, GPT-5, Claude, Deep Research, 布雷顿森林委员会 (Bretton Woods Committee), 美国法学会 (American Law Institute) ## Video description Deal-Making Reimagined: AI's Investment Banking Revolution. Investment banking is evolving through the integration of advanced technology with traditional advisory expertise. Modern transactions require analysis of complex data sets that exceed human processing capabilities - from real-time market signals to cross-border regulatory requirements. The conventional banking model can be significantly enhanced by AI systems that analyze global companies continuously, identifying potential transactions through predictive modeling of corporate trajectories and strategic fits. This approach complements relationship-driven deal-making with systematic opportunity identification - recognizing optimal timing and structure for M&A, IPO, and financing transactions. Digital twin simulation enables testing of thousands of potential deal outcomes before approaching companies, while autonomous due diligence systems analyze real-time company data to validate assumptions. This shift enhances established investment banks' ability to serve clients by combining trusted senior banker judgment with sophisticated data analytics and modeling capabilities. The technology enables more precise transaction design and execution, reducing the uncertainty in deal outcomes while maintaining the crucial human elements of negotiation and strategic advisory. This represents a fundamental enhancement in capital markets advisory, suggesting a significant evolution in investment banking practices and a reorganization of the investment banking playing field. Sarah Hammer, Executive Director at the Wharton School and Adjunct Professor at the University of Pennsylvania Law School; Co-Chair of the International Expert Consortium on AI, The Wharton School [Source](https://www.youtube.com/watch?v=Xri5WZyPcUY)

Source quotes

[9:02] Sarah Hammer: AI 永远无法替代人与人之间的真实关系，无法替代我们建立的信任，也无法替代我们注视彼此时所产生的情感共鸣。 [9:02] Anton Korinek: 我很抱歉，Sarah，但 AI 连这一切都能做到。 [9:02] Sarah Hammer: AI 无法替代人与人之间的关系，无法替代我们之间的信任，更无法替代人类在真实互动中所感受到的情绪与直觉。 [9:02] Sarah Hammer: 有谁认为 AI 可以包办一切？ [11:44] 沃顿商学院学生: AI 会彻底取代初级分析师吗？ [17:10] Sarah Hammer: 现场有谁敢把自己的全部投资组合交给 AI 打理？ [17:10] 现场观众: 比信任普通人类更信任 AI，因为绝大多数普通人的所谓直觉至少有一半时间是错误的。 [19:34] Sarah Hammer: 如果你现在想就银行业危机提出几个核心问题，你究竟会更信任坐在台前的资深专家，还是去向 Claude 寻求答案？ [20:18] Sarah Hammer: 交易的本质依然依托于人与人之间通过深入互动建立起来的信任壁垒。

Source quotes

[7:02] Anton Korinek: I'm really sorry, Sarah, but it can do all of that. [7:02] Sarah Hammer: Goldman Sachs estimated that generative AI could drive a 7% increase in global GDP, a macroeconomic surge rivaling the transformative technological shifts of prior generations. [10:13] Sarah Hammer: An analyst may surface an attractive investment thesis, but seasoned partners rely on lived experience, relational context, and institutional memory to decide whether to risk capital—qualities an automated model cannot replicate. [14:41] Sarah Hammer: AI functions as an efficient drafting and design partner, but it possesses no inherent narrative perspective. [14:41] Sarah Hammer: Even as algorithms automate routine financial modeling, spread adjustments, and initial deck formulation historically assigned to first- and second-year analysts, the net effect elevates total institutional capacity rather than diminishing the industry's economic output. [17:10] Sarah Hammer: Maybe your close family, your best friend, but beyond that, do you trust me? Give me your money. I'll invest it for you... All right, so if we trust you, we'll give you all our money after this presentation. [17:10] Sarah Hammer: Mathematical competence alone does not bridge the emotional divide required for clients to relinquish control of capital without a trusted human intermediary. [19:34] Sarah Hammer: AI operates strictly as a model, whereas human beings are driven by emotion and are inherently prone to making mistakes. However, when evaluating complex transactions where nuance and accountability are paramount, the human touch remains indispensable. [20:18] Sarah Hammer: While technology can model outcomes, deals ultimately close through human rapport, strategic advisory, and mutual confidence. [20:18] Sarah Hammer: While machine intelligence optimizes analytical precision and risk detection, relationship equity and seasoned judgment remain the deciding factors in strategic client negotiations.

Original transcript

[0:04] get started right on time uh to keep the [0:06] conference moving. I'm Sarah Hammer. Uh [0:09] I'm really excited to be here today and [0:12] uh I am I was executive director at the [0:16] Wharton School and a professor at the [0:18] University of Pennsylvania Law School, [0:20] uh CEO of Wharton Cipher Accelerator and [0:23] AI uh blockchain and quantum computing [0:26] accelerator. and I recently returned to [0:29] the private sector, in fact, just in the [0:31] past few weeks. So, I'm excited to be [0:33] here today to talk to you about [0:35] investment banking and about artificial [0:38] intelligence. And I'll be talking about [0:40] not just what AI can do in investment [0:44] banking, but what the industry is [0:47] actually using it for and based on [0:49] research and my experience, what we are [0:53] seeing internationally in terms of how [0:55] we as businesses can position ourselves [0:58] to use AI and to work alongside it as [1:02] humans to maximize its potential. So, [1:05] I'm going to talk a little bit about my [1:07] experience as I go through the [1:08] presentation. Uh, but just a quick [1:10] overview. Prior to Wharton, uh, I was at [1:13] US Treasury as deputy assistant [1:15] secretary for financial institutions. I [1:18] oversaw the office of US financial [1:20] institutions policy, the federal [1:23] insurance office, including the [1:24] terrorism risk insurance program, and [1:27] uh, the office of cyber security and [1:29] compliance policy. While at Wharton in [1:32] 2023, I took a short-term leave of [1:35] absence to serve as secretary of banking [1:37] and securities uh for Pennsylvania. I [1:40] was appointed by the governor and I led [1:42] the department through the banking [1:44] crisis of 2023. So if we may not have [1:48] time in this presentation, but if any of [1:50] you would like to talk about what it was [1:52] like to lead through that period, not [1:55] just for banks, but as regulators and of [1:57] course the technology companies that we [1:59] saw moving cash during that time, I'd be [2:02] glad to do it. Uh but prior to that I [2:05] was in financial services, in trading at [2:08] a large bank uh JP Morgan, in investment [2:11] strategy, portfolio analytics, a [2:14] macroeconomic research, and I ran a [2:16] business on the endowment and foundation [2:18] investment side. So I'm going to bring [2:20] my experience and the research today [2:23] into the presentation and I'm looking [2:26] forward to talking with all of you about [2:27] that. I'll just mention briefly also I [2:30] co-chair an international uh an [2:33] international consortium of experts on [2:35] artificial intelligence that are looking [2:37] not just at regulation but also at the [2:40] economic impact of artificial [2:42] intelligence and engineering solutions [2:45] that will bring us closer to [2:47] understanding and addressing some of the [2:49] issues that you all are learning about [2:51] today at this conference. I'm also [2:54] co-chair of the in uh artificial [2:55] intelligence working group uh at the [2:58] Bretonwoods committee and I serve on the [3:00] member consultative group for civil [3:02] liability for AI at the American Law [3:05] Institute. All right, so enough about [3:08] me. Uh I'm going to talk a little bit [3:10] about those different perspectives that [3:13] I mentioned and we're going to continue [3:14] throughout the presentation to talk [3:17] about the human element and why it is [3:20] crucial even when we use AI and finance [3:23] especially I would pause it to say [3:26] especially and we need the human element [3:28] in finance and this is an area where a [3:31] lot of folks think it's very data [3:33] intense AI is a great prospect for [3:36] helping us solve some of the problems s [3:38] dealing with both structured and [3:40] unstructured data as you heard in the [3:42] last presentation. Uh but we need the [3:45] human element and we're going to talk a [3:46] little bit more about why. So as uh [3:50] deputy assistant secretary at US [3:52] Treasury, the perspective I bring to the [3:55] table is sitting in the cash room at [3:58] Treasury and talking with industry [4:00] folks, especially banks about how they [4:03] are engaging with the regulators, what [4:06] challenges they face based on the [4:08] existing framework and how they are [4:10] using technology. As secretary of [4:13] banking, I mentioned the perspective [4:15] that I bring is from the boardroom [4:18] during the March 2023 banking crisis and [4:22] beyond and seeing technology companies [4:25] both crypto as was discussed in the last [4:27] round table uh and the role of [4:30] artificial intelligence and social media [4:32] of course in moving liquidity during [4:35] that time. So you all recall during the [4:38] banking crisis uh SVB experienced stress [4:42] but prior to that there was a bank [4:44] called Silvergate that also failed. Uh [4:46] Silvergate was a well-known crypto bank. [4:49] In fact it had taken over some of the [4:51] assets of FTX. Uh and the failure of [4:55] Silvergate was actually just weeks [4:57] before the failure of SVB. At the time [5:00] that SBB began to experience stress, a [5:04] large amount of cash, both from the [5:06] crypto community and from many other [5:08] startups, including AI startups, move to [5:12] different banks. And as Secretary of [5:14] Banking and Securities, I saw some of [5:17] that cash move to my Pennsylvania banks. [5:20] And that's public information uh because [5:23] a lot of those companies were actually [5:24] posting on social media how they were [5:27] handling that cash. So one application [5:30] that I'll mention of AI in the banking [5:34] space that I helped build later when I [5:38] was back at Wharton was a large language [5:40] model to evaluate banking liquidity [5:43] considerations especially in stress [5:46] periods based on social media data. And [5:49] that's the magic formula that we've all [5:51] been looking for. How can we know that a [5:54] crisis is coming based on data on social [5:58] media on sentiment uh and it's a very [6:01] challenging opportunity. Uh but I do [6:04] think that artificial intelligence uh [6:06] has brought us a long way in that. All [6:09] right. I mentioned I was a trader and [6:10] portfolio manager and I will say that as [6:13] far as deal making goes, this is an [6:16] experience that I had uh in my more than [6:19] 25 years of experience in the industry [6:22] that helped me understand what is the [6:24] human element in dealmaking and we're [6:26] going to talk a little bit more about [6:28] that. and then as CEO of Cipher [6:31] Accelerator at Wharton, which was [6:34] identifying and scaling AI, blockchain [6:38] and other emerging technology companies. [6:40] AI was essential not just for those [6:44] companies and their business models, but [6:46] for us in understanding the industry and [6:49] that applies to what we are looking for [6:51] in investment banking. Can we use data [6:54] to learn about the industry to develop [6:56] an investment thesis and to make great [6:59] investment decisions? [7:02] Okay, I'm sure that all of you have uh [7:05] heard and learned a lot today about the [7:08] potential for AI. So, I'm just going to [7:10] briefly poll the audience. Um, but who's [7:13] familiar with some of these statistics? [7:15] How will AI help economic growth? [7:20] Right? So Goldman Sachs had a study [7:22] recently that said that AI will help [7:25] increase global GDP by 7%. That's an [7:30] incredible number. Incredible. That's a [7:33] bigger contribution than we've seen in [7:35] many other industries over the past [7:37] several decades. AI and investment [7:39] banking, the value there actually was [7:42] assessed by McKenzie also in a recent [7:44] study. Who's familiar with that [7:46] statistic? [7:48] I'm happy to share the citations after [7:50] the presentation if all of you are [7:52] interested. Um, and then Bloomberg GPT [7:56] outperformed based on a study in Europe [7:59] the analysis of financial disclosures. [8:02] Another area where the vast amount of [8:05] information that we all receive that we [8:07] all have access to can be synthesized [8:10] and better understood by AI. And in [8:13] fact, the study said um that it [8:15] outperforms human beings. And that study [8:17] was done in 2023. [8:19] All right. And then large language [8:21] models of course outperform an economic [8:24] interpretation and that was done by an [8:26] official resource the uh European [8:29] Central Bank. And again I'm happy to [8:31] share that uh study with all of you. Uh [8:33] but I thought it was interesting because [8:35] again comparing the large language model [8:38] that my team and I built uh after the [8:41] banking crisis at Wharton with the [8:44] findings of the ECB tell us there's [8:47] incredible potential to use AI to get [8:50] deals done to increase global domestic [8:53] product or uh global domestic product [8:56] right gross domestic product on a global [8:58] level but based on experience what can [9:02] AI I not do. So, who has some thoughts [9:05] on what AI cannot do? Or do you all [9:08] think that AI can do everything? [9:12] Who thinks AI can do everything? [9:17] Wow, there's always a couple that are [9:19] like AI can do everything. So, uh I [9:22] actually have a colleague friend Anton [9:24] Cornuk who's a world-renowned economist [9:27] and expert in artificial intelligence. [9:29] You can look him up. uh he's in the AI [9:32] consortium that I co-chair and Anton [9:35] believes that AI will replace everything [9:38] that humans can do. And I said to Anton [9:41] very recently, but AI cannot replace the [9:45] human relationship. It cannot replace [9:47] the trust that we have with each other. [9:50] It cannot replace emotions that we feel [9:53] when we see someone. All of you have a [9:56] perception or emotions watching me right [9:58] now. And Anton looked at me and he said, [10:01] "I'm really sorry, Sarah, but it can do [10:03] all of that." So, you know, I'm I'm not [10:07] gonna worry about when that day comes. [10:09] I'm just gonna enjoy it. But I don't [10:11] agree with Anton. [10:13] What can AI do? Well, it can help us [10:16] with data, right? We know that it can [10:19] help us with finding and interpreting [10:22] information. And so that means it can [10:25] help us align with our deal teams to [10:28] look at the data to analyze it and [10:31] discuss it as a team, but it can't make [10:34] human decisions for us. Only humans can [10:37] do that. Uh I suppose if you had a chip [10:39] implanted in your brain, you could have [10:41] AI make the decision for you, but we're [10:44] not there yet for all of us, right? It [10:46] can help us work with our boards. If I [10:49] go before my board and I say, "Here's [10:51] the data that I am seeing in banks and [10:55] this is what it's telling me. Banks are [10:57] not going to grow. They're too highly [10:59] regulated. They're too compressed. The [11:02] liquidity conditions are hard. The [11:04] interest rate environment is difficult [11:06] and it's too volatile for banks to be [11:09] successful in the long run. Everybody [11:11] move to FinTech." Right? I'm not going [11:13] to go to my board and say that unless I [11:16] have really good data and AI might be [11:19] able to help me make that argument and [11:22] aligning with the regulators. So, I can [11:24] tell you as a former regulator, if you [11:27] come in to talk to me about a deal or a [11:30] challenge or maybe a merger, having that [11:33] data and analysis from AI is key. And AI [11:37] is an incredible tool to help us reach [11:40] that alignment as we move forward. [11:44] Okay. So, what about the investor's [11:46] lens? Uh, at Warden, I have to tell you, [11:49] I have so many incredible students and [11:53] they have gone on to do amazing things. [11:56] They are leaders in banks, in consulting [11:59] firms, uh, in tech companies, in the [12:02] digital asset space, in artificial [12:04] intelligence all around the world. And [12:08] the question that I'm often asked is, is [12:11] AI going to replace their jobs? Right? [12:15] That's a common question. Raise your [12:16] hand if you've thought about that. Is AI [12:19] going to replace investment analysts? Is [12:22] it going to replace analysts in banking? [12:24] Is it going to replace analysts in [12:26] venture capital? So, who thinks that AI [12:29] will replace analysts in banking? [12:32] I think there there's just like only so [12:33] much like um throughput a human has in [12:37] their brain and the ability to just [12:39] ingest all the information from all the [12:41] sources and create meaningful insights [12:43] is well suited for AI [12:46] >> and um it could just periodically [12:48] monitor and stay up to date and [12:50] >> it's just tireless. [12:52] Right? So that's a really good reason. [12:55] There's a limit to what humans can [12:59] ingest with data and how we can analyze [13:01] it especially in the world of [13:03] unstructured data uh where we have [13:05] social media data and we have chats and [13:08] we have uh text uh text data for [13:11] example. So AI can be better at us at [13:15] finding and analyzing and telling us [13:17] what data is important. And for an [13:20] analyst in investment banking firm for [13:23] example, they should be looking at tons [13:26] of data both structured and unstructured [13:29] to identify trends in an industry to [13:33] help develop an investment thesis and to [13:36] build conviction over that thesis. And [13:39] by the way, I would posit that's [13:41] something that AI can't do. AI can't [13:44] give us conviction. You can show me the [13:46] best data in the world for an incredible [13:48] deal, but if I have experience, for [13:52] example, or I've had a human interaction [13:55] related to that deal, I'm not going to [13:58] believe you. All right? Um, so that's [14:01] one thing that AI can do. And in some [14:04] ways that can replace what we ask [14:07] analysts to do for venture capital [14:09] analysts for example they look at [14:12] hundreds if not thousands of startups [14:14] and it's very very difficult to [14:16] ascertain what data is important how am [14:19] I going to do it for all of these [14:21] companies and if I don't know these [14:23] founders how am I going to decide who's [14:25] worthy and who's not all right so that's [14:28] data and in fact some of the studies [14:30] show us how AI can improve improve [14:33] things like investment analysis, looking [14:35] at data, productivity, for example, from [14:38] the Stanford MIT study. All right, what [14:42] about writing or powerpoints? Who of you [14:46] use AI for PowerPoint presentations? [14:50] All right, that's good. But that's a [14:52] small percentage. I would say like maybe [14:54] 5% in this room. If you haven't done it, [14:57] you might want to try it. Uh it's [15:00] amazing. I used AI to create this [15:02] presentation, but I the human being [15:05] actually decided what I wanted to talk [15:08] to all of you about and I put together [15:11] the notes and I built the presentation. [15:14] AI only helped me with design with ideas [15:17] and what can AI not replace? It cannot [15:21] replace my experience. So AI can make a [15:25] presentation for me, but AI can't say, [15:27] "You know what, Sarah, during the [15:28] banking crisis, here was your experience [15:31] dealing one-on-one with those banks. [15:34] Let's put it into the presentation." [15:36] Writing, for example. So, uh, investment [15:39] banking analysts are well known for [15:42] putting together deal decks, for [15:44] sometimes writing memos, preparing [15:46] things for investment committee. AI can [15:49] certainly do that, right? And we have [15:51] studies that show that. I don't have the [15:53] full oh I have a partial citation up [15:55] there from noi in 2023. So that was a [15:57] very interesting research paper and [15:59] again I can provide the citations to all [16:01] of you. Investment banking gains who's [16:04] familiar with the deote study in 2023. [16:08] Oh good. Do you want to tell us about [16:09] it? [16:11] Okay. Also I know I surprised sorry I [16:14] surprised you with a mic again. [16:17] >> No. [16:18] >> Oh he doesn't want to tell us but I saw [16:20] a hand here. No problem. But I saw a [16:22] hand here. There was another person who [16:25] is familiar with the Deoid study. [16:28] Anybody? Now they're like, "No, I I [16:30] actually didn't raise my hand." All [16:32] right. Very interesting Deoid study from [16:35] 2023. [16:37] Uh that said that we um can increase our [16:41] investment banking gains by using [16:43] artificial intelligence by 27 to 35%. [16:48] Uh that's interesting. 27 to 35%. Wow, [16:52] amazing. So even if we are using AI to [16:56] replace some of the wrote tasks that [16:59] investment analysts or associates have [17:02] been doing, we still see a gain overall [17:05] in the industry and that's what that [17:07] study is telling us. All right, so let's [17:10] talk about the human edge again. What [17:13] can AI not do? Well, AI cannot build [17:17] trust. So we talked about for example [17:20] conviction. We talked about your [17:22] personal experience. AI can't know your [17:26] personal experience unless you tell it. [17:28] Uh and AI can't apply your personal [17:31] experience uh precisely in the way that [17:34] a human would. And if you believe in [17:36] things like intuition, AI can't do that. [17:39] It's amazing because if you use the [17:41] tools, um, you might think that it can, [17:44] but only humans have human intuition by [17:47] definition. Who's used GPD5 that just [17:50] came out this week? All right. Do you [17:53] like it? Yeah. Do you think it's as Have [17:55] you used Deep Research? [17:59] >> Okay. Do you think it's as good as Deep [18:01] Research or better? You like this one? [18:04] Yeah. So, we like this one. We like uh [18:06] GPT5. I like I prefer it. I much prefer [18:10] it over deep research. It's easier to [18:12] use. It's much more accessible and of [18:14] course um it's priced much lower than [18:17] deep research. But instead of having to [18:21] um give more direction, you can do it [18:24] all sort of in one prompt and GPT5 [18:27] will provide you with a lot of [18:30] information in a shorter time based on [18:32] your question. Okay. So what about [18:36] trust? [18:38] Who here trusts AI? Does anybody here [18:40] trust AI to their entire investment [18:43] portfolio? Why do you trust AI? You you [18:47] trust it more than a human [18:49] >> than the average human. And going back [18:51] to intuition just a couple moments back, [18:53] I think it's important to realize that [18:55] most people's intuition is wrong at [18:56] least half the time. Um so yeah, [19:00] >> that's a great point. So you trust AI [19:03] more than many humans and [19:06] >> more than the average human? [19:07] >> More than the average human. So when you [19:09] say average, do you mean intelligence or [19:12] skills? [19:14] >> Can't answer a single question in the [19:15] test they're giving the AI to score it. [19:17] Right? I I've looked [19:19] >> but going back who do you trust? Right? [19:21] Maybe your close family, uh your best [19:23] friend, but beyond that, do you trust [19:25] me? Give me your money. I'll invest it [19:27] for you. [19:27] >> Right? [19:29] All right. So if we trust you, we'll [19:31] give you all our money after this [19:32] presentation. So those are all great [19:34] points, right? Some people do trust AI [19:37] more than humans. And um that's [19:41] understandable, right? AI is a model. [19:43] Humans have emotion. We make mistakes. [19:46] Um and in deal making, I suppose you [19:49] could go one way or the other. Um I [19:51] personally feel the human touch is [19:53] important. And I told all of you that I [19:55] have 25 years of experience as a [19:58] regulator, as an academic, as a business [20:00] leader, as a trader and portfolio [20:02] manager. So maybe you trust me more. [20:05] Right now, as compared to Claude, if you [20:08] wanted to ask Claude some questions [20:10] about the banking crisis, which one [20:12] would you trust more? And I'm just [20:15] posing that question for all of you to [20:16] think about. I know I have to wrap it [20:18] up. Okay. Why not leave deals to AI? [20:22] Well, we said that AI can process data [20:25] faster and sometimes better, that models [20:28] can achieve synergies, right? And that [20:31] AI can flag problems that humans miss. [20:34] We don't always trust humans. We might [20:36] trust AI in certain circumstances. Well, [20:40] the one thing that we haven't talked [20:41] about yet is relationship equity. So, I [20:44] spent 20 minutes up here with all of you [20:47] talking about some of these issues. We [20:50] haven't necessarily come to conclusions, [20:52] but I can tell you from my professional [20:54] experience, we know that investment [20:56] banks and private equity companies and [20:59] venture capital companies are rapidly [21:01] moving forward with deploying AI in the [21:04] investment space. And I can tell you [21:06] that based on surveys that I have been [21:09] involved with with the financial [21:11] services industry and the the signals [21:14] from some of those places who have hired [21:16] data analytics es experts at the top to [21:19] look at their portfolios. But I spent 20 [21:22] minutes with all of you today. And so I [21:25] have more relationship equity with you [21:27] now than I did before this presentation. [21:30] Right? So maybe based on what I said [21:32] today, you might be more inclined to [21:34] come and talk to me about what I think [21:37] about investment banking, dealmaking. [21:39] And I hope that's true. Thank you so [21:41] much for the time today.