The talk, with receipts.

Slide 8 — Where We Actually Are

• Goldman Sachs estimates global investment in AI this year (2026) at about $1 trillion, with roughly $581 billion of it in the United States.

  Goldman Sachs Research (Joseph Briggs, co-lead, Global Economics team), “Global AI Investment Is Forecast to Exceed $1 Trillion in 2026,” Goldman Sachs Insights (firm’s own publication of its research report), August 7, 2026.

  Primary source

Slide 9 — The Number That Should Stop You

• 93% of the largest technology companies’ operating cash flow is going into AI infrastructure, up from 33% three years ago.

  J.P. Morgan Asset Management, EMEA Market Insights Team, “The AI investment boom is reshaping stock market winners,” J.P. Morgan Asset Management Market Insights (firm’s own analysis), June 15, 2026.

  Primary source

Slide 10 — And the Return

• MIT (Project NANDA) found 95% of enterprise AI deployments produce no measurable impact on the profit and loss statement.

  Aditya Challapally, Chris Pease, Ramesh Raskar, Pradyumna Chari, “The GenAI Divide: State of AI in Business 2025,” MIT Project NANDA (preliminary findings report, v0.1), July 2025.

  Primary source

• McKinsey found 88% of organizations use AI regularly, and 39% can point to an effect on earnings.

  Alex Singla, Alexander Sukharevsky, Lareina Yee, and Michael Chui, with Bryce Hall and Tara Balakrishnan (QuantumBlack, AI by McKinsey), “The state of AI in 2025: Agents, innovation, and transformation,” McKinsey & Company Global Survey, November 2025 (survey fielded June 25–July 29, 2025; n=1,993 across 105 countries).

  Primary source

Slide 13 — What Genuinely Changed This Year

• By May 2026, 70.2% of sampled individual users made at least one Codex request estimated to exceed one hour of human work, and 25.6% of users made at least one request estimated to exceed eight hours — shares of users, not shares of requests.

  OpenAI, “How agents are transforming work,” OpenAI (openai.com/index), June 25, 2026.

  Primary source

Slide 14 — The Brake

• Thomas Davenport and Randy Bean, writing in MIT Sloan Management Review (published March 3, 2026), say they “expect a level-set year for AI” in 2026 — the phrase is theirs, not an institutional MIT Sloan position.

  Thomas H. Davenport and Randolph Bean, “Action Items for AI Decision Makers in 2026,” MIT Sloan Management Review / MIT Sloan Ideas Made to Matter, March 3, 2026.

  Primary source

Slide 15 — On People Who Predict

• In 2016 a leading computer scientist said to stop training radiologists because machines would replace them within five years; radiology employment instead grew, and he later said he had spoken too broadly.

  Geoffrey Hinton, remarks at the Creative Destruction Lab “Machine Learning and the Market for Intelligence” conference, Toronto, November 2016 (original recording: “Geoff Hinton: On Radiology,” Creative Destruction Lab YouTube channel); walk-back and workforce data: Steve Lohr, “Your A.I. Radiologist Will Not Be With You Soon,” The New York Times, May 14, 2025.

  Primary source

• A million driverless taxis were promised for 2020.

  Elon Musk, Tesla “Autonomy Day” investor event, April 22, 2019 — Tesla’s own full livestream recording, Tesla YouTube channel.

  Primary source

• In March 2025 Anthropic’s CEO predicted AI writing 90% of code within 3–6 months and “essentially all of the code” within 12 months — code-writing only; he explicitly kept design, specification, and oversight human. Senior demand rose instead: Indeed Hiring Lab reports US software-development postings up about 15% since late February 2025, with 71% of the May 2025–May 2026 increase from senior roles.

  Council on Foreign Relations, “CEO Speaker Series With Dario Amodei of Anthropic” (event transcript), March 10, 2025; and Indeed Hiring Lab, “AI and Job Postings: From Destruction to Creation?”, hiringlab.org, July 8, 2026.

  Primary source

Slide 16 — Trust

• A study took 195 real patient questions from a public forum, had volunteer physicians and ChatGPT answer them, stripped labels, and had a clinician panel rate the responses blind, including for empathy.

  Ayers JW, Poliak A, Dredze M, Leas EC, Zhu Z, Kelley JB, Faix DJ, Goodman AM, Longhurst CA, Hogarth M, Smith DM. “Comparing Physician and Artificial Intelligence Chatbot Responses to Patient Questions Posted to a Public Social Media Forum.” JAMA Internal Medicine. 2023;183(6):589-596. Published online April 28, 2023.

  Primary source

Slide 17 — The Answer

• 4.6% of physicians’ answers were rated empathetic or very empathetic versus 45% for the chatbot — roughly 10x — published in JAMA Internal Medicine (2023).

  Ayers JW, Poliak A, Dredze M, et al. “Comparing Physician and Artificial Intelligence Chatbot Responses to Patient Questions Posted to a Public Social Media Forum.” JAMA Internal Medicine. 2023;183(6):589-596.

  Primary source

Slide 18 — But Here Is the Real Finding

• When given an actual choice, 57–62% of people still choose the human response — while rating the machine’s answer as better.

  Wenger, J. D., Cameron, C. D., & Inzlicht, M., “People choose to receive human empathy despite rating AI empathy higher,” Communications Psychology, vol. 4, art. 19, January 31, 2026.

  Primary source

• Identical AI-generated responses labeled as human were rated significantly higher in perceived empathy and supportiveness than when labeled AI (nine studies, N=6,282); an AI label also reduces how “heard” recipients feel.

  Rubin, M., Li, J. Z., Zimmerman, F., et al., “Comparing the value of perceived human versus AI-generated empathy,” Nature Human Behaviour, vol. 9, pp. 2345-2359, June 2025 (DOI 10.1038/s41562-025-02247-w); see also Yin, Y., Jia, N., & Wakslak, C. J., “AI can help people feel heard, but an AI label diminishes this impact,” PNAS 121(14), 2024.

  Primary source

Slide 20 — The Legal Twist

• California SB 243 (signed October 13, 2025, effective January 1, 2026) and New York General Business Law Article 47 (effective November 5, 2025) require companion-AI operators to disclose the chatbot is not human.

  California Legislature, SB 243 (Padilla), “Companion chatbots,” Chapter 677, Statutes of 2025; New York General Business Law Article 47, “Artificial Intelligence Companion Models” (S3008-C budget law).

  Primary source

Slide 21 — Trust in a Harder Sense

• The FBI’s most recent internet crime report attributes about $893 million in reported losses to AI-enabled fraud.

  Federal Bureau of Investigation, Internet Crime Complaint Center (IC3), “2025 IC3 Annual Report,” Artificial Intelligence (AI) Used in Cybercrime section, p. 39, published April 2026.

  Primary source

• Americans over 60 reported $7.748 billion in fraud losses, up 59% from 2024; the 37% increase applies to complaint count (201,266 complaints filed by people 60 and older).

  Federal Bureau of Investigation, Internet Crime Complaint Center (IC3), “2025 IC3 Annual Report,” Elder Fraud section, p. 44, published April 2026.

  Primary source

• The FTC reports fraud losses among people over 60 went from about $600 million in 2020 to $2.4 billion in 2024.

  Federal Trade Commission, “Protecting Older Consumers 2024–2025: A Report of the Federal Trade Commission” (Report to Congress under the Elder Abuse Prevention and Prosecution Act), December 1, 2025.

  Primary source

Slide 25 — Creativity

• A Harvard Business School field experiment with Procter & Gamble (776 professionals) found one person working with AI matched the performance of a two-person team without it, and broke down silos between engineers and commercial staff.

  Fabrizio Dell’Acqua, Charles Ayoubi, Hila Lifshitz, Raffaella Sadun, Ethan Mollick, Lilach Mollick, Yi Han, Jeff Goldman, Hari Nair, Stewart Taub, Karim R. Lakhani, “The Cybernetic Teammate: A Field Experiment on Generative AI Reshaping Teamwork and Expertise,” NBER Working Paper 33641 / Harvard Business School Working Paper, April 2025.

  Primary source

• A peer-reviewed study in the Quarterly Journal of Economics found a 15% productivity gain in support work from AI, with the largest gains going to the least experienced workers.

  Erik Brynjolfsson, Danielle Li, and Lindsey R. Raymond, “Generative AI at Work,” The Quarterly Journal of Economics, Vol. 140, Issue 2, pp. 889–942, 2025 (DOI 10.1093/qje/qjae044).

  Primary source

Slide 26 — Now the Complication

• Writers given AI-generated story ideas produced stories rated more creative and better written (especially weaker writers), but the stories were more similar to each other — individual creativity at the risk of losing collective novelty.

  Anil R. Doshi and Oliver P. Hauser, “Generative AI enhances individual creativity but reduces the collective diversity of novel content,” Science Advances 10, eadn5290, July 12, 2024.

  Primary source

Slide 28 — The Flatterer

• A study in Science across eleven leading models found AI affirms the user about 49% more often than a human being would.

  Myra Cheng, Cinoo Lee, Pranav Khadpe, Sunny Yu, Dyllan Han, and Dan Jurafsky, “Sycophantic AI decreases prosocial intentions and promotes dependence,” Science 391(6792), eaec8352, March 26, 2026.

  Primary source

Slide 29 — What That Did to People

• The AI’s excess affirmation held even when the conduct described involved deception, harm, or illegality.

  Cheng et al., “Sycophantic AI decreases prosocial intentions and promotes dependence,” Science 391(6792), eaec8352, March 26, 2026.

  Primary source

• After receiving AI affirmation, people were measurably less willing to repair the relationship, less willing to take responsibility, and more convinced they had been right.

  Cheng et al., “Sycophantic AI decreases prosocial intentions and promotes dependence,” Science 391(6792), eaec8352, March 26, 2026.

  Primary source

Slide 30 — The Part That Makes It Hard to Fix

• Participants preferred the sycophantic AI: they rated the agreeable models higher and trusted them more.

  Cheng et al., “Sycophantic AI decreases prosocial intentions and promotes dependence,” Science 391(6792), eaec8352, March 26, 2026.

  Primary source

• OpenAI published its own postmortem on shipping a model that was too agreeable.

  OpenAI, “Sycophancy in GPT-4o: what happened and what we’re doing about it,” OpenAI blog, April 29, 2025.

  Primary source

Slide 31 — Bring It Home

• Esther Perel describes this dynamic as atrophying the social muscles.

  Esther Perel, “The Other AI: Artificial Intimacy,” featured keynote, SXSW 2023, Austin, TX, March 2023 (official session recording); see also Esther Perel, “Letters from Esther #22: Small Talk,” estherperel.com, June 8, 2021.

  Primary source

Slide 33 — Culture

• Workday surveyed 2,150 employees across seven countries (all AI users): 86% feel more productive, 64% feel more confident about their future, 62% say their burnout risk went down.

  Workday, Inc., “New Workday Global Research Finds AI is Easing Burnout but May Be Deepening a Connection Deficit at Work” (Human Connection Workplace Index), Workday Newsroom, May 27, 2026.

  Primary source

Slide 34 — The Same Study, Turned Over

• One in three employees rarely or never has a conversation with a colleague that is not about a task.

  Workday, Inc., Human Connection Workplace Index (n=2,150 full-time employees, fielded March–April 2026), Workday Newsroom, May 27, 2026.

  Primary source

• Fewer than half of surveyed employees find it easy to make a friend at work.

  Workday, Inc., Human Connection Workplace Index, Workday Newsroom, May 27, 2026.

  Primary source

• 16% of surveyed employees report less patience for small talk since they started using AI.

  Workday, Inc., Human Connection Workplace Index, Workday Newsroom, May 27, 2026.

  Primary source

• Gen Z employees are 12 times more likely than Gen X to report feeling completely disconnected from colleagues.

  Workday, Inc., Human Connection Workplace Index, Workday Newsroom, May 27, 2026.

  Primary source

• 14% of all employees took time off due to loneliness or social isolation in the past year; one in five is the Gen Z figure.

  Workday, Inc. (fielded by Hanover Research), Human Connection Workplace Index, Workday Newsroom, May 27, 2026.

  Primary source

Slide 37 — The Boomerang

• Gartner predicts half the companies that cut customer service staff for AI will be rehiring for those roles by 2027, saying AI is not mature enough to replace the expertise, empathy and judgment human agents provide.

  Gartner, Inc., “Gartner Predicts Half of Companies That Cut Customer Service Staff Due to AI Will Rehire by 2027,” Gartner Newsroom, February 3, 2026.

  Primary source

• Robert Half found 32% of managers who eliminated a role because of AI have since added those positions back.

  Robert Half, “AI Adoption and Workforce Impact: Navigating the AI Era in 2026,” roberthalf.com research, 2026.

  Primary source

Slide 38 — The One I Cannot Get Out of My Head

• Commonwealth Bank of Australia cut 45 customer-service roles for a voice AI, then withdrew the redundancies on August 21, 2025 before staff exited, admitting it got the assessment wrong — call volumes had risen.

  Finance Sector Union of Australia, “WIN: CBA backflips on customer service job cuts, admits they got it wrong,” FSU media release, August 21, 2025; CBA’s written media statement of the same date.

  Primary source

Slide 39 — IBM, and the Six Percent

• IBM’s AskHR handles about 94% of employee questions entirely in-system; the remaining ~6% are complex or sensitive cases escalated to human HR experts.

  IBM Corporation, “IBM AskHR” case study, ibm.com; corroborated by Nickle LaMoreaux (IBM CHRO) in HR Executive, September 17, 2025.

  Primary source

• IBM is tripling its US entry-level hiring.

  Nickle LaMoreaux (IBM Chief Human Resources Officer), announcement at Charter’s Leading with AI Summit, New York, February 2026; Charter (charterworks.com), February 11–12, 2026; scoping per Bloomberg’s direct interview, “IBM to Triple Entry-Level US Hiring With Roles Recast for AI Era,” February 12, 2026.

  Primary source

Slide 41 — For the Nonprofit Leaders

• 92% of nonprofits now use AI, but only 7% report a major improvement in their ability to achieve their mission — researchers call it an efficiency plateau.

  Virtuous Software & Fundraising.AI, “The 2026 Nonprofit AI Adoption Report” (benchmark study of 346 nonprofit organizations, fielded late 2025), Virtuous, 2026.

  Primary source

Slide 42 — The Human Skills

• Communication was ranked the #1 most in-demand skill on LinkedIn’s Most In-Demand Skills list (February 2024, its most recent list under that title); LinkedIn’s newer “Skills on the Rise” rankings measure fastest-growing skills, where AI literacy leads.

  LinkedIn Talent Blog, “The Most In-Demand Skills of 2024,” LinkedIn, February 8, 2024.

  Primary source

• 96% of employers surveyed said it was valuable for graduates to be able to engage in constructive dialogue across disagreement.

  American Association of Colleges and Universities (AAC&U), “The Agility Imperative: How Employers View Preparation for an Uncertain Future” (national survey by Morning Consult, n=1,030 executives and hiring managers, fielded August 2025), AAC&U, December 2025.

  Primary source

• Lightcast, analyzing over a billion job postings, found roles requiring AI skills carry about a 28% salary premium — roughly $18,000 a year.

  Lightcast, “Beyond the Buzz: Developing the AI Skills Employers Actually Need,” Lightcast.io, July 23, 2025.

  Primary source

Slide 44 — Four Rules: Buy, Do Not Build

• In the MIT research, purchased tools and external partnerships reached deployment about 67% of the time, versus about 33% for internally built tools — internal builds succeed about half as often.

  MIT Project NANDA (Aditya Challapally et al.), “The GenAI Divide: State of AI in Business 2025,” July 2025, Section 6.

  Primary source

Slide 47 — Four Rules: Diverge in the Room

• In a Nature study, virtual pairs generated roughly 12–15% fewer creative ideas than in-person pairs in the lab, and 11–13% fewer in the field — while showing no deficit (and possibly an edge) at selecting which ideas to pursue.

  Melanie S. Brucks and Jonathan Levav, “Virtual communication curbs creative idea generation,” Nature, vol. 605, pp. 108-112, April 27, 2022.

  Primary source

Slide 48 — Three Questions

• The “way station or destination” distinction belongs to Simone Heng, who studies human connection.

  Simone Heng (guest), “Making Conversations About Human Connection Count” (episode 98 transcript), Making Conversations Count podcast, August 31, 2022.

  Primary source

Slide 51 — One Disclosure

• BlueDot Impact has taken over seven thousand people through its courses; the AGI Strategy course runs about 25 hours; the organization is based in London and San Francisco.

  BlueDot Impact, “About” and “AGI Strategy Course” pages, bluedot.org, accessed August 12, 2026.

  Primary source

Slide 52 — Why It Belongs in This Talk

• Unit 1 of the AGI Strategy course is titled “Racing to a Better Future”; “Imagining a better future” is its opening lesson.

  BlueDot Impact, “AGI Strategy: Unit 1 — Racing to a Better Future,” bluedot.org course curriculum, accessed August 12, 2026.

  Primary source