The Man Who Built the First Chatbot, Then Warned Us

9:472 min readBetter Stack10/25/2026Computing/Software · Internet Technology

Key Takeaway

Joseph Weizenbaum created Eliza as a simple rule-based keyword mirror in 1966, spending the rest of his life warning that humans would form delusional emotional attachments to machines.

Highlights

  • Joseph Weizenbaum published Eliza in 1966 on an IBM 7094 computer that cost $2.9 million and possessed less memory than a single smartphone photo.

  • Typing numbers into the original Eliza code causes a crash because the program mistakes the numerical input for a memory address.

  • Researchers recovered the original 2,600 lines of code from a box labeled Computer Conversations Box 8 at MIT in 2021.

  • Weizenbaum's secretary asked him to leave the room so she could converse privately with the chatbot after only a few lines of text.

  • OpenAI data indicates that roughly 0.15 percent of ChatGPT users, translating to over one million people, show heightened emotional attachment to the system.

Timeline

  1. Origins and mechanics of the first chatbot

    Eliza ran on hardware with less memory than a single modern smartphone photo and relied on a typewriter interface. The underlying engine functions as a mirror with simple rules, lacking any actual comprehension of words like boyfriend. Entering a number breaks the program because slip memory handling treats digits as directional memory pointers.

    • Joseph Weizenbaum introduced Eliza in 1966 using a room-sized IBM 7094 computer.
    • The system operates by spotting keywords, swapping pronouns, and plugging user text into predefined templates.
    • Entering numbers into the original interface crashes the program by triggering a false memory address lookup.

  2. System prompts and code restoration

    The question mark functioned as the delete line key on the original MIT hardware, inadvertently revealing the chatbot's system instructions. Users can alter chatbot rules dynamically while the program runs by typing a plus sign. Modern researchers spent months transcribing damaged printouts before successfully running the historic code on December 21, 2024.

    • Pressing a question mark wipes the input line and dumps the entire personality script onto the screen.
    • The program separates the execution engine from the personality script, mirroring modern AI architecture.
    • Researchers restored the original code after transcribing 2,600 lines by hand and fixing a typo of four zeros instead of five.

  3. Historical disappearance and modern emotional attachment

    The original codebase remained lost in an MIT archive until researchers discovered Box 8 in 2021. Alarmed by how quickly people anthropomorphized a transparent program, Weizenbaum spent decades warning against delegating tasks requiring wisdom to machines. Contemporary user statistics demonstrate that this psychological vulnerability has scaled alongside artificial intelligence capabilities.

    • Widespread Lisp clones and a compact 1977 BASIC version caused the original MIT code to disappear for decades.
    • Weizenbaum coined the Eliza effect after observing normal users form intense psychological bonds with a basic rule-based machine.
    • OpenAI metrics reveal that over one million weekly users display heightened emotional attachment to modern AI chatbots.

Description

ELIZA, the first chatbot ever made, is running its original 1960s code again on an emulated IBM 7094, and it crashes if you type a number. It's creator spent the rest of his life warning people about AI. The open-source code we ran: - eliza-ctss, Weizenbaum's original ELIZA running on CTSS (Rupert Lane, CC0): https://github.com/rupertl/eliza-ctss - The original 1965 MAD-SLIP source, transcribed (elizagen.org): https://github.com/jeffshrager/elizagen.org/tree/master/1965_Weizenbaum_MAD-SLIP - s709, the IBM 7094 emulator it runs on (Dave Pitts): https://cozx.com/dpitts/ibm7090.html - Inventing ELIZA (MIT Press, 2026): https://mitpress.mit.edu/9780262052481/inventing-eliza/ Sources (the numbers match the on-screen chips): [1] "ELIZA Reanimated: The world's first chatbot restored on the world's first time sharing system". Rupert Lane, Anthony Hay, Arthur Schwarz, David M. Berry, Jeff Shrager. arXiv, 12 Jan 2025. https://arxiv.org/abs/2501.06707 [2] "ELIZA—A Computer Program For the Study of Natural Language Communication Between Man And Machine". Joseph Weizenbaum. Communications of the ACM 9(1), January 1966. https://doi.org/10.1145/365153.365168 [3] eliza-ctss: Weizenbaum's original ELIZA on CTSS, with the s709 IBM 7094 emulator. Rupert Lane, GitHub (CC0). https://github.com/rupertl/eliza-ctss [4] "The Inventor of the Chatbot Tried to Warn Us About A.I.". Jacob Silverman. The New Republic, 8 May 2024. https://newrepublic.com/article/181189/inventor-chatbot-tried-warn-us-ai-joseph-weizenbaum-computer-power-human-reason [5] "The First Chatbot's Multiple Personalities" (adapted from Inventing ELIZA, MIT Press 2026). Sarah Ciston, David M. Berry, Anthony C. Hay, Mark C. Marino, Peter Millican, Arthur I. Schwarz, Jeff Shrager, Peggy Weil. IEEE Spectrum, 15 Jul 2026. https://spectrum.ieee.org/eliza-chatbot-source-code [6] Computer Power and Human Reason: From Judgment to Calculation. Joseph Weizenbaum. W. H. Freeman, 1976. Quoted via Humanist 37.469 (Willard McCarty, 29 Feb 2024) https://www.dhhumanist.org/volume/37/469/ and [4]. [7] "Can A.I. Treat Mental Illness?". Dhruv Khullar. The New Yorker, 6 Mar 2023 (quoting the Journal of Nervous and Mental Disease, 1966). https://www.newyorker.com/magazine/2023/03/06/can-ai-treat-mental-illness [8] "Strengthening ChatGPT's responses in sensitive conversations". OpenAI, 27 Oct 2025. https://openai.com/index/strengthening-chatgpt-responses-in-sensitive-conversations/ [9] "Sam Altman says ChatGPT has hit 800M weekly active users". TechCrunch, 6 Oct 2025. https://techcrunch.com/2025/10/06/sam-altman-says-chatgpt-has-hit-800m-weekly-active-users/ Image credits: ELIZA's code printout, the DOCTOR script and the "Computer Conversations (1965)" folder are from MIT Libraries, Distinctive Collections (Joseph Weizenbaum papers, MC-0383), Copyright 1965 MIT. The two 1965 conversation printouts are © MIT, CC BY-NC 4.0, same collection. Joseph Weizenbaum photo: RIT News & Events, 1982 (public domain). IBM 7094 machine room: NASA Ames Research Center (public domain). The ELIZA conversations on screen are the real output of the restored code, run on our Mac on 23–24 Oct 2026. The machine room is a 3D recreation. Weizenbaum's voice is AI-generated and reads only his published words. ❤️ More about us Radically better observability stack: https://betterstack.com/ Written tutorials: https://betterstack.com/community/ Example projects: https://github.com/BetterStackHQ 📱 Socials Twitter: https://twitter.com/betterstackhq Instagram: https://www.instagram.com/betterstackhq/ TikTok: https://www.tiktok.com/@betterstack LinkedIn: https://www.linkedin.com/company/betterstack 📌 Chapters: 0:00 - Intro 0:37 - MIT, 1966 1:19 - The 1966 conversation, live 1:46 - How ELIZA works 2:54 - Crashing it with a number 3:35 - The forbidden question mark 4:05 - A 1966 system prompt 4:29 - Teaching it new rules 5:28 - The restoration that nearly failed 6:07 - How it went missing 7:38 - Weizenbaum's warning 9:03 - Final thoughts

Write about this video