Doors to AI a project

Non-fiction → Fiction

Workshops

A more direct pathway between research and imagination. Below are accessible summaries of real presentations from cutting-edge technology and policy conferences, with responses from the fiction, entertainment and arts communities.

  1. Workshop 01

    Curious Robot Children

    Toddlers use all their senses to experience things that interest them, creating self-directed, well-organized data. Allowing AI to have bodies, curiosity and intention may make training new AI much easier and quicker.

    Original talk
    “Coherence statistics, self-generated experience and why young humans are much smarter than current AI” — Linda Smith, Indiana University [NeurIPS 2023]
    Accessible summary
    under construction
    Creative responses
    open call — come write something

    Image: “Frustrated Total Reflection: Its Application to Proximity Problems in Metrology”

  2. Workshop 02

    Impossible Social Norms

    Ambiguity about societal norms is measurable, structured, and complex. Asking an AI to "not harm humans" is impossible given conflicting definitions of harm.

    Original talk
    “The Many Faces of Responsible AI” — Lora Aroyo, Google [NeurIPS 2023]
    Accessible summary
    slides
    Creative responses
    open call — come write something

    Image: “The Atomic Clock, an Atomic Standard of Frequency and Time”

  3. Workshop 03

    How to Pay Attention

    AI could learn for itself what it should pay attention to, and be more efficient as a result. But self-learned attention might allow AI to become obsessed (overfit).

    Original talk
    “Sketching: core tools, learning-augmentation, and adaptive robustness” — Jelani Nelson, UC Berkeley / Google [NeurIPS 2023]
    Accessible summary
    slides
    Creative responses
    open call — come write something

    Image: “Macromolecular crystallography and structural biology databases at NIST”

  4. Workshop 04

    Democratizing Access to AI

    Making a smart AI requires data and computation power at an incredible, unsustainable scale. With improvements to architecture and training, it might be possible to train AI on everyday consumer technology.

    Original talk
    “NextGenAI: The Delusion of Scaling and the Future of Generative AI” — Björn Ommer, University of Munich [NeurIPS 2023]
    Accessible summary
    slides
    Creative responses
    open call — come write something

    Image: “Expansive characteristics of hydrated limes and the development of an autoclave test for soundness”