Impromptu: Amplifying Our Humanity Through AI consciousness as well as GPT-4 can? Far more novel. Interact- ing with them can be uncanny, even unsettling.11 Since LLMs are so novel and seemingly agentic, it doesn’t sur- prise me that we’re already seeing New York Times op-eds calling to “protect society” from rogue AIs, or Substack dis- patches like a recent one from Gary Marcus, a cognitive psy- chologist and computer scientist who often critiques LLMs, who lamented the current “Wild West” environment where “anyone can post any chatbot they want” without prior permis- sion from Congress. Wanting to protect society from bad tech outcomes is not a new phenomenon, of course. In fact, it’s exactly this sentiment that led OpenAI’s founders to create their organization in 2015. So what’s the most effective and inclusive way to achieve good outcomes for society in the long term? In recent years, the predominant critique of AI is that it is some- thing that has largely been happening to individuals rather than for them—an under-the-radar force deployed by Big Tech without much public knowledge, much less consent, via tech- nologies like facial recognition and algorithmic decision-mak- ing on home loans, job applicant screening, social media rec- ommendations, and more. A founding goal of OpenAI was to develop technologies that put the power of AI directly into the hands of millions of people. In this way, AI might function as a decentralized, personally empowering force, rather than a top-down, totalizing one. Broadly distributed and easily accessible to individuals making 11 I imagine this is especially true when one of these tools starts be- having like Microsoft’s Sydney has on at least some occasions. (I haven’t had anything like that experience yet.) 214