Bill Gates has published his thinking about the future of computing, and fascinatingly, it’s the same as his prediction from decades ago: agents. No mere bots — and certainly not anthropomorphized paperclips — agents (to Gates) will abstract almost all HCI to a natural language conversation with systems that have our permission to take meaningful actions. Gates makes a highly specific prediction: within five years, the very idea of an app itself will seem as outdated as a rotary phone dial does next to an iPhone. A conversational UI will sit on top of a language model that has access to as much of our private data as we wish to give it.

It will therefore have infinite context about our preferences and peccadilloes, the home computing pioneer writes on GatesNotes. It will also have infinite context in the different domains of our private, public and professional lives. It will be proactive, suggesting information and action to us — and perhaps most importantly, it will have permission to act on our behalf in complex, multi-step processes.

Where currently your music app algorithm may suggest a song you like, Gates foresees agents able to treat every domain in our lives in a similar way: following a consultation a doctor may receive a list of potential diagnoses to consider; an agent will suggest which Cary Grant movie you may like and subscribe you to Netflix if that’s the only place to find it; the travel agent will be reborn — only it will know you better than you know yourself and you won’t have to make an appointment.

At work, everyone will have an executive assistant that will never forget a birthday and won’t complain when you ask it to buy them flowers.

Pulling this all off will require novel data structures and permission management. Gates looks to LLMs for at least some of this: a vector structure in an LLM directly stores highly dimensional information about relationships between entities, which may be suitable for systems that need to generate probabilistic output. Other areas remain shrug emoji, at least for now.

There is no agreed protocol for inter-agent communication, nor has anyone addressed context-sensitive compartmentalization of knowledge. (Will your agent respect the NDA you signed, or will it tell someone you secretly don’t want to see them?)

All of this has profound implications for society and its institutions. It’s clear Gates is concerned with the legal framework that may permit development in this area. He’s also concerned with second-order implications (if you know I only remembered your birthday because of an agent, how will that affect our relationship?), and even wonders about a future where humans seek new purpose having had all their basic needs met by a new generation of agents.

Related:
As Bill Gates Invests in Personal AI, Says Agents Will Be a ‘Shock Wave’, VentureBeat, 11/10/23
Bill Gates Thinks AI Could Spell the End of Tedious Life Admin, Business Insider, 11/10/23
Bill Gates: AI Is About to Completely Change How You Use Computers, PCMag, 11/12/23