Design Intelligence 04: Living with synths
Research notes on tech, ecology, design futures, and design. This month: the next computing platform
The idea: AI agents
For 18 months, the biggest and most frequently updated file in my research folder has been the one titled Generative AI. One of the most exciting ideas in that file has been the AI agent - an autonomous bot that can carry out tasks for its owner and make decisions on their behalf, communicating with other AI agents to do so.
An AI agent is a digital personal assistant; it’s the idea of AI that comes immediately to mind for many of us - a variant of the virtual companion in the 2013 film Her. And because of the hurtling speed of development within AI at the moment, it’s a concept that is fast becoming real.
The key capabilities that distinguish an AI agent from another AI tool are that it can take decisions proactively, complete a self-directed chain of tasks, and communicate with other AI programs to do so. Researchers at Nvidia have gotten ChatGPT to play Minecraft by itself, which requires it moving through long sequences of decisions, tasks and actions, while AI startup Luminance has used LLMs to negotiate a contract end-to-end. Both happened without any human intervention.
Now, this ability is moving from research and testing to commercial availability. OpenAI’s new GPT Builder enables paying customers to generate custom GPTs from a prompt; the custom GPT can then be connected to third party platforms to act.
AI agents can be deeply personalised. Dot, an “intelligence guide” by New Computer, was trained by a playwright and acts as a conversational assistant to individual people, remembering each exchange with them and offering up information, ideas and reminders as “gifts”.
Beside the technological advancements this will bring - with Bill Gates saying AI agents will replace most software and become “the next platform” - this imminent ‘bot pop’ will change digital spaces to the extent that we will live with synths and interact closely with them, sometimes unwittingly. Essentially, the internet will fill with NPCs: non-playable characters, or in this context, non-people characters.
We already live with synthetic 'people' to some extent - many of the announcements at train stations are made by robotised voices, we speak to automated phone messages, and we type questions to chatbots - but this feels like a new line to cross. China’s streaming platforms are currently being populated by AI influencers; synthetic humans can be used to fill focus groups; Instagram is developing AI friends with customisable personalities; social media is increasingly synthetic.
Personal LLMs go still further: they can be trained on one person’s writing or text messages, ingest that person’s personal history and writing style, and then mimic it. In short, synthetic media can be trained to be you.
This might or might not happen: every future is a choice. So here are some open questions: how does the internet look when it's being navigated by AI agents rather than by people? How do human relationships change if they're entirely mediated through machines, through two AI agents acting out the mechanics of a friendship? Will this move us even further from taking responsibility for the consequences of our consumption? And if we outsource this many decisions - from planning holidays to sending flowers and a card to a loved one - why are we doing this, and what is left afterwards?
📄 Research notes
The Everyday Automation Observatory is collecting examples of automated processes, like delivery robots or self-service machines, as they become more common
Films are directed differently for streaming (because executives live by drop-off data) and TV shows are increasingly being written to be on in the background while you look at your phone
Apple is considering healthcare use cases for the Vision Pro, including using its eye tracking to assess mental health, and its cameras and microphones to gauge emotional responses. The device could also offer treatment or management techniques such as light and sound therapy
Nieuwe Instituut is rethinking the museum shop at Dutch Design Week 2023 with the New Store, which includes soap made from urine and lighting provided by visitors' smartphones
Let’s remake everything in paper: Pulpatronics’ RFID tags are made from paper and can be recycled as household waste; the Good Cup is recyclable and biodegradable and replaces plastic-coated cups and plastic cup lids
“A big part of my job now is removing things. How can we get rid of a layer while still giving the product a surface and texture that we want?” CMF designer Jenny Lechner on Urbanears' impressive progress towards being a circular brand
“Say goodbye to forever chemicals”: Patagonia, after 15 years of work, will be able to phase out use of forever chemicals in textile coatings by 2025. Like the Lego news from last month, showing your work in public like this is a win for sustainability strategy
An interesting use case for virtual reality headsets: HTC is sending them up to space to help keep astronauts from getting lonely
Skip aims to broaden perceptions around exoskeletons and create a new product category called “powered movement”
Designing Peace, an exhibition by Cooper-Hewitt at Museum of Craft and Design, San Francisco, is looking at “ways to create and sustain more durable peaceful interactions”
💻 My work
For Dezeen: “The tech hype-cycle is spinning ever-faster”
On Medium: Futures Thinking: 5 Brilliant Books on Tech, Culture & Society
🧼 Interesting products
Leica’s disinformation-proof camera stamps photos with Content Credentials metadata
Carbon-capture soap by CleanO2
Trustandsafety.fun, a game about content moderation
Humane’s Ai Pin is the biggest tech debut since the iPhone
Clever and cutting-edge health tech: a constipation pill that vibrates on arrival at destination
🎧 Listening
Pearson Lloyd on the Material Matters podcast, discussing how to design industrial products to have a lower environmental impact
If Books Could Kill: The “Organized Retail Crime” Panic for a class in data sourcing and finding the structural issues that lie beneath the symptom (in this case, shoplifting)