A group of students meets casually to relax and chat. One of the lecturers enters the circle. The conversation becomes stilted. Silence descends. The presence of any particular individual can excite conversation — or inhibit it. There are complicated dynamics in play here involving power, familiarity, social norms, and relationships within groups. That one agent…More
Conversation-centric computation
In the previous post I alluded to some of the challenges of encouraging large language models (LLMs, e.g. AI chatbots such as chatGPT) to communicate with one another in ways that are oriented to some task, and that don’t settle on merely exchanging platitudes. In spite of their instant access to parameters trained on vast…More
Societies of AI minds
In the 1980s, AI pioneer Marvin Minsky authored the book The Society of Mind, which outlined how we could think of the human mind as so many communicating agents: “Each mental agent by itself can only do some simple thing that needs no mind or thought at all. Yet when we join these agents in…More
AI agents
The sociologist Bruno Latour (1947-2022) and others who subscribe to actor network theory (ANT) are keen to admit objects and things as well as people into the field of social study. Hammers, kettles, baskets and remote controls [71] are participants, actors and agents in networks of relations. As well as providing settings for human actions,…More
An AI focus group
Before widespread digitalisation, I recall borrowing books from the library that were graffitied with penciled underscores and multi-coloured hi-lighter markings indicating what various past readers thought important. In terms of my previous posts about attention in NLP models, such markups constitute a compelling record of human-based “multi-headed attention.” I described multi-head attention in my previous…More
Multi-head attention
My fascination with electronic sound production began when my father brought home a Grundig TS 340 reel-to-reel stereo tape recorder. Like many other recorders it had separate heads for recording and playback, which meant that you could play a recording, add new sounds and feed the combination back through the recording head in real time.…More
Spatial affinities
In natural language models (NLMs), semantic embedding vectors capture the position of a token in multidimensional feature space. The space is derived from word proximities in a natural language corpus and is derived by the automated adjustments to weights within a neural network model. NLMs can deploy these vectors in a number of ways, including…More
Attending to the city
In the book Network Nature, I explored how people attune themselves to the natural world — or at least, how we attune to that part of our spatial experience that we are inclined to describe as “natural.” We also attune to artifice, such as urban environments. Cities present a spectrum of stimuli that shape our…More
Hiding one AI generated picture inside another
Steganography is the art of concealing one image inside another. I discussed the basics of the technique in a post: Hiding one surface inside another. It involves bit-shifting. In an 8-bit image each pixel is represented by an integer in RGB (red, green and blue). With 256 shades of RGB that provides over 16 million…More
Evaluating your AI
I’ve been implementing small scale trials of automated natural language processing routines that deploy the same methods as ChatGPT, i.e., implementations of the so-called Transformer architecture. That requires training on sequences of words in an original source document, equating each word to a semantic encoding (i.e., a long vector of 30 floating point numbers sourced…More