UK AI Firm Unveils Foundation Model for Human Behaviour

London-based firm Artificial Societies has launched Rothko, a foundation model for human behaviour, using a proprietary architecture and training corpus to predict how individuals and groups make decisions.

The firm was founded two years ago this month by James He (CEO), Patrick Sharpe (CPO) and Tom Whittle (CTO). Its core product, AS, uses a database of more than 500,000 AI personas to 'simulate human societies', and predict both trends and audience reactions to specific content and messaging. The product can be used to optimize product and marketing decisions, and according to the company has demonstrated more than 80% accuracy in predicting social media performance, far better than the mean 62.5% achieved by other 'traditional' large language models, of which it lists Claude, Gemini and GPT-4o.

Rothko's launch follows more than three years of frontier AI research, the firm says. It works by taking a pretrained model and changing its architecture so that it produces decisions as outputs rather than natural language text. It is then trained on an 'enormous' volume of human behavioural data which includes anonymised social media information, transaction histories, mobility traces, unstructured expert interviews and traditional surveys.

The result is a large decision model that takes information about an individual and their context and predicts their behaviour. Artificial Societies says it improves significantly on the success rates of general-purpose LLMs in this field, which tend towards common answers, removing the variation found in real human behaviour. In addition to predicting purchase intent and public reactions to high-stakes corporate strategies, the model also has potential for increasing democracies' resilience against 'misinformation, polarisation, and other cognitive attacks.'

In an evaluation published by the firm on 1st October, seven sources of real survey data, including the General Social Survey (GSS), were used to compare the new tool's results against nine frontier LLMs, and found that Rothko had comfortably the lowest error on every measure.

Market researchers can use Rothko to test purchase intent and public reaction to products, campaigns and corporate strategies prior to launch, as well as to simulate reactions among hard-to-reach audiences.

However the firm says its own research suggest that the new model's most valuable application will be 'at the interface of human-agent collaboration, particularly in exploring novel techniques to make swarms of super-intelligent models safe.'

co-founder and CEO James He (pictured) comments: 'Rothko marks a new frontier in behavioural simulation, bringing together world-class social science and AI research to provide clarity and conviction for the world's most consequential decisions.'

Home page: www.societies.ai .

All articles 2006-23 written and edited by Mel Crowther and/or Nick Thomas, 2024- by Nick Thomas, unless otherwise stated.