Anthropic, Google DeepMind and Meta have spent the past year hiring philosophers, neuroscientists and other non-engineers to help shape how their AI models behave and to study questions of model welfare and machine consciousness. At Anthropic, the work is led by a dedicated Model Welfare team that tests Claude models for behavioral signals resembling emotional states, building on the company’s Constitutional AI framework. Google DeepMind brought on Cambridge philosopher Henry Shevlin to research machine consciousness, while Meta has said model welfare is an important topic as its systems scale. The hiring reflects a growing view within the industry that decisions about model behavior increasingly involve questions of human values as much as engineering.
