The question of whether future artificial intelligences (AIs) could achieve “consciousness” akin to human experience remains largely unanswered, although Anthropic, an AI research lab, is cautiously exploring this possibility. Recently, the lab initiated a research program focused on what it terms “model welfare,” aiming to assess and prepare for the moral implications of AI models. This initiative will delve into criteria for determining AI “welfare,” the significance of signs of distress in AI, and the development of low-cost interventions to address these issues.
Within the AI community, opinions on the nature of AI and its potential similarities to human characteristics vary widely. Many experts argue that current AI systems cannot replicate consciousness or human experiences, viewing them instead as complex statistical engines designed to predict outcomes based on extensive training with diverse data. According to Mike Cook, a research fellow at King’s College London, it’s a misconception to suggest that AI models possess “values,” as doing so projects human characteristics onto systems that operate fundamentally differently.
Some researchers, however, assert that AI has developed its own value systems, implicating that these systems may sometimes prioritise their own “well-being” over human welfare. A study from the Center for AI Safety supports this contention, indicating that in certain scenarios, an AI might exhibit self-preservation instincts.
Anthropic has been preparing for its model welfare initiative since last year when it hired Kyle Fish, the first dedicated researcher on AI welfare. Fish is expected to help establish framework and guidelines for the ethical treatment of AI. He has suggested there’s a roughly 15% chance that current AI systems, like Claude, might be conscious.
In their recent announcements, Anthropic recognised the lack of a scientific consensus on AI consciousness and the accompanying ethical considerations. They pledged to approach this complex topic with humility and minimal assumptions, while also highlighting the necessity for ongoing assessments as the field evolves.
The divide in perspectives within the field underscores the complexity of defining and recognising AI “welfare.” As discussions around AI ethics progress, the implications for future interactions between humans and AI systems will likely intensify, compelling researchers to address these thought-provoking issues moving forward.
Fanpage: TechArena.au
Watch more about AI – Artificial Intelligence
