Walter Sinnott Armstrong on AI and Morality
Philosophy Bites
Deep Dive
How does Walter Sinnott Armstrong define artificial intelligence?
Artificial intelligence is broadly defined as occurring whenever a machine learns something, as learning involves intelligence. It often involves the machine being given a goal and learning new and better means to achieve that goal.
What is the challenge of programming AI with human morality?
The challenge lies in choosing which moral principles to program into the AI, as different ethical systems like utilitarianism and Kantian ethics conflict. There is no consensus on which moral system should dictate the AI's decisions.
What is the hybrid approach to introducing ethics into AI?
The hybrid approach combines top-down principles with bottom-up data collection. It involves asking people which moral features matter in a situation, refining those features, and building conflicts to train the AI to predict human moral judgments.
How does AI handle moral dilemmas like kidney allocation?
AI collects features that matter to people, such as age, dependents, or criminal records, and predicts which patient should receive a kidney based on these factors. It can also confirm or challenge a doctor's decision, aiding in the decision-making process.
What are the limitations of using AI for moral decision-making?
AI can inherit human biases from data, and its decision-making process is often a 'black box,' making it difficult to understand the reasoning behind its conclusions. Additionally, local values and expertise must be considered to ensure fairness.
How far are we from using AI for ethical decisions in hospitals?
While AI is already used in some kidney transplant centers for medical efficiency, integrating moral considerations is still in development. Walter Sinnott Armstrong estimates it could take about 10 years for such systems to be refined and widely adopted.
What are potential applications of ethics in AI beyond healthcare?
AI can be applied to dementia care, hiring decisions to ensure fairness regarding gender and race, and even military operations. These applications aim to introduce moral considerations into various decision-making processes.
- Defining AI broadly as machines that learn and achieve goals.
- Challenges of top-down (pre-programmed ethics) and bottom-up (learning from internet data) approaches.
- The issue of human biases in AI algorithms.
- The "black box" problem of understanding AI decision-making processes.
Shownotes Transcript