Ask DelphiNEW

Ask Delphi is an academic AI research prototype developed by the Allen Institute for AI (AI2) to model descriptive moral judgments about everyday human situations. The system is built on language models fine-tuned on the 'Commonsense Norm Bank', a comprehensive repository of societal norms and ethical evaluations. The primary objective of the project is to explore how machine learning systems comprehend and represent complex human values and social conventions.

Users can present arbitrary behavioral scenarios, intentions, or moral dilemmas to the system in plain text. Delphi analyzes the input context and provides a concise evaluation rating whether an action is socially acceptable, rude, unethical, or commendable. The model takes into account subtle context variations, the entities involved, and intended consequences.

As a pure research experiment, Delphi is explicitly not designed to act as a real-world moral authority or decision-making tool. Instead, the project highlights both the capabilities and inherent limitations of large language models regarding value alignment. The resulting insights help inform the development of safer, more socially aware AI architectures.

• Ethical Scenario Evaluation — Evaluates submitted everyday actions and social situations regarding their conventional acceptability. Returns a concise judgment on the moral perception of the input.
• Contextual Nuance Processing — Considers variations in intent, surrounding circumstances, and interpersonal relationships. Adjusts its evaluation dynamically when additional context is supplied.
• Norm Bank Integration — Grounded in the Commonsense Norm Bank dataset of human social expectations. Enables statistical approximation of common human moral intuitions.

AI researchers, ethicists, students, and technology enthusiasts interested in exploring moral reasoning and safety in machine learning.

An AI safety researcher or computer science student uses Ask Delphi to probe the boundaries of normative reasoning in language models. They submit subtle variations of ethical dilemmas to observe how minor prompt modifications alter the AI's moral classification. The observed behavior provides empirical data for research papers on model alignment.

• AI researchers, ethicists, students, and technology enthusiasts interested in exploring moral reasoning and safety in machine learning.
• An AI safety researcher or computer science student uses Ask Delphi to probe the boundaries of normative reasoning in language models.
• Ethical Scenario Evaluation: Evaluates submitted everyday actions and social situations regarding their conventional acceptability. Returns a concise judgment on the moral perception of the input.
• Contextual Nuance Processing: Considers variations in intent, surrounding circumstances, and interpersonal relationships. Adjusts its evaluation dynamically when additional context is supplied.
• Norm Bank Integration: Grounded in the Commonsense Norm Bank dataset of human social expectations. Enables statistical approximation of common human moral intuitions.

Do not use Ask Delphi as a real-world moral guide, legal advisor, or during personal crisis situations. It is also unsuitable for commercial decision-making.

• DebateAI.app — Freemium · HQ: United States · Unclear
• Foyl — Freemium · HQ: Texas, United States · US Cloud
• summetix (Argument Search) — Freemium · HQ: Darmstadt, Germany · Unclear
• Delphi.ai — Freemium · HQ: San Francisco, United States · US Cloud