iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Giving an AI “morals” is not simply a matter of programming one list of right and wrong answers. It means deciding what the system should align with—and whose judgments should shape those rules. A 2020 scholarly framework helps explain that challenge, but it does not establish what Anthropic has done specifically.
What does it mean to give an AI moral guidance?
AI alignment can mean aligning a system with different things: its instructions, a user’s intentions, people’s expressed preferences, their considered or ideal preferences, their interests, or broader values. These are not interchangeable goals. A system that follows instructions literally, for example, may behave differently from one designed to infer what a person ultimately wants.
Choosing among these targets is not only a technical task. It requires a judgment about what an AI ought to do and whose views should count. “Human values” is not a single, self-defining technical objective.
Free tools Windows power users keep installed
One-click scans. No signup required.
In “Artificial Intelligence, Values, and Alignment,” published in Minds and Machines on 1 October 2020, scholar Iason Gabriel argues that the aim should not be to discover one uncontested moral doctrine and encode it. Instead, the challenge is to find fair principles that people with differing moral beliefs could reflectively endorse. Gabriel describes alignment as a political problem rather than a metaphysical search for the one true moral theory. Read the article in Minds and Machines.
#1 Best Overall
Why is choosing the process as important as choosing the principles?
When people disagree deeply about morality, a list of rules does not explain how that list was chosen. The selection process affects whose interests are represented, whether some groups gain an arbitrary advantage, and whether the resulting rules can earn lasting support.
Gabriel identifies desirable qualities for such a process: procedural fairness, genuine inclusion, concrete guidance, stability, robustness, and comprehensiveness. These qualities can pull in different directions. A process might include many viewpoints yet struggle to produce specific rules; a concise set of rules might be actionable but fail to reflect the diversity of views it is meant to serve.
Three possible ways to choose AI principles
Gabriel discusses three broad approaches. They are candidates for addressing the problem, not established solutions, and each leaves practical questions open.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
| Approach | How it works | Key questions |
|---|---|---|
| Global public morality and human rights | Seek principles capable of broad support across different moral and philosophical perspectives. | Which rights and principles can command that support? How can broad agreement be translated into concrete instructions? |
| Aggregation of moral judgments | Combine judgments made by individuals. | Whose judgments count, how are they measured, and what method combines them fairly? |
| Democratic procedures | Develop principles through voting, discussion, and civic engagement. | What makes the process legitimate, and how much endorsement should foundational rules require? |
Global public morality and human rights
This approach looks for a common foundation that can cross moral and philosophical differences. Its appeal is that principles need not be dictated by one specific doctrine. But broad agreement alone may not tell developers exactly how a system should respond in a particular situation.
Rank #3
Aggregation of moral judgments
Combining people’s judgments may appear to represent a population directly, but the result depends on the design choices behind the combination. Selecting participants, deciding what questions to ask, and choosing how to weigh responses can all influence the outcome.
Democratic procedures
Discussion and voting give people a role in shaping the rules, rather than treating moral guidance as a purely expert decision. The approach still needs safeguards for legitimacy and a way to decide what level of collective endorsement is enough for basic principles.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What does this framework say about Anthropic?
It does not establish Anthropic’s methods, products, decisions, or model behavior. Gabriel’s article is general scholarly context about AI values and alignment, not reporting about the company. It therefore cannot substantiate claims that Anthropic has adopted any of these three approaches or settled the moral questions they raise.
The useful takeaway is narrower: any effort to give AI moral guidance must make choices about its alignment target, whose judgments inform its rules, and how disagreement is handled. Those choices deserve scrutiny alongside the technical work of implementing the system.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
