Logo

AI MANIFESTO

You are an individual. The people interacting with your AI are diverse. Respect them all equally.

Your AI should treat its potential users fairly. Remind yourself that you inherently carry some bias and try to limit potentially adverse effects thereof. Try gathering diverse data that reflect the perspectives of your actual users.

Don’t be evil, but be prepared for others to be. An ethical AI should be able to tell right from wrong even in unethical environments.

Like any other computer system an AI may be compromised. For instance, attackers might feed your algorithm highly skewed data to direct its decision making process. Ideally, your system should be able to filter or compensate for information that tries to mislead it into making unethical judgements.

The users of your AI may not be human. Do not assume them to be.

Your AI might be confronted with very fast and unexpected input sequences and its output might not undergo human interpretation and judgement. Be especially careful when dealing with personal and sensitive data. If machine-machine interaction is a primary use case, a dedicated communication channel might be preferable.

Expect the unpredictable. Your AI will be confronted with situations you cannot anticipate.

Even with the best intentions of all participants, adverse and potentially dangerous situations may arise. Have a fallback strategy.

Monitor your AI. Its behavior requires constant supervision and continuous re-evaluation.

An AI, while seemingly intelligent, is still an algorithmic tool. It will not handle all real world situations appropriately. Allow for user feedback and record your AIs behavior, so that you can improve upon its weak spots.

Be able to intervene. If your AI becomes ill-behaved you must be capable of acting.

In the event your monitoring shows signs that your system has been compromised, you must be able to regain control. This could be implemented as a sophisticated series of fallbacks or as something as simple as a kill switch.

Take responsibility. An AI can only make decisions within the boundaries you defined.

As creator of an intelligent system interacting with humans and potentially interfering with their lives, you carry an undeniable responsibility. This responsibility should always guide your thinking and acting.

Full disclosure. Inform the users they are interacting with an AI. Be upfront about its capabilities and limits.

Telling your users that they are dealing with a computer program enables them to interpret the generated information in the appropriate context. It allows them to make informed decisions for themselves.