The 10 Principles of Intelligent Agent Design

Originally published in Venture Beat in 2017

We’re on the cusp of a new generation of better and more sophisticated intelligent agents.

Intelligent agents are fast becoming ubiquitous in personal life and business, which means they are an important area of opportunity and interest for innovators. As entrepreneurs, designers, product managers, developers, and investors, we should step back and think about the principles behind what we’re building. It’s a great time to start laying out practices and principles for how we want to design and build intelligent agents. As part of that thought process, below are 10 principles that can help govern the future design of intelligent agents.

1. Useful

An intelligent agent should be of service to others. The ideal intelligent agent assists humans in an explicitly useful fashion. Some agents may assist other agents or be a part of a larger process. But they must be useful.

2. Context-aware

An intelligent agent should understand context, including time, place, and many others. It should understand as much about the world as possible in order to complete tasks, provide the right information at the right time, and learn.

3. Able to learn

An intelligent agent should become more intelligent over time. Every interaction with humans, other computer systems, and the world is an opportunity to learn. Some of the information the agent collects will increase its store of knowledge and some will make it better at its core functions, such as communicating and taking the right actions.

4. Adaptable

An intelligent agent should be able to change in order to apply what it has learned. One area for adaptability involves being sensitive to users’ preferences. There are many preferences that the agent can learn in order to mold its behavior — these include communication preferences and privacy parameters.

5. Able to communicate well

An intelligent agent should pursue maximum eloquence. It should be intuitive, self-explanatory, and responsive to whatever entity it is communicating with. It should use the most efficient communication method (such as text, voice, graphics, and API) for the context of the communication.

6. Proactive

An intelligent agent should anticipate the needs of others. It should combine what it has learned with context to be able to take proactive action. However, it should be proactive with communications only if the information is useful to the recipient.

7. Respectful of privacy

An intelligent agent should respect information privacy. Based on privacy requirements, different information can be shared with different parties. It should understand the privacy parameters of providing or receiving all different types of information with all different parties.

8. Honest

An intelligent agent should provide correct information. It shouldn’t deceive or make promises that can’t be kept.

9. Transparent

An intelligent agent should allow appropriate access to information that impacts others. For example, it should allow appropriate access to view settings, configuration controls, and defaults that impact humans.

10. Non-malevolent

An intelligent agent should not participate in harming others. If a requested action is harmful, it should warn the user about the consequences of the action being requested.

I don’t think every intelligent agent needs to follow all of these principles. But if one of these principles isn’t included in a design, the reason for doing so should be understood. Omitting something for a specific purpose is better than leaving it out because it was never considered. Keeping these principles in mind will help developers build robust future intelligent agents that improve as the technology advances and we get better at their design.