Human-Centered AI: Ensuring Human Control, Enhancing Human Performance, by Ben Shneiderman
When you move to automation you lose some of human control… right? Not necessarily, if you have human-centered AI. Human control can be low or high, and computer automation can be low or high.
For example, a bicycle has human control, and low computer automation. But a camera or an elevator has a high amount of human control (buttons give control over where to go or what gears to use) and high computer automation. This is the HCAI framework.
That said, there are things that use too much automation – when the Boeing 737 overrode pilot control. There are also situations with too much human automation. Morphine drips are an example where computer automation could control the drip and put a limit on how much a person could ask for, but it’s still patient guided, in that the patient doesn’t get it without asking.
Examples and metaphors of human-centered AI
The HCAI (Human-Centered AI) framework is a table with an X axis of computer control, and a Y axis of human control. Of all the types of wheelchairs, there are robotic wheelchairs, push chairs, essentially all types within the HCAI framework!
- Intelligent agents – now we think of supertools
- Teammates – now we call them telebots
- Assured autonomy – now we have control centers
- Social robots – now we talk about active appliances
Lots of AI, machine learning, and automation, but also lots of human control.
- Digital camera controls – they automate a lot, but you make the final choice/can override
- Navigation choices
- Texting/search autocompletion – you don’t have to take the suggestions
- Recommended spelling correction – you can make the choice to take the correction
- Tele-bot from Da Vinci surgical system. “Robots don’t perform surgery. Your surgeon performs surgery with da Vinci by using instruments that he or she guides via a console.” -da Vinci website