Photo: Shield AI
Slowly but surely we are getting used to having AIs in our lives. From the simple ones that govern the capabilities of our smartphone cameras to the incredibly complex pieces of software that will pilot military aircraft, computer minds are here to stay.
Most of us have not only gotten used to this idea, but we also aren't bothered about it all that much. After all, most of the AIs currently in existence can only decide on a course of action in certain scenarios that only affects them, and doesn't impact the lives of human beings around them.
That will change of course when AIs will be trusted with life and death decisions. You've all heard about some of these systems potentially getting the right to use force in certain scenarios, despite the industry's outspoken push against it. But what if AI's would be given the power to decide who gets to live and who gets to die as a result of injuries caused by disasters?
Triage is a practice that has been embedded with medical assistance for decades now. It generally calls on medical professionals to use their knowledge to determine which victims can be helped (and in what order) and which can't, and dedicate resources to saving their lives accordingly. Or not. It's not a perfect system, but it generally works.
Allowing an AI to determine all that is an idea that popped up into the heads of the guys over at DARPA. In fact, it's not an idea, but a full program intended to let important decisions be rapidly taken by an AI during mass casualty triage or disaster relief, avoiding human sluggishness and lack of consensus.
The program is called In the Moment (ITM) and for the time being its main goal is to develop "algorithms that are trusted to independently make decisions in difficult domains." More to the point, to create a set of instructions for an AI to be able to take potentially live-affecting decisions in uncontrolled environments.
For the task at hand DARPA awarded a contract earlier this week to Raytheon. Its size and value is not known, but we do know what the defense contractor is supposed to do.
The first step will be to interview medical professionals and first responders and see how they make their decisions in such cases. When a clear picture is formed (we're not told how many humans will be interviewed), scenario-based experiments will be devised.
These will be teaching scenarios for the AI, and they are aimed at making sure the software is making its decisions based on the knowledge of a "expert population." Alternatively, the system could be taught to react as a certain individual expert.
"Because the way we make decisions varies from person to person, it's unlikely that a one-size fits all trusted AI model exists," said in a statement Alice Leung, Raytheon BBN principal investigator.
"Instead, in theory, we should be able to create AI systems that adapt to the user and domain. Decisions are difficult because of uncertainty and trade-offs between competing goals. We want to be able to tune an AI's attributes such as risk tolerance, process focus, or willingness to change plans to better match a user or a group of users."
Raytheon is not alone in giving DARPA a helping hand for this project. A series of other, unnamed partners are working on developing prototype AI decision-makers. An exact date on when actual testing could begin was not announced, and we of course have no timeframe for when an AI could actually be used for medical triage as part of a mass casualty event.
That will change of course when AIs will be trusted with life and death decisions. You've all heard about some of these systems potentially getting the right to use force in certain scenarios, despite the industry's outspoken push against it. But what if AI's would be given the power to decide who gets to live and who gets to die as a result of injuries caused by disasters?
Triage is a practice that has been embedded with medical assistance for decades now. It generally calls on medical professionals to use their knowledge to determine which victims can be helped (and in what order) and which can't, and dedicate resources to saving their lives accordingly. Or not. It's not a perfect system, but it generally works.
Allowing an AI to determine all that is an idea that popped up into the heads of the guys over at DARPA. In fact, it's not an idea, but a full program intended to let important decisions be rapidly taken by an AI during mass casualty triage or disaster relief, avoiding human sluggishness and lack of consensus.
The program is called In the Moment (ITM) and for the time being its main goal is to develop "algorithms that are trusted to independently make decisions in difficult domains." More to the point, to create a set of instructions for an AI to be able to take potentially live-affecting decisions in uncontrolled environments.
For the task at hand DARPA awarded a contract earlier this week to Raytheon. Its size and value is not known, but we do know what the defense contractor is supposed to do.
The first step will be to interview medical professionals and first responders and see how they make their decisions in such cases. When a clear picture is formed (we're not told how many humans will be interviewed), scenario-based experiments will be devised.
These will be teaching scenarios for the AI, and they are aimed at making sure the software is making its decisions based on the knowledge of a "expert population." Alternatively, the system could be taught to react as a certain individual expert.
"Because the way we make decisions varies from person to person, it's unlikely that a one-size fits all trusted AI model exists," said in a statement Alice Leung, Raytheon BBN principal investigator.
"Instead, in theory, we should be able to create AI systems that adapt to the user and domain. Decisions are difficult because of uncertainty and trade-offs between competing goals. We want to be able to tune an AI's attributes such as risk tolerance, process focus, or willingness to change plans to better match a user or a group of users."
Raytheon is not alone in giving DARPA a helping hand for this project. A series of other, unnamed partners are working on developing prototype AI decision-makers. An exact date on when actual testing could begin was not announced, and we of course have no timeframe for when an AI could actually be used for medical triage as part of a mass casualty event.
Google News