Software Agents Can Help Time-Stressed Teams

Roland Piquepaille writes "Penn State researchers have developed software agents which can help human teams to react more accurately and quickly in time-stressed situations than human teams acting alone. According to this news release, the software was tested in a military command-and-control simulation. "When time pressures were normal, the human teams functioned well, sharing information and making correct decisions about the potential threat." But when the pressure increased, the human teams made errors who would have cost lives in real situations. The decisions taken by agent-supported human teams were much better. Now, it remains to be seen if this software can be used in other stressful situations, such as for emergency management operations. Read more for other details, references and illustrations about this project."

Comments

Assisted Operations

[Renraku]

Software assistance isn't so bad when you can click cancel and it takes all of one second out of your life. In a combat situation? What if something outside of the plan happened? Which often does during times of war and/or duress.

I look at it this way...

[Announcer]

Anything that can be used to save the lives of our brave young men and women is worth discussing and implementing.

If this software can be set up and used by our Military, even if it only saves *one* life, it will have been worth it.

Re:I look at it this way...

[mikelieman]

"Anything that can be used to save the lives of our brave young men and women is worth discussing and implementing."

I agree with you, we need to get out of Iraq yesterday.

Pulled both ways

[soma_0806]

This is a tough one. My initial reaction is, "Egads, (yes, I talk like this), with the possibility for bugs, hang-time, etc., obvious NO!"

However, if in reality, the possibility of some glitch causing a bad decision is, say, 1 in 100, and the frequency of these pressurized teams making the obviously wrong decision is around 1 in 10, then I'd say go for it.

The point is, we need to know the rates/probablities of failure for both systems. Failure of some sort is inevitable, just how often and how badly are possible to control.

The same model works in self defense

[Colin Smith]

If you have to consciously think then you lose. Thinking takes too long. It has to be reflex/muscle memory/autopilot or whatever you want to call it.

Which is why all the repetitive training, the high pressure fighting during practice. And it's also why books and videos though good for imparting information can never help when the real thing happens.

Not on my watch

[zappepcs]

I have been part of teams working under pressure, and there is little that I can see of value in a clippy if the team is actually well trained and have worked together.

Human interaction changes dramatically under pressure to perform. Long sentences can become single words or syllables, yet full communication is achieved. An well trained team member begins to anticipate the action of other team members, in ways that clippy cannot do.

The parallel like processing of the human mind still outperforms that of any computer in small paradigms. Even in military situations, no computer application can apply all the relelvant information from other team members and information sources in a way that can replace or even assist in those decisions. If a team member forgets part of their job, it is usually not because s/he is under stress, it is because of lack of training or experience. Substituting computer assistance for training and experience is an EXTREMELY dangerous thing in my opinion, especially where human life is at risk.

Just my two cents.

Re:Not on my watch

[zappepcs]

The trouble, as I see it in what you say is that the nuances of human communication cannot be fully categorized or recorded. Today in training, the word 'gotcha' may indicate target lock for two team members working in a tank, tomorrow, or even in three hours, it could signify that one team member understands the other. Situational relevance of communication causes extreme difficulty for computers to immitate humans, or even play along. That is why experience and training cannot be substituted with a computer.

Talk about biased "tests"

[Shadowlore]

From the link:
In the simulation, team members had to protect an airbase and supply route which were under attack by enemy aircraft. The scenarios were configured with different patterns of attack and at different tempos. The situation was complicated because team members had to determine at first if the aircraft were neutral or hostile. Furthermore, two team members were dependent on the third whose role was to gather information and communicate it to them.

So a three person team set up to be fully dependant on a single person. Hello? Any CS major with half a brain can tell you what will happen there. So could any decent sysadmin. Resource contention caused by a bottleneck. So their second "team" of agent assisted humans ... take a look at the illustration. It is one person and two (three actually) "agents".

They basically made a very simple RTS. The they "discovered" that it is faster when your information distribution is faster. This is NOT "agent assist". And who didn't think that a computer program with direct data input, that doe snot need to move input devices, scan a screen and process would NOT be faster in disseminating simple information like that?

This scenario is so far from reality in any situation that you cannot call it a simulation of reality. The conditions are far, far too simple and remove *any* intelligence from the "s3" and "s4" roles. If you are told to kill it, you do - and you get penalized heavily if you were told wrong information. This is important. It basically means that the role of "s3" is best suited for a computer. Combined with the inherent speed boost for information distribution and simple tests for the role of S2 this along will produce "better" results. Their S4 role is essentially less intelligent than "s3". "Move from A to B unless told to run away".

On top of that, they set it up such that one unit was defending two different areas; one in motion.

Ironically, the one area agents could in theory help out here is the one they specifically stated the human brain is better at; spatial reasoning. Go figure.

What some of the other posters need to be aware of is that this scenario is not the same as in self defense or life-death sequences. So rants about that are basically off topic.

One final interesting observation. They stated that at maximum speed no human team could destroy any target, but the computer .. oops I mean the HA teams could still destroy 36% of their targets. Seems to me that the limitations of the human body in move-click-process-move-click for the S2 role are a serious limiting factor they did not account for. Any RTS/FPS gamer will tell you these factors are not small. A better interface would have been a three key layout. Press I to identify, N for neutral, A for attack. Perhaps using the tab key to select from available targets. This would have improved the human performance merely by decluttering and improving their effective reaction times.

Then to further eliminate inherent diferences that have nothing to do with agent and decision making, there should have been a delay incorporated into the agents to account for the remaining difference in UI effects. At least then it would have been interesting.

Software agents may yet have a purpose in such condtions, but it won't be at this level, and this "study" doesn't demonstrate they would have any real value; it only demonstrates that you need battlefield intel to be disseminated quickly. Agents may have a use at a much higher level than was used in this experiment.

I've done battlefield intelligence. We don't need agents to identify friend or foe. We need a fast, easy to grok at a glance view of assets, terrain, and intel.

This is one reason that on the battlefield, attack units are assigned *directly* to intel units - so they can react and respond without waiting on information to filter up and down the chain.

Military victories are nearly always based on who has the better intelligence and data. When you've got a superior method of information distribution, I'll be interested. When you just want to tell me that computers can do some things that are irrelvant to the application domain, it is a waste of time and resources.

YES

[Moraelin]

Actually, we already _are_ using software agents to work a lot faster. Compilers, IDEs, frameworks, you name it. That's what they're there for.

Agents to help decision making? Well, that's what syntax highlighting, auto-completion, help files, and other tools in the IDE do for me. They let me decide faster what can I use there.

(Which also addresses the flood of "ugh! they're making Clippy!" posts. There are at least a dozen tools I use every day that aren't Clippy. Just because one tool is retarded, doesn't mean they all are.)

And they _do_ allow us to achieve deadlines that were unthinkable back in the days of coding in hex/octal and counting the bytes by hand.

The problem isn't the reliance on _good_ tools. The problem is, well, bad management. (Including buying the wrong tools, but that's a topic for itself.)

I really hope more managers will read threads like these, because there's one important message there: stressed people make more mistakes. And according to other studies, some of which were linked to by /. too, tired people make more mistakes too.

And between those two, you have the whole picture of what's wrong with 84 hour weeks and other PHB-style management techniques. It's not that programmers aren't soldiers. It's just that humans (programmers, soldiers, etc) are not machines. A computer can work on SETI packets 24x7 and do proportionally more work than 8x5. A human can't.

Since in programming most of the time is spent in debugging and maintenance, not in just typing code, past a point it's exactly that making more mistakes (which need to be debugged... again) and taking weird shortcuts (which will bog down maintenance) that's ending up costing more time than it saves.

Not that I'm setting my hopes too high, though. There are managers which do have a clue, and then there are the PHB's. Those who fall in the second category, well, I just can't see them getting a clue, even if it was written in big letters on a billboard in front of their office.