Hi,
Natstrackalpha words:
As i told in my
post the comment was from
TTex600:
in my mind two differing interfaces need discussion. the "man - machine" interface and the "human intelligence - computer intelligence" interface.
I agree with safetypee:
I have some agreement with this view,
(my bold)
The "dangerous confusion" may come from an automated interface (with no AI built in). Simple protections can present "inputs" to the crew difficult to be understood in certain situations. This can occur with just combinational logic. E.g. a simple interlock. When we introduced microprocessors (i designed with the first one: Intel 4004) you put the so called "Finite State Machines" in the "loop". They memorize and it's outputs DEPENDS on new inputs and the MEMORIZED data. look, i am not talking of AI. Just saying crew may face situations where an almost immediate understanding could be vital and could be forced by training to clear several memorized states, just to allow the System to return to normal thinking
(operation).
Problem as i see is: The interface must be with K.I.S.S. principle built in to allow a safe operation (in abnormal situations). Even dumb interfaces can present challenges to a crew. And training is concentrated on predictable issues. We are capable before new situations to deliver an interface that really ALWAYS help the crew? I think is just not possible to assure. What's the solution? Very simple: a well prepared pilot (not just by Pavlov training, memory items, P&P approach, etc.) can do miracles. Why. Because his Natural intelligence is "open" to highly creative solutions. We may list examples of this and examples of incidents and accidents where confusion was installed from even dumb interfaces. like the LOC of Thiells 727.
You know, you should start a thread on Artificial Intelligence.
After understanding the issues presented by microprocessor based interfaces with just algorithms built in we could discuss something on AI in an "off topic" way. I don't see reasons to complicate the thread with this component. The problems we observe in the current interfaces IMHO recommends attention. Because seems complex enough to, in some cases, cause "CRM issues" between the pilots and the machine. In a sense tha both (partners) don't work together to effectively use all available resources to save the A/C.
AF447 may be is a dramatic example:
1) The plane had no failures* (the sensors were operating as per design)
2) Crew including Capt. had all resources available (engines, system, etc.)
3) The had no chance to "decipher" the interface outputs and were forced to rely on other inputs
4) They aggravated the initial issue and were caught in a "unsolvable surprise"
5) They never understood what really happened
Despite the interface being MUCH MORE advanced than the 727. Confusion. And the result was the same. Crew were put in a "terminal state" requiring a super pilot. Even test pilots could not leave this "state"
* Just a temporary "cold" in 3 sensors
In the aftermath of AF447 case what could be made? It seems to me this is a (dramatic) wake up call to the importance of the interface (and it's characteristics) to the Safety of the "advanced planes". I don't operate "the thing". I know what an interface MUST present:
ALWAYS present to the crew the possibility to NEVER be caught in a surprise impossible to be solved, specially when the plane has no failures. Design (technology limitations, whatever) never could put the crew in the situation PF+PM, then Captain entered.
This (confused crew with no time to even understand) could happen again? I think so.
Why? IMO because the interface could be improved. And probably as i understand better without adding extra complexity, like AI features