[Jeroen]

Originally Posted by Motard_Blr

But AI is tricky; can we trust it in safety-critical systems? Humans are social beings, understand morality, the concept of action and consequences, and seek social approval and respect. What is to stop an AI system from killing us all even if it is following the laws of robotics https://en.m.wikipedia.org/wiki/Laws_of_robotics?

Thanks for those comments. I think there is far more immediate and pressing problem with AI. It will be quite some time before we have AI system that become independent and start killing us.

The biggest problem we have with AI today is that already it is becoming increasingly more complex to understand what they are doing or more importantly why. Monitoring AI systems in terms on how they are self adapting and learning and understanding they are still doing what we want them to do is already a challenge. It is one thing if a FB AI algorithm goes off track and start throwing adds to the wrong persons.

But AI picking up bad habits from pilots is something very different.

One of the problems I can see for AI in the cockpit is the following. The cases where, at least in theory, AI could make a huge difference are actually quite small. E.g. emergency situations where based on multiple (part erroneous) reading from various instruments and sensor a decision needs to be made. (E.g. the Max and AF44 accidents)

How are they going to train these AI algorithms? How do you select the correct meaningful scenarios that give sufficient data to train an algorithms?

Jeroen

[Sutripta]

Originally Posted by Jeroen

Everybody jumped onto the "blame Boeing bandwagon" early on.

Early on? That I don't think is accurate.

It was only after the second crash, when information started coming out that the baying for Boeing's blood started. I think it was because people realised that Boeing had blood on its hands, but had almost got away with it scottfree. (In terms of penal action, it still might).

Regards
Sutripta

[Motard_Blr]

Originally Posted by Jeroen

The biggest problem we have with AI today is that already it is becoming increasingly more complex to understand what they are doing or more importantly why. Monitoring AI systems in terms on how they are self adapting and learning and understanding they are still doing what we want them to do is already a challenge. It is one thing if a FB AI algorithm goes off track and start throwing adds to the wrong persons.

...

Jeroen

Yes, the lack of transparency is beginning more apparent with each successful AI implementation. Whether it is in medicine or in logistics. Actually, the problem might not be that we don't know how an AI system works or how it has arrived at a solution to a problem, but, that we don't have the intellectual capacity to understand the working of the AI because of the non-human way it thinks.

Trusting such an AI system will definitely be a leap of faith.

[Sutripta]

Originally Posted by Jeroen

The biggest problem we have with AI today is that already it is becoming increasingly more complex to understand what they are doing or more importantly why.

More than a decade now, but would any engineer at your company put his life in the hands of a system which designed the 'Borg antenna' after taking a look at the antenna! (I believe it was actually used by NASA).

[Jeroen]

Originally Posted by Sutripta

More than a decade now, but would any engineer at your company put his life in the hands of a system which designed the 'Borg antenna' after taking a look at the antenna! (I believe it was actually used by NASA).

No self respecting engineer would accept any other engineer’s solution at face value. Let alone a solution that was gobbled together by a bunch of PC’s.

But seriously, what that approach showed: conventional wisdom/engineering can be very limiting. The blasted Borg came up with something that did not look like anything, but still met the specifications better than anything conventional wisdom and engineering could envisage.

We are using AI in our products and our services more and more.

https://www.ericsson.com/en/managed-services

We have been relying on all sorts of automation for our managed service delivery since we started Managed Services, AI is just one of the next steps. It is only one component in what we are changing. What is more important is that the Ericsson Operations Engine as mentioned here is not just about AI. It is a fundamentally different way of how we support the business of our customers. It is also far more than a bit of automation. With it comes a completely new way of interacting with our customers, a different organisational set up, a new way of contractual models, T&Cs, pricing models, different and new jobs with new roles and responsible, completely new processes, adjustment in ways of working, culture etc.

AI has tremendous potential, but it takes a lot of thinking on how to make effective use of it. AI standalone is pretty useless. It can do some fancy analytics for you, but in order to make use of that you need to make some major chances into your organisation as well.

Jeroen

[pganapathy]

The catch with relying on upgraded legacy systems is that some minor bug that was irrelevant 20 years ago begins to matter. The only solution to this is designing things from scratch, but that is expensive. Which explains why the 737 has been manufactured for so many years.

Evolution improvements can only fix so many things. Look at the Porsche 911 for example. It is definitely less tail happy than earlier iterations courtesy of a better weight distribution, but with a clean template the 911 could be a better car.

[.sushilkumar]

Originally Posted by Jeroen

The biggest problem we have with AI today is that already it is becoming increasingly more complex to understand what they are doing or more importantly why. Monitoring AI systems in terms on how they are self adapting and learning and understanding they are still doing what we want them to do is already a challenge. It is one thing if a FB AI algorithm goes off track and start throwing adds to the wrong persons.

I work in Data centre space and organisation are moving into anything "Software defined", be it network or storage. Nothing wrong in there but all these technologies will need 2 or 3 dacades to mature before i can go an risk my bussiness with them. I remember compute virtualization started in circa 2003 -04 and now in 2019. 2 decades later, they are still not that reliable That they can work without a bug check for a year. They are just there but bugs do keep track and hit when not needed most. Amost every customer is burning their hand with software defined and going back to where they started.

Some thing similiar with these AI and big Data. We need to give them atleast 2 or 3 decades to mature and be mainstream. I am just horrifed to see such critical systems being handed over to robots / computers. Planes are meant to be flown by Piltos not computers . Some level of automation is acceptable but having a system like MCAS taking over is just plain stupid and dangerous. I always remember this MCAS while boarding a Boeing aircraft & recent deals in Dubai air show proves that airlines are not willing to risk with 737 MAX or boeing in general anymore. IMO it's cursed name now and even if they came back , people will be reluctant to board one.

Not that Airbus is any better in terms of automation but still they have been in limits so far and have proven reliable so far in all my multiple 16 hours atlantic crossing so far

[Jeroen]

Originally Posted by .sushilkumar

I work in Data centre space and organisation are moving into anything "Software defined", be it network or storage. Nothing wrong in there but all these technologies will need 2 or 3 dacades to mature before i can go an risk my bussiness with them.

Not sure I understand, You say you work there, but you will not risk your business with them? I take it you do not work at their PR department?

Originally Posted by .sushilkumar

I remember compute virtualization started in circa 2003 -04 and now in 2019. 2 decades later, they are still not that reliable That they can work without a bug check for a year. They are just there but bugs do keep track and hit when not needed most.

So are you saying that all other engineering then software is fault free? Each type of technology comes with its intrinsic limitations and risks. A proper design takes those into consideration.

Originally Posted by .sushilkumar

Amost every customer is burning their hand with software defined and going back to where they started.

So your company is going out of business fast then? Not sure, that is how most
Data Centre experts see their business going?

Originally Posted by .sushilkumar

Planes are meant to be flown by Piltos not computers .

Says who? Most aviation expert will tell you and so do aviation accidents that the pilot is actually the weakest link in the whole safety chain! Pilot error is still the major contributing factor in the vast majority of incidents and accidents. It has come down considerably over the last decades, partly because of automation.

Originally Posted by .sushilkumar

Some level of automation is acceptable but having a system like MCAS taking over is just plain stupid and dangerous.

The way Boeing and FAA went about it was stupid and dangerous. MCAS is a trim system. MCAS is unique in how it was envisaged, designed and introduced in the plane (without anybody knowing). Any modern air liner has a bunch of automated trim system already and they have been fine for decades.

What level of automation would you find acceptable? Could you elaborate?

There are dozens and dozens of fully automated systems on a modern aircraft. Remember flight engineers, navigators, radio operators? We had all of them in the cockpit not too long ago. All replaced by clever automation in the space of a few decades.

I do not want to worry you unduly, but aviation is relying heavily on all sorts of automated systems. Whether it is air traffic control, weather forecast or the pilots in the cockpit or the maintenance department. When you are happily cruising at 35.000 feet, it is the autopilot (essentially a bunch of computers and automated systems) who is controlling the plane.

If you find yourself in RNAV airspace you are only allowed to enter with working autopilots, because humans can not be relied upon controlling vertical and horizontal separation to these narrow standards.

Originally Posted by .sushilkumar

Not that Airbus is any better in terms of automation but still they have been in limits so far and have proven reliable so far in all my multiple 16 hours atlantic crossing so far

There is no statistical difference in reliability and or accident rates between Boeing and Airbus, despite a different approach to flight controls. At least I am not aware of any research saying so, one way or the other.

Hope I am not making you nervous but (almost) all Airbuses are fly by wire. The pilot moves a joy-stick. That provides inputs to some computers who are actually flying the plane. There is no mechanical back up on an airbus. If the computers go wonky, or to your point, have a bug, you are toast!

When you are crossing the Atlantic it is computers that give your plane the actual clearance to enter the Atlantic traffic scheme. Pilots do not talk to ATC anymore to get this particular clearance. It was to cumbersome, too slow and too many errors. So all done by computers and automation these days.

So to my earlier question; what automation would you like to do away with?

Jeroen

[.sushilkumar]

Originally Posted by Jeroen

The way Boeing and FAA went about it was stupid and dangerous. MCAS is a trim system. MCAS is unique in how it was envisaged, designed and introduced in the plane (without anybody knowing). Any modern air liner has a bunch of automated trim system already and they have been fine for decades.
Jeroen

Thats what my problem is in bold above. We need to rely on a well tested and enggineered system rather than a half tested. Airbus do use fly by system but it was well tested and implemented and have proven to be reliable with multiple redundant systems in place.

Incase of Max, We should have stopped at Lion air crash itself and questions were raised there as well on MCAS but it was hidden well by boeing and FAA, who were found sleeping on same bed. it took life of few more hundreds to bring them to senses.

Technology is good but it must be tested when it involves human life.

[Jeroen]

Originally Posted by .sushilkumar

Thats what my problem is in bold above. We need to rely on a well tested and enggineered system rather than a half tested. Airbus do use fly by system but it was well tested and implemented and have proven to be reliable with multiple redundant systems in place.

Technology is good but it must be tested when it involves human life.

Sorry, but you write a lengthy post about software being unreliable, you make multiple suggestion about the need to have less automation in planes

You state:

Some level of automation is acceptable but having a system like MCAS taking over is just plain stupid and dangerous

Which again suggest you have concerns on automation. (At least that is how i read it all)

But you believe systems need to be properly designed and tested. I do not think you will get anybody to argue against that. It would have been helpful if you said so in the first place.

Data centres still going out of business though? Not properly designed and tested?

Jeroen

[.sushilkumar]

Originally Posted by Jeroen

Sorry, but you write a lengthy post about software being unreliable, you make multiple suggestion about the need to have less automation in planes

You state:

Which again suggest you have concerns on automation. (At least that is how i read it all)

But you believe systems need to be properly designed and tested. I do not think you will get anybody to argue against that. It would have been helpful if you said so in the first place.

Data centres still going out of business though? Not properly designed and tested?

Jeroen

Apologies if i am not clear. All i am saying is that Boeing should have tested MCAS thoroughly before putting in actual use. It did not helped that regulator itself was found wanting . In Hind sight, i beleive they went a step ahead while trying to create a system which helps the pilots in command. Instead the system overruled the pilots and precious human lives were lost.

i was just creating a analogy with data center's as i see similiar happening here as well in the race to be "software defined.

[Jeroen]

Originally Posted by .sushilkumar

All i am saying is that Boeing should have tested MCAS thoroughly before putting in actual use.

They should, obviously. But then there is never 100% guarantee.

Your beloved Airbus is suffering from design flaws as well!

OJK also found that a "design flaw" allowed a single event – a shallow rebound during the landing – to trigger a consolidation logic discrepancy between the spoiler elevator computer's two channels, with one computing elevator orders in 'flight' law and the other computing orders in 'ground' law.

https://www.flightglobal.com/news/ar...cciden-462014/

Everybody survived but the aircraft was a write off!

Jeroen

[Sutripta]

Would someone explain to me the difference between automation and AI. (In ones own words please. Not links to learner articles which would just further confuse us laymen). Or are these one and the same? Because that is the way it seems to be used in this thread.

Regards
Sutripta

[Thad E Ginathom]

AI requires that the system be fed a huge number of examples so that it can learn. It can then work with other stuff that is within it's "experience."

There is no such thing as a machine That is intelligent "out of the box."

[Jeroen]

Originally Posted by Sutripta

Would someone explain to me the difference between automation and AI. (. Or are these one and the same?

They are two very different thing all together.

In essence automation means something happens without human intervention. A very simple example: The outside light on my shed turns on automatically when I walk near to it. It has a simple sensor that can detect movement and it will automatically (no human intervention) turn on the light.

A more complex example would be an auto-pilot. An auto-pilot is a system which is capable of flying an aircraft though various vertical and lateral pre-defined parts. The pilots provides input to the autopilot. The autopilot will manipulated the flight controls, throttle etc without any further human intervention.

Another simple automation example is a macro in an Excel sheet.

Automation enables some sort of automated (non human) output, based on pre-defined inputs/condition.

AI is in essence a very advanced data analytics capability. AI allows you to process vast amounts of data. AI is a technique that is very good at finding trends, pattern in vasts amount of data. Which also means it can be used to predict outcomes.

In it’s most rudimentary form AI is just some software that sits on one or multiple computers and can be run stand alone. So somebody would feed it data and it would spit out a data analysis report. The report can be used to take certain actions.

It will require automation (the term actuation is often used in conjuction with AI) to do something with the dat, to take action. That could be done by a human, or you could automate it.

What AI brings is superior data analysis that can have an element of predictive in it.

An example: We use so called User Cases in our AI. One of these Use Cases is predictive maintenance. Currently we have reactive and preventive maintenance. Reactive means something happens in the network, alarms are raised, correlated and our opeartors decide a course of action. They might log remotely into a certain node in the network, they might dispatch a field technician Etc.

Preventive maintenance is essentially a scheme at which at certain interval (usually based on time and or cycles) we send a technician to a site to do some preventive maintenance.

With AI we are able to predict with a very high degree of reliability where we are likely to see problems in the near future. This allows us to become pro-active. So we can respond before anything happens.

In analogy with my earlier example, our AI capabilities provide our operators with a report which shows what sites could be facing what sort of problems, when. They can then decide the most appropriate intervention. We might be able to fix it remotely, we might have to send a field technician etc.

But of course, we can also automate whereby the system itself determines what the best cause of action is. It could log on remotely with no human intervention and maybe do an automatic software reset, or upload the latest patch.

Or it could automatically dispatch a field technician. Again, no human intervention

So automation means no, or less human intervention, to action something.

AI is just very nifty data analytics.

Incredible amounts of data can be analysed and it can be used to predict what might happen. What you do with the result of such an analysis will depend. The required action can be automated or not.

AI due to its very nature is a bit more complex of course. Even obtaining (data ingress), storage and manipulating the input data so it can be used require high level of automation themselves.

AI is often used as sort of the next level of automation. Technically that is incorrect. AI is a capability (provided by several components, e.g. computers, software, storage etc) . What is true is that the combination of automation and AI gives us unprecedented capabilities, far beyond what we can achieve with automation without AI.

Hope this helps

Jeroen