Internet of Everything

How much work is being done in this regard and what are the successful applications so far?

------------------------------
RAFI AHMED
Mobily
------------------------------

Hi Rafi,

This is a great question, and TM Forum are just about to start the research for an AI report that will come out in December. I have been doing some initial research and at the moment AI is still in it's nascent stage and there has been little work one in AI in M2M that one can measure from an industry point of view. Most of the analysis that has been done to date, relates to the market size, worth, and opportunity. Moreover, the biggest success to date in AI is around chat bots. This are by far the largest deployed "intelligent" systems, but one could argue is that really AI. 

Where we will see the biggest deployment of AI from a telecom perspective in the next 18+ months will be around product catalog as a lot of work is being done in this area. AI can fine tune the product catalog like never before – it can process the optimal price, content, size, validity and other parameters. It can also configure it based on analysis of competitor analysis and can use things such as customer feedback, BSS data etc. Following this , using AI to improve customer centricity will evolve as key area for investment/deployment and then looking at network optimisation and analytics. Early adopters from a CSP standpoint are Telefonica with Aura, and Orange with Djingo. 

From an M2M perspective, from the research i have conducted we are still in the mode of collecting data from machines/devices and manually -using that to predict faults etc but we are not yet using it to intelligently optimise/make decisions.  A good example of this was written about in a report by PWC which stated that a" GE jet engine collects 500Gb of data per flight, taking a 'snapshot' every second of over 5,000 parameters including air speed calibration, altitude, cooling, exhaust gas temperature and flow, and ground speed. This is in stark contrast to previous generations of jet engine technology, where just 1 kb of data was generated per flight from three snapshots (take-off, cruising, landing) on 30 parameters. The resulting insights enable GE to boost performance by 287 times while also delivering a seven-fold reduction in costs." - so the machine is thus recording data that engineers are using, machines are not optimising themselves. 

The convergence of IOT solutions and AI will really drive the need for AI in M2M and it is expected that investment will rise to about US $236 billion by 2020 (up from $72 billion). I will cause "dumb" devices to become more intelligent. 

So to conclude, we don't have the conclusive work/analysis done in this area, but it will certainly gather pace in the next 18+ months. 

I hope this was of help to you, and this is my perspective from an a market/analyst view. I don't know if anybody else has come across any other work? Would be great input to our next report if so.... 

Best wishes,

Aaron

------------------------------
Aaron Boasman
TM Forum
------------------------------

Original Message:
Sent: 07-06-2017 17:03
From: RAFI AHMED
Subject: AI in M2M

How much work is being done in this regard and what are the successful applications so far?

------------------------------
RAFI AHMED
Mobily
------------------------------

hmmmmm....AI has been around about 62 years. At least, that was when it was first nominated as an area of research/science/technology in its own right. If it is nascent it is taking a while to grow up...and doing so in fits and starts. Every so often there is a fit of enthusiasm for it. Normally associated with funding for it. Not surprising that the most apparent instantiations of "AI" you see are chatbots.  But I suspect that the most successful real applications of AI are not so prominent. Many of us learned in the 1990s that a good model for a real application of AI is to not tell anyone at all that you are using AI technology....just use it for competitive advantage. Let your competitors go and wonder why they just can't get their attempts at an algorithm for the application to work properly.

I don't think there has been any real "breakthrough" in AI of late. 

Understanding a problem space which actually needs AI is the beginning of successful AI application, I think. So where in the telco space do we really need AI? (I can think of a few places, but would like to see what others think.)

------------------------------
Peter Duxbury-Smith
Gigaclear

------------------------------

Original Message:
Sent: 07-06-2017 17:03
From: RAFI AHMED
Subject: AI in M2M

How much work is being done in this regard and what are the successful applications so far?

------------------------------
RAFI AHMED
Mobily
------------------------------

We see usage of bots in the operations area where keystroke level or L6 level process needs to be automated and that will drive efficiency and also OPEX reduction.
If we put an analytics engine below it can make intelligent decision making too.

------------------------------
Shyamtanu Banerjee
Wipro Technologies
------------------------------

Original Message:
Sent: 08-02-2017 06:31
From: Peter Duxbury-Smith
Subject: AI in M2M

hmmmmm....AI has been around about 62 years. At least, that was when it was first nominated as an area of research/science/technology in its own right. If it is nascent it is taking a while to grow up...and doing so in fits and starts. Every so often there is a fit of enthusiasm for it. Normally associated with funding for it. Not surprising that the most apparent instantiations of "AI" you see are chatbots.  But I suspect that the most successful real applications of AI are not so prominent. Many of us learned in the 1990s that a good model for a real application of AI is to not tell anyone at all that you are using AI technology....just use it for competitive advantage. Let your competitors go and wonder why they just can't get their attempts at an algorithm for the application to work properly.

I don't think there has been any real "breakthrough" in AI of late.

Understanding a problem space which actually needs AI is the beginning of successful AI application, I think. So where in the telco space do we really need AI? (I can think of a few places, but would like to see what others think.)

------------------------------
Peter Duxbury-Smith
Gigaclear


Original Message:
Sent: 07-06-2017 17:03
From: RAFI AHMED
Subject: AI in M2M

How much work is being done in this regard and what are the successful applications so far?

------------------------------
RAFI AHMED
Mobily
------------------------------

Hi

Picking up on Shyamtanu's point.

The 'bot API solution to AI seems well suited to answering the "narrow" problem.  In my experience the 'bots seemed to struggle once they had more than ten intents to choose from.  That would be in alignment with a level 6 process.

This is not to say there is not much value in applying bots. 
Firstly, ten intents can still allow for thousands of untrained or random inputs, which increases system flexibility by orders of magnitude.  This is incredibly valuable in a fast paced environment. 
Secondly, there is no reason the architecture should not be a multilayered set of 'bots.  In this case the ten intent limitation becomes a hundred with the addition of a second layer, or in a three layer solution it becomes a thousand intent limitation. 

Perhaps my suggested limit is due to implementation skills, but it is what I have seen to be a pragmatic scale that can be maintained on a continuous basis.

Enjoy!

------------------------------
Hugo Vaughan
Crowd Frame Consulting Limited.
------------------------------

Original Message:
Sent: 08-02-2017 06:47
From: Shyamtanu Banerjee
Subject: AI in M2M

We see usage of bots in the operations area where keystroke level or L6 level process needs to be automated and that will drive efficiency and also OPEX reduction.
If we put an analytics engine below it can make intelligent decision making too.

------------------------------
Shyamtanu Banerjee
Wipro Technologies

Original Message:
Sent: 08-02-2017 06:31
From: Peter Duxbury-Smith
Subject: AI in M2M

hmmmmm....AI has been around about 62 years. At least, that was when it was first nominated as an area of research/science/technology in its own right. If it is nascent it is taking a while to grow up...and doing so in fits and starts. Every so often there is a fit of enthusiasm for it. Normally associated with funding for it. Not surprising that the most apparent instantiations of "AI" you see are chatbots.  But I suspect that the most successful real applications of AI are not so prominent. Many of us learned in the 1990s that a good model for a real application of AI is to not tell anyone at all that you are using AI technology....just use it for competitive advantage. Let your competitors go and wonder why they just can't get their attempts at an algorithm for the application to work properly.

I don't think there has been any real "breakthrough" in AI of late.

Understanding a problem space which actually needs AI is the beginning of successful AI application, I think. So where in the telco space do we really need AI? (I can think of a few places, but would like to see what others think.)

------------------------------
Peter Duxbury-Smith
Gigaclear


Original Message:
Sent: 07-06-2017 17:03
From: RAFI AHMED
Subject: AI in M2M

How much work is being done in this regard and what are the successful applications so far?

------------------------------
RAFI AHMED
Mobily
------------------------------

Interesting, Hugo.

Would there be a comparison between the quantities of bot intents and what a human decision maker is capable of?

I recall studies of medical diagnosticians: Whereas medical syndromes typically involved 11 or more possible diagnoses, given presentation of a set of symptoms average diagnosticians could not cope with considering more than 5 or 6 as they proceeded through diagnosis.  The top specialists managed up to 9. 


------------------------------
Peter Duxbury-Smith
Intelligent Solutions Ltd

------------------------------

Original Message:
Sent: 08-03-2017 03:08
From: Hugo Vaughan
Subject: AI in M2M

Hi

Picking up on Shyamtanu's point.

The 'bot API solution to AI seems well suited to answering the "narrow" problem.  In my experience the 'bots seemed to struggle once they had more than ten intents to choose from.  That would be in alignment with a level 6 process.

This is not to say there is not much value in applying bots.
Firstly, ten intents can still allow for thousands of untrained or random inputs, which increases system flexibility by orders of magnitude.  This is incredibly valuable in a fast paced environment.
Secondly, there is no reason the architecture should not be a multilayered set of 'bots.  In this case the ten intent limitation becomes a hundred with the addition of a second layer, or in a three layer solution it becomes a thousand intent limitation.

Perhaps my suggested limit is due to implementation skills, but it is what I have seen to be a pragmatic scale that can be maintained on a continuous basis.

Enjoy!

------------------------------
Hugo Vaughan
Crowd Frame Consulting Limited.

Original Message:
Sent: 08-02-2017 06:47
From: Shyamtanu Banerjee
Subject: AI in M2M

We see usage of bots in the operations area where keystroke level or L6 level process needs to be automated and that will drive efficiency and also OPEX reduction.
If we put an analytics engine below it can make intelligent decision making too.

------------------------------
Shyamtanu Banerjee
Wipro Technologies

Original Message:
Sent: 08-02-2017 06:31
From: Peter Duxbury-Smith
Subject: AI in M2M

hmmmmm....AI has been around about 62 years. At least, that was when it was first nominated as an area of research/science/technology in its own right. If it is nascent it is taking a while to grow up...and doing so in fits and starts. Every so often there is a fit of enthusiasm for it. Normally associated with funding for it. Not surprising that the most apparent instantiations of "AI" you see are chatbots.  But I suspect that the most successful real applications of AI are not so prominent. Many of us learned in the 1990s that a good model for a real application of AI is to not tell anyone at all that you are using AI technology....just use it for competitive advantage. Let your competitors go and wonder why they just can't get their attempts at an algorithm for the application to work properly.

I don't think there has been any real "breakthrough" in AI of late.

Understanding a problem space which actually needs AI is the beginning of successful AI application, I think. So where in the telco space do we really need AI? (I can think of a few places, but would like to see what others think.)

------------------------------
Peter Duxbury-Smith
Gigaclear


Original Message:
Sent: 07-06-2017 17:03
From: RAFI AHMED
Subject: AI in M2M

How much work is being done in this regard and what are the successful applications so far?

------------------------------
RAFI AHMED
Mobily
------------------------------

Picking up on Shyamantu's observation: Are there particular organisations you observe doing this kind of application? Is it widespread practice or found with just a few pioneers?

Are there any specialists offering support to develop such applications?

------------------------------
Peter Duxbury-Smith
Gigaclear

------------------------------

Original Message:
Sent: 08-03-2017 04:15
From: Peter Duxbury-Smith
Subject: AI in M2M

Interesting, Hugo.

Would there be a comparison between the quantities of bot intents and what a human decision maker is capable of?

I recall studies of medical diagnosticians: Whereas medical syndromes typically involved 11 or more possible diagnoses, given presentation of a set of symptoms average diagnosticians could not cope with considering more than 5 or 6 as they proceeded through diagnosis.  The top specialists managed up to 9.


------------------------------
Peter Duxbury-Smith
Intelligent Solutions Ltd


Original Message:
Sent: 08-03-2017 03:08
From: Hugo Vaughan
Subject: AI in M2M

Hi

Picking up on Shyamtanu's point.

The 'bot API solution to AI seems well suited to answering the "narrow" problem.  In my experience the 'bots seemed to struggle once they had more than ten intents to choose from.  That would be in alignment with a level 6 process.

This is not to say there is not much value in applying bots.
Firstly, ten intents can still allow for thousands of untrained or random inputs, which increases system flexibility by orders of magnitude.  This is incredibly valuable in a fast paced environment.
Secondly, there is no reason the architecture should not be a multilayered set of 'bots.  In this case the ten intent limitation becomes a hundred with the addition of a second layer, or in a three layer solution it becomes a thousand intent limitation.

Perhaps my suggested limit is due to implementation skills, but it is what I have seen to be a pragmatic scale that can be maintained on a continuous basis.

Enjoy!

------------------------------
Hugo Vaughan
Crowd Frame Consulting Limited.

Original Message:
Sent: 08-02-2017 06:47
From: Shyamtanu Banerjee
Subject: AI in M2M

We see usage of bots in the operations area where keystroke level or L6 level process needs to be automated and that will drive efficiency and also OPEX reduction.
If we put an analytics engine below it can make intelligent decision making too.

------------------------------
Shyamtanu Banerjee
Wipro Technologies

Original Message:
Sent: 08-02-2017 06:31
From: Peter Duxbury-Smith
Subject: AI in M2M

hmmmmm....AI has been around about 62 years. At least, that was when it was first nominated as an area of research/science/technology in its own right. If it is nascent it is taking a while to grow up...and doing so in fits and starts. Every so often there is a fit of enthusiasm for it. Normally associated with funding for it. Not surprising that the most apparent instantiations of "AI" you see are chatbots.  But I suspect that the most successful real applications of AI are not so prominent. Many of us learned in the 1990s that a good model for a real application of AI is to not tell anyone at all that you are using AI technology....just use it for competitive advantage. Let your competitors go and wonder why they just can't get their attempts at an algorithm for the application to work properly.

I don't think there has been any real "breakthrough" in AI of late.

Understanding a problem space which actually needs AI is the beginning of successful AI application, I think. So where in the telco space do we really need AI? (I can think of a few places, but would like to see what others think.)

------------------------------
Peter Duxbury-Smith
Gigaclear


Original Message:
Sent: 07-06-2017 17:03
From: RAFI AHMED
Subject: AI in M2M

How much work is being done in this regard and what are the successful applications so far?

------------------------------
RAFI AHMED
Mobily
------------------------------