BPM.com
  1. Peter Schooff
  2. Sherlock Holmes
  3. BPM Discussions
  4. Tuesday, 13 June 2017
  5.  Subscribe via email
As Connie Moore writes in this blog, "The hype surrounding artificial intelligence (AI) in today’s marketplace cannot be overstated. Most of the excitement is just that—hype and hot air, plus a whole lot of confusion about what AI actually encompasses." So in terms of Artificial Intelligence's impact on BPM, what do you think?
Accepted Answer Pending Moderation
I think BPM may be one of the areas where the hype about AI might actually be understated.
It seems perfectly rational that the Process Manager could "learn" from processes as they execute - the initial process definition might come from Process Mining, or might be built by Humans using BPMN - but from that point on machine learning would tweak the process to improve performance.
Founder at John Reynolds' Venture LLC - Creator of ¿?Trules™ for drama free decisions
Comment
Process Mining does not produce working processes unless you already have a procces in place. The machine learning has to happen from user action in the process environment. Hence your orthodox BPM system is suddenly obsolete ... and so is BPMN. They guiding and constraining of work for compliance has to happen in business language rules and not flow diagrams.
Max I think you are quite wrong if anything BPM has significant contribution to make in bring order and structure to AI. BPM has no limitations in thinking which will be recognised as AI evolves with new software capabilities. Yes old supporting software such as BPMN will not be up to the job....
  1. Emiel Kelly
  2. 1 year ago
  3. #4082
Since when is BPMN software? ;-)
  1. John Morris
  2. 1 year ago
  3. #4083
Re: "process mining does not produce working processes unless you already have a process in place" -- that's not saying much, because as long as you have date stamps and some sort of unique ID (e.g. an order form), which could even be acquired in a paper-based environment, you can mine the process and create a nice graph. What business doesn't have at least an implicit process? The power of process mining is how much you can achieve from "tools you probably have around the house" already . . . . : )
  1. more than a month ago
  2. BPM Discussions
  3. # 1
Accepted Answer Pending Moderation
Walk before you run!

With many BPM apps locked in time, the logical sequence is

a). Move to ACM/BPM where run time experiences are made up of a series of process fragments plus ad hoc interventions, threaded together by users, software, machines such that you really do NOT have a "process" until you close a Case (all, in theory, will be different).

b) Put in place the ability to take all of the fragments plus the ad hoc interventions at a Case and map these out on a graphic canvas (with time as your horizontal dimension).

c) Use data mining to overlay "similar cases" and apply high-level filtering so you can identify such activity as ". . . .most of the time, they went that way"

d) Get your environment to auto-add branching decision boxes where there were none based on the data mining statistics (no rules needed for these new branching decision boxes).

d) Now background new Cases with probabilities at decision branching boxes that post to increase user decision-support (sub-path #1: 40%; sub-path #2: 20%; sub-path #3: 10%; sub-path #4 30%).

Sit back and track how your now "auto-process-improvement" environment augments your Case experiences, then, perhaps (2-3 years to get the above working and settled in), blend in some AI.
Comment
Would work but is very complex, but our User-Trained Agent learns from the users in real-time.
  1. more than a month ago
  2. BPM Discussions
  3. # 2
Accepted Answer Pending Moderation
I refer you to @neilwd's newsletter of yesterday morning, talking about McKinsey's article on Burned by the bots: Why robotic automation is stumbling and the following relevance to today's QOTD:

"Among other blinding insights: you need to think properly about how you’re going to manage the change to business activities, the bots themselves and any changes that might occur over time in back-end systems.

It’s a bit like advising an organisation buying a vehicle fleet to remember to hire qualified drivers, and make sure they get the fleet serviced at regular intervals.

Who would have guessed…?"

The same is, would be true of intelligence. Machine learning and AI aren't just going to automagically enhance BPM to some utopian state without some up-front input and guidance. The up-front homework is the part that always gets left out on technical panaceas.

Just my tuppence.
Comment
Yes, buying AI software and trying to do some process management with it will be a disaster. Chatbots are just stupid. But a machine learning component that learns how to execute a process from the expert user replaces also the analysis phase and does not need a optimization cycle. We have been doing it for years and now we are suddenly cool ....
Nice to see there's at least one person who reads our newsletters Mr L! Thanks!
  1. more than a month ago
  2. BPM Discussions
  3. # 3
Accepted Answer Pending Moderation
AI will have no impact on BPM! There is no such thing. But MACHINE LEARNING will make all flowchart design obsolete. Our User Trained Agent makes task recommendations to users since 2009. It builds a complete goal-oriented use-case in a value stream through user interaction.

Up till now most CIOs were really worried about it and did not dare to use it. Suddenly they can be all cool and do 'AI' with what they have installed.

BPM flowcharting was a waste of time before and now it is simply obsolete, GREAT!!!
References
  1. Http://isis-papyrus.com
  2. https://isismjpucher.wordpress.com/2016/12/15/artificial-intelligence-and-bpm/
Comment
  1. more than a month ago
  2. BPM Discussions
  3. # 4
Accepted Answer Pending Moderation
Great comments and original item on AI hype. As we learned in the 80's and with expert systems, which was the AI of the time, it's very difficult to do, including especially modeling any given domain. Fast forward 30 years and we have only now built a machine that can win at the game of Go -- and real life is how many orders of magnitude more complex? AI is useful -- if you can actually codify some model (possibly ignoring the tacit). But note that such models are only useful for narrow distributions of outcomes -- but will be useless for the outliers or "black swans" (see Nassim Taleb) found in the larger fat-tailed worlds where most business lives. Today's AI is still difficult (i.e. expensive) to do for non-trivial things.

However, given that BPM is very much about microeconomics-driven process repetition patterns, i.e. characterized by narrow distributions, AI is especially suitable as a complement to BPM technology. The narrow scope of business automation where BPM thrives also defines a narrow scope of data and behaviour where AI is more technically feasible.
Comment
  1. more than a month ago
  2. BPM Discussions
  3. # 5
Accepted Answer Pending Moderation
Connie gets a +1 from me for “a whole lot of confusion about what AI actually encompasses”. Back when I was at school, AI meant a bunch of nerds doing neural nets in LISP, in a “temporary” (it's still there, over 30 years later) trailer parked in the shadow of the computer science building. Today, my son (admittedly, also a nerd) is specializing in AI, and so I've gotten a glimpse of just how rich and complex that field has become. Here's an actual slide from his ML class:

https://www.bplogix.com/wp-content/uploads/2017/06/ML-slide-26pct.jpg
Yikes: good thing I already got my degree.

ML and AI have implications anywhere there are decisions to be made. But is AI intrinsic to BPM? Or is it a service, like any other, that informs BPM behaviors? Does it make sense for BPM vendors to reinvent AI/ML code for their platforms, or would it be more reasonable to be able to plug into technology from different vendors depending on the nature of the problem being addressed?

I guess we'll all find out soon enough.
http://www.bplogix.com/images/icon-x-medium.png
-Scott
Comment
Or our users could just train their smart agents and the angels' choir would rejoice. ;)
  1. more than a month ago
  2. BPM Discussions
  3. # 6
Accepted Answer Pending Moderation
As E Scott points out, it seems of utter importance to define artificial intelligence in the context of BPM first. AI seems to have the potential to apply to different aspects in BPM, in different ways. The impact of AI on "process flow charting" that Max stresses, seems very interesting. Within the process innards, predictive analytics, driven by lineal or multiple regressions of statistical relevant process data sets (5+ years of complete, monthly observations for instance) can lead to pattern recognition, which then can support an AI as part of a BRE, being part of a iBPMS.
In general, I do think AI will impact BPM greatly, starting with business rules. It's just very depended on the user BPM maturity level with current technologies (ref. Karl's post, outlining a possible road map).
NSI Soluciones - ABPMP PTY
Comment
  1. more than a month ago
  2. BPM Discussions
  3. # 7
Accepted Answer Pending Moderation
To me AI/ML is about turning data into insights which hopefully leads to action.

Like a self driving car. With all of it sensors, it is collecting lots of data (which in itself don't mean anything). Next, with algorithms and computing power, it turns this into insight (too fast, not in the right lane, cars in front of me are slowing down, etc). And luckily the software in the car can turn these insight into action (turn left, brake, etc)

I look the same at the application of AI/ML in processes (or better "managing cases" ).

During the execution of work for a case, a lot of data can be collected. In itself this data doesn't mean anything, but with some algorithms this could become insight in "how work is done here" or some indicators of, what we think shows, the performance of a process. Process Mining is an example of that. But I agree with Max that is still "insight staring". It should be about turning that insight into action.

That could be immediate action (do this for this case!) or data for more traditional process improvement. Which might be done by a "machine" that changes some process parameters.

So AI/ML will definitely have it's impact on processes. Not the same for any type of process, but where data must be turned into action (actually, isn't that in any process?), it hopefully helps to create better insights and faster action.

But to me it's all about the action. Because a self driving car can be a masterpiece of sensors and algorithms, but is useless without a throttle or a brake.

And wheels that can turn.
Sharing my adventures in Process World via Procesje.nl
Comment
  1. more than a month ago
  2. BPM Discussions
  3. # 8
Accepted Answer Pending Moderation
Yes as ever over hyped on impact. It will be an evolutionary experience aiding business efficiency but remember people are at all angles of this move from "build" of AI to the managing and using outcomes. Economies need to see efficiency improvements to prosper and AI will have a role in this. BPM has a role to ensure compliance and accountability. As previously discussed in this forum if such discipline is not there more VW scandals will happen.

In terms of the intelligence gained from data creation in processes that journey started where decisions can be automated to enrich the user experiences and become more productive. AI needs the outside in approach as it becomes incorporated into operational business processes as such the BPM discipline will expand as such capabilities evolve and grow.
Comment
  1. more than a month ago
  2. BPM Discussions
  3. # 9
Accepted Answer Pending Moderation
By the way, AI/ML should likely be paired with decision making, rules repositories, DMN etc. etc. And the reason is that the only use for AI is ultimately "better decisions".

So, in that corner of decisioning which is amendable to rules, AI/ML will likely work better when integrated into such a comprehensive rules-based decision management environment. And of course, good DMN decision technology-based solutions are already available for integration with BPM deployments.

One can make an argument that the rules-oriented corner of AI/ML is the most important application of AI/ML. The alternative idea that big data, AI and ML can be at the heart of discovering something new, i.e. to drive a new strategy or breakthrough product development for example, cannot reliably be the basis for many sound departmental business plans.
Comment
@John.. agree that AI/ML is not likely to see its best use "discovering something new". We have crystal balls for this.

As for the use of AI/ML, I see it's main value as providing "better suggestions", not "better decision-making".

The whole idea of ACM/BPM is we have best practices, here they are, we know that consistent use of best practices gives better outcomes but, go ahead, based on experience, situation/context, judgment, logic, intuition, . . . and deviate from the best practices.

BPM will provide orchestration, Case-level rules will provide governance (e.g. remember the model of the centerline on the highway, guardrails on both sides).

No different IMO with AI/ML - you get to a branching decision box where you have rules that engage certain processing based on algorithms (upstream insurance checkbox was checked/was not checked ) or based on user choices (able to take this on, not able to take this on) and the ideal contribution of AI/ML is to post stats that say (40% of the time prior cases went this way, 60% of the time prior cases went that way)

Summary - we already have decisions that can be automated, we already have ways to constrain "arbitrary" choices via governance rule sets, and what AI/M adds is not decisions but "decision support" i.e. suggestions.

In the world of inputs/outputs, you cannot rely on logic connections to guarantee that it is OK to perform the next-in-line task (i.e. there may not have been a context/situation relevant predecessor).

Nor can we rely on the processing itself between Input and Output (i.e. the designer may have failed to take into consideration some aspect essential to a solution, so, this is why I keep pushing for pre-processors and post-processors. (are you sure its ok to engage? are you sure it's ok to go away with this result?).


  1. John Morris
  2. 1 year ago
  3. #4085
I think we are in agreement @Karl. "Suggestions" are a way of loosening up BPM guardrails to account for more of the richness or variability (or chaos and complexity) of real business processes. I see suggestions therefore as just part of a sensible 'decision process' -- not that AI/ML will necessarily dictate a given branch or case sub-process.
@John

LOL, I suspect that before we hand over decision making to robots/AI, we have to get to where scenarios like the following can be dealt with.

I remember a boss who gave me some instructions and then came back the next day with opposite instructions.

I reminded him that yesterday he wanted that and that today, it seems, he wants this.

The response was " . . . .yesterday was yesterday, today is today"

Most rule sets are likely to have trouble with this.

2nd example . . .

At uni, we had a professor who did not have much patience for idiotic dialogue.

One student asked whether the questions on the final exam would be the same as last year.

Response: Yes, same questions, different answers.
  1. more than a month ago
  2. BPM Discussions
  3. # 10
Accepted Answer Pending Moderation
It seems that AI/ML is just yet another coordination technique - flow-chart, business rules, event-based, queue-based and, finally, intelligence-based. Sure, this is a very important coordination technique, because there are models and associated data, and, we want AI/ML to tell us how to change our models to get desired outcomes from them.

Thanks,
AS
Comment
  1. more than a month ago
  2. BPM Discussions
  3. # 11
Accepted Answer Pending Moderation
Comment
  1. more than a month ago
  2. BPM Discussions
  3. # 12
Accepted Answer Pending Moderation
I'd love to get to pick the users that train those agents :p (and see some agents self-shutdown due to desperation!)

And I would love to see an actual audit of an AI-driven-decision-gone-wrong :D Who's guilty: the developer, the integrator, the trainer, the operator, the data feeder?

The most hypocritical discourse however is the one where AI is replacing tedious human jobs so people can focus on more fulfilling, higher-IQ jobs. Since the transition from man to machine will happen in months, I wonder if those people "relieved" from the "basic, undignified" jobs will also get quickly equipped with the incremental IQ required not to do the task, but to operate the task-doer.

You see this [NOT] happening in RPA: all those people inputting invoices are being replaced by robots and suddenly move into Analyst jobs, which were just waiting for them. / sarcasm
CEO, Co-founder, Profluo
Comment
  1. Emiel Kelly
  2. 1 year ago
  3. #4087
Drew the same conclusion here. http://procesje.nl/wordpress/?p=419 You might need a translation AI-robot to understand ;-)
Oh I tried AI translation before, on your tweets - we know how THAT went LOL
  1. John Morris
  2. 1 year ago
  3. #4092
Bravo your comment @Bogdan. And just for the record I don't see your comments in any way in conflict with my notes above. I said that AI/ML is nicely complementary to BPM given the fact that BPM is necessarily (due to economics) about repetitive workflows. (I also noted that AI is typically much more difficult to do than hype would have it, especially due to the world of the tacit.)
So, a business case can be made for AI/ML in the context of some workflows.
At the same time, you are completely correct about the social implications and the fantasy of retraining -- and the condescension too -- concerning so-called "basic, undignified jobs" as you say. I keep harping on "the tacit": the tacit is a proxy for all the things that staff (how about we say "workers") know in order to do any kind of task -- and tacit is present even in things that seem highly structured and ideal for automation and AI, such as RPA.
There's probably some sort of "U-shaped" curve that maps semantic content of work over time, against automation. In the beginning, humans do everything and the semantic content is high. Then with automation, humans are removed from the process, that tacit is forgotten, and semantic content falls (along with costs). The futurist hope is that after a while with ever-more powerful AI, that the semantic content will begin to rise again.
Big topic. Lots of implications.
  1. more than a month ago
  2. BPM Discussions
  3. # 13
Accepted Answer Pending Moderation
Link below just published by Computer Weekly to add to thoughts

Majority of Brits would use artificial intelligence, survey finds
The level of trust for using artificial intelligence for customer services has increased among British people, although they may have little choice as businesses flock towards the technology.
References
  1. http://ed-link.techtarget.com/r/JE0S18F/QP1845/X4GYVL/ONRQN9/NRYWLI/85/h?a=20465073
Comment
  1. more than a month ago
  2. BPM Discussions
  3. # 14
  • Page :
  • 1


There are no replies made for this post yet.
However, you are not allowed to reply to this post.