Man or Machine: Hybrid Intelligence - Wil van der Aalst
0 likes10 views
Process mining is not a magic wand. It is a discipline that requires a smart human being who can make the connection between the data and the underlying business process — with the help of the process mining tool. You have to apply your domain knowledge to interpret the results and develop improvement ideas. Pundits love to tell us which professions will be replaced by robots any day now. All the while, the smart devices in our pockets and at home are not as smart as we would like them to be. Where exactly lies the boundary between what machines can do, and what will humans continue to do well into the future? In his closing keynote, Wil explored the questions of where technology outperforms humans, where human intelligence is needed, and where both blend together to combine the best of both worlds.
Man or Machine: Hybrid Intelligence - Wil van der Aalst
- 1. © Wil van der Aalst (use only with permission & acknowledgements) prof.dr.ir. Wil van der Aalst RWTH Aachen University W: vdaalst.com T:@wvdaalst Man or Machine: Hybrid Intelligence
- 2. © Wil van der Aalst (use only with permission & acknowledgements) How can Artificial Intelligence (AI) & Machine Learning (ML) improve processes? Automate & improve individual tasks (link to Robotic Process Automation) Automate & improve process management (link to Process Mining) Combine data, algorithms, and people (link to Hybrid Intelligence) Focus on processes and IT, not on new products, materials, machines, etc.
- 3. © Wil van der Aalst (use only with permission & acknowledgments) Machine Learning = Learning based on examples without being programmed. Supervised and Unsupervised Neural networks: First a failure now the dominating paradigm Focus on specific tasks (speech, images, etc.). input output 0.2345 0.7765 0.7745 0.8845 0.2342 0.6545 0.1895 0.765 0.2345 0.2667 0.9845 0.5653 0.4455 0.9943 0.2247 0.2345 0.7765 0.7745 0.8845 0.2342 0.6545 0.1895 0.765 0.2345 0.2667 0.9845 0.5653 0.4455 0.9943 0.2247 0.2345 0.7765 0.7745 0.8845 0.2342 0.6545 0.1895 0.765 0.2345 0.2667 0.9845 0.5653 0.4455 0.9943 0.2247 0.7765 0.7745 0.2345 0.2667 0.5653 0.2247 0.7765 0.7745 0.2345 0.2667 0.5653 0.2247
- 4. © Wil van der Aalst (use only with permission & acknowledgments) How about managing and improving operational processes? We need process models that are understandable! We do not have labeled data, we have SAP, Salesforce, Oracle, Microsoft, Infor, etc. (holding thousands of tables)! We are interested in improving end-to-end performance and compliance (not a single task)! Process mining is 20% of analytics that drive 80% of the ROI (Dave Langer) Neural networks are 20% of analytics that drive 110% of the hype (Marco Pegoraro)
- 5. © Wil van der Aalst (use only with permission & acknowledgements) It is difficult to make predictions, especially about the future Niels Bohr (1971)
- 6. © Wil van der Aalst (use only with permission & acknowledgements) Theodore Gordon and Olaf Helmer- Hirschberg, Report on a Long-Range Forecasting Study, RAND Corporation, Santa Monica, CA, RP-2982, 1964 Predictions in 1964 about 1984: • “A permanent lunar base will have been established” • “Manned Mars and Venus fly-bys will have been accomplished” • “Automation will span the gamut from service operations to decision making at the management level” (For 2020: Breading of apes for low-grade labor and ground combat.) 1964
- 7. © Wil van der Aalst (use only with permission & acknowledgements) Tim Cross/Tom Gauld (The Economist, June 2020)
- 8. © Wil van der Aalst (use only with permission & acknowledgements) Our home • 9 Alexa-capable devices 1x Echo plus, 2x Echo show, 2x Echo, 2x Echo dot, 1x Fire TV, 1x Fire TV Cube (not counting laptops) • 12 Siri-capable devices 4x iPhone, 8x iPad, 1x watch controlling 10+ Philips hue lights, 2 Blink cameras, 2 Kasa switches, etc.
- 9. © Wil van der Aalst (use only with permission & acknowledgements) “… we are probably only a month away from having autonomous driving at least for highways and for relatively simple roads. My guess for when we will have full autonomy is approximately three years.” (Elon Musk, 2015)
- 10. © Wil van der Aalst (use only with permission & acknowledgements) “… People should stop training radiologists now. It’s just completely obvious that within 5 years, deep learning is going to do better than radiologists …. We’ve got plenty of radiologists already” (Geoffrey Hinton, 2016)
- 11. © Wil van der Aalst (use only with permission & acknowledgments) © Wil van der Aalst (use only with permission & acknowledgments) Human Intelligence Machine Intelligence Hybrid Intelligence fast efficient cheap scalable consistent flexible creative emphatic instinctive commonsensical data and algorithms people and experiences
- 12. © Wil van der Aalst (use only with permission & acknowledgments) © Wil van der Aalst (use only with permission & acknowledgments) human machine
- 13. © Wil van der Aalst (use only with permission & acknowledgments) © Wil van der Aalst (use only with permission & acknowledgments) human machine
- 14. © Wil van der Aalst (use only with permission & acknowledgments) © Wil van der Aalst (use only with permission & acknowledgments) human machine Key question: how to distribute the work? RPA, ML, AI, PM, etc. revive this old question
- 15. © Wil van der Aalst (use only with permission & acknowledgements) PwC report by John Hawksworth, Richard Berriman and Saloni Goel, 2018
- 16. © Wil van der Aalst (use only with permission & acknowledgements) “AI will power 95% of customer interactions by 2025” (Servion)
- 17. © Wil van der Aalst (use only with permission & acknowledgements) The Future of Jobs Report 2020 by the World Economic Forum
- 18. © Wil van der Aalst (use only with permission & acknowledgements) How about process automation and management?
- 19. © Wil van der Aalst (use only with permission & acknowledgements) SCOOP 1977 (Michael Zisman, Wharton, Philadelphia) OfficeTalk 1978 (Skip Ellis, Xerox, Palo Alto) Domino 1984 (Thomas Kreifelts, GMD, Sankt Augustin) start register request examine thoroughly examine casually check ticket decide pay compensation reject request reinitiate request end c1 c2 c3 c4 c5 Skip Ellis first African- American to earn a Ph.D. in Computer Science (1969) The origins of WFM/BPM Did it work?
- 20. © Wil van der Aalst (use only with permission & acknowledgements) reality check ahead This is what I believed in 1995:
- 21. © Wil van der Aalst (use only with permission & acknowledgements) Very difficult to get simulation/BPMN models to behave as real processes. Most workflow projects failed. Too expensive. Not trustworthy, surreal, … Models seen as wallpaper.
- 22. © Wil van der Aalst (use only with permission & acknowledgements) This was not in my model …
- 23. © Wil van der Aalst (use only with permission & acknowledgements) classic BPM human behavior data-driven systems actual/effective process improvements O2C, P2P processes having hundreds of thousands of variants. Information systems having thousands of tables with complex dependencies recording what people really do. Transforming insights into actions (not models or PowerPoints). What did we miss?
- 24. © Wil van der Aalst (use only with permission & acknowledgements) Miguel Valdes CEO and co-founder of BonitaSoft at BPM 2017 † Gartner BPM summits (RIP)
- 25. © Wil van der Aalst (use only with permission & acknowledgements) Process Mining (initiated from academia) Robotic Process Automation (initiated from industry) Formalisms Modeling WFMS/BPMS Data as evidence Actionable insights Augmenting systems shift in focus
- 26. © Wil van der Aalst (use only with permission & acknowledgements) You know about process mining …
- 27. © Wil van der Aalst (use only with permission & acknowledgements) 80 / 20
- 28. © Wil van der Aalst (use only with permission & acknowledgements) • 80% of the cases are described by 20% of the variants. • 80% of the cases cause only 20% of the friction (rework, complaints, etc.). • The remaining 20% of the cases account for 80% of the variants. • The remaining 20% of the cases account for 80% the friction (rework, complaints, etc.). P2P: 100% P2P: 80%
- 29. © Wil van der Aalst (use only with permission & acknowledgements) 20% of cases is causing 80% of the friction! Rework Delays Deviations Ping-pong Loops Fraud Lost cases Unresponsiveness
- 30. © Wil van der Aalst (use only with permission & acknowledgments) research commercial tools adoption
- 31. © Wil van der Aalst (use only with permission & acknowledgments) discover align replay enrich apply compare information systems extract process models explore select filter clean conformance performance diagnostics predictions improvements transform act show model adapt show interpret drill down ML + + event data
- 32. © Wil van der Aalst (use only with permission & acknowledgments) discover align replay enrich apply compare information systems extract process models explore select filter clean conformance performance diagnostics predictions improvements transform act show model adapt show interpret drill down ML + + event data How long will this case take? Will this case deviate? Will this case be rejected? What is the next activity? Will there be a bottleneck tomorrow? Should I accept new cases? Should I reallocate people? Etc.
- 33. © Wil van der Aalst (use only with permission & acknowledgments) discover align replay enrich apply compare information systems extract process models explore select filter clean conformance performance diagnostics predictions improvements transform act show model adapt show interpret drill down ML + + event data Warning: Do not start here! Start here!
- 34. © Wil van der Aalst (use only with permission & acknowledgements) More on RPA
- 35. © Wil van der Aalst (use only with permission & acknowledgements) Robotic Process Automation • Unlike PM initiated from industry • Keep original system • Bottom-up / quick wins • Complements process mining - automate versus analyze - task level versus process level
- 36. © Wil van der Aalst (use only with permission & acknowledgments) RPA: The Poor man’s WFM system user interface database system application database system database system application application application user interface user interface application application application user interface user interface database system application database system database system application application application user interface user interface application application application user interface Humans are often the glue between applications
- 37. © Wil van der Aalst (use only with permission & acknowledgements) process variants sorted in frequency frequency traditional automation candidates for RPA (traditional automation is not cost effective) low-frequent process variants that cannot be automated and still require human involvement process mining is able to diagnose the full process spectrum from high-frequent to low-frequent and from automated to manual RPA shifts the boundary of cost-effective automation
- 38. © Wil van der Aalst (use only with permission & acknowledgments) Process Mining for RPA open file copy paste type type copy paste aggregate correlate
- 39. © Wil van der Aalst (use only with permission & acknowledgments) Example Hybrid Intelligence aiconomix.com
- 40. © Wil van der Aalst (use only with permission & acknowledgments) Interplay Between Process Mining and RPA process mining process mining Analyze interactions rather than backend systems
- 41. © Wil van der Aalst (use only with permission & acknowledgments) Interplay Between Process Mining and RPA process mining process mining Replace people by robots for selected tasks
- 42. © Wil van der Aalst (use only with permission & acknowledgments) Interplay Between Process Mining and RPA process mining process mining People can hand-off tasks to robots (like an autocomplete). Robots can call for human assistance and detect drifts.
- 43. © Wil van der Aalst (use only with permission & acknowledgements) Conclusion
- 44. © Wil van der Aalst (use only with permission & acknowledgments) © Wil van der Aalst (use only with permission & acknowledgments) Human Intelligence Machine Intelligence Hybrid Intelligence fast efficient cheap scalable consistent flexible creative emphatic instinctive commonsensical data and algorithms people and experiences
- 45. © Wil van der Aalst (use only with permission & acknowledgements) Process Mining (initiated from academia) Robotic Process Automation (initiated from industry) Formalisms Modeling WFMS/BPMS Data as evidence Actionable insights Augmenting systems shift in focus
- 46. © Wil van der Aalst (use only with permission & acknowledgements) BPM+ PM+RPA human behavior data-driven systems actual/effective process improvements O2C, P2P processes having hundreds of thousands of variants. Information systems having thousands of tables with complex dependencies recording what people really do. Transforming insights into actions (not models or PowerPoints). Addressing the problems of classical BPM
- 47. © Wil van der Aalst (use only with permission & acknowledgements) W.M.P. van der Aalst. Process mining and RPA. In Robotic Process Automation: Management, Technology, Applications, pages 223-242. De Gruyter, 2021. W.M.P. van der Aalst. Hybrid Intelligence: To Automate or Not to Automate, That is the Question. International Journal of Information Systems and Project Management, 10(2), 2021. Learn more? W: vdaalst.com T:@wvdaalst