How would AI shape Future Integrations?
1 like343 views
This document discusses how AI could shape future integrations. It begins by explaining different types of tasks that AI can perform, such as those that can be precisely explained versus those requiring examples and feedback to learn. The document then covers benefits of AI like speed, lower costs, and ability to learn and extrapolate. It discusses using AI for cost savings, competitive advantages, and new revenue streams through insights. Challenges of AI like lack of data and skilled professionals are presented along with risks such as bias, privacy issues, and how mistakes can be more harmful than for humans. Various use cases of AI in integration are explored such as enhancing inputs, security, and automatic integration. The document concludes that AI will create many new integration opportunities
How would AI shape Future Integrations?
- 1. How would AI shape Future Integrations? Srinath Perera, Ph.D. VP Research WSO2, Apache Member, ( [email protected]) @srinath_perera
- 5. !5 • Some tasks (e.g., make a coffee) we can explain, precisely. We can write a code to do them. • Other tasks, e.g., drive a car, we can do, but can’t explain how. We can’t teach our children how to drive by writing it down. Such “intelligent tasks,” we have to give them examples and feedback and train them. • AI can learn from examples to do such “intelligent tasks.” • Andrew Ng’s rule - “AI can do anything human can do in 10 sec or less.” What is AI?
- 6. What is the big deal? • Compared to humans, Computer doing these tasks (With AI) are • Faster (1K-1000K times) • Cheaper to replicate • Reliable • AI ability to learning can be used to extrapolate the future !6
- 7. Why AI in Integration?
- 8. Why? (1) Cost Savings !8 • For repeated tasks, AI is often much cheaper than Humans (e.g., video surveillance, quality control) • Having humans in the loop often leads to complex integrations because those interactions take time and because humans make mistakes • AI can pinpoint problems. Some times AI can predict and proactively fix problems. • Increase the usability of systems • Optimize systems
- 9. Why? (2) Competitive Advantage !9 • AI makes tasks that have been infeasible before now feasible (e.g., image or video- based inputs, personalized and targeted recommendations). • Furthermore, AI enables us to do tasks better and more frequently. (e.g., effective diagnoses, detecting anomalies). • AI can lead to better decisions and forecasts. This happens because AI makes better decisions or aids humans in understanding and exploring data. • AI-based solutions have predictable performance, while human performance vary.
- 10. !10 • An organization may share insights derived through AI with other parties. (e.g., Walmart shares data with suppliers). • Organization may creating new revenue streams by selling insights delivered through the data • e.g., a telco may share discoveries about local trends with customers (e.g., sports events). • e.g., Targeted Marketing ( market to demographic) • e.g., Credit card companies sells data about customer behavior Why? (3) New value and Revenue:
- 11. Feasibility !11 • AI is beating human performance in many cases • (e.g., Voice recognition, character recognition, Go) • Most tools are available as open- source projects • (e.g., TensorFlow, Torch, Sklearn) • Wide range of public datasets has enabled multiple parties to build on each other’s results • AI has Significant investments by internet companies, academia, and VCs.
- 12. Challenges !12 • Shortage of Skilled Professionals • The data scientists, programmers, and architects are in short supply and expensive (300-500K year). • It is hard for medium and smaller organizations to attract and hire enough skilled people. • Solutions ( Wizards, Mapping AI to DB (e.g., Bays DB) or spreadsheets, Automatic Statisticians) • Most AI solutions will be delivered as cloud APIs • Only possible when If data formats are well known (e.g., banking or healthcare data) and key performance indicators (KPIs) are well defined. Then reusable models can be built for those use cases. • Examples are disease detection in healthcare, marketing insights, spatiotemporal models, and analytics for specific sports.
- 13. Challenges (contd.) • Lack of Large Enough Data Sets. • With 10,000 data points per day, it takes 3 years to collect 1 million data points • Labeled data sets are hard to find • Solutions • Transfer learning • Semi & Unsupervised learning • Interpretability • Some models are black boxes • Tracking provenance of data used
- 14. AI mistakes are Dangerous !14 • AI weaknesses can cause much more potential harm than the same weakness in humans • Broad integration - AI will be everywhere (e.g., selecting people for an interview, picking suspicious people for further investigation). • Wide application - because the cost to replicate AI operations is small, AI is applied on a broader scale than humans • Transparent application without oversight - diversity of views among humans detect and control flowed applications, while AI will be applied transparently avoid such oversight. CC https://www.maxpixel.net/House-Of-Cards-Fragile-Sensitive- Statics-Patience-763246
- 15. Risks • Bias • AI can learn and repeat the inherent bias in the data caused by human behavior (e.g., Book “Weapons of Math Destruction” by Cathy O’Neil.) • Removing bias is hard - For example, an address in a certain neighborhood might act as a proxy for the race; the name might act as a proxy for gender, or name of the degree might act as a proxy for age. • AI tug war • If the use case ends up being a fight between two AI algorithms, it can lead to lose-lose situations !15
- 16. Risks (contd.) • Privacy: AI can infer sensitive data from seemingly harmless information • E.g., detect where you live using accelerometer data • Detect what you watch on the TV using electricity data • Your travel history by automatic number plates reading • GOV policies • Data collections rules limiting data collection • Laws limiting where AI can be used (e.g., SF limiting face recognition) !16
- 17. Macro Trends • Data Moats • AI depends on data. Hence data can be a significant advantage. Also, products that get market adoption first will get more data, creating data network effects • In some cases, data network effects do not work due to Diminishing returns of adding data • This gives rise to AI offered as Cloud APIs • Let vendors collect data • Concentrate both expertise and data • remove the complexity of managing AI models • AI provides a significant opportunity for large companies such as telecom, credit card, and retail sectors, which already have access to large and rich data sets. !17
- 18. AI Use cases
- 19. Use Cases1: Integrations required to enable AI • Most AI use cases in organizations will have significant Integration component • AI requires organizations to collect data from many sources, reconcile their differences, and manage them. • Carrying out actions suggested by AI • Exposing data and insights via APIs !19
- 20. Use Cases 2: Enhancing Inputs • GUIs, CLIs, computer languages are not natural for humans • AI enables us to ingest audio, video, natural language that is a lot more natural to humans • e.g., object detection, NLP, chatbots • Follow a football game just through video • Extracting information from twitter !20
- 21. Use Cases 3: AI support within existing data stores • Vendors that build data storages (e.g., CRM, ERP, databases) can build AI as turnkey solutions into their products • e.g. Salesforce Einstein • Users get AI without additional investments or coding for predictable cost • User will benefit from the data from all customers !21
- 22. Use Cases 4: Self Driving System Operations • Use of AI to reduce or eliminate the need for human intervention while managing (integration) systems • Examples: • Auto tuning • Predictive maintenance • Detect anomalies, trigger deep observations, rise alerts, help analysis and root cause • Chat-based control • Automatic compliance and QoS checks !22
- 23. • Use of AI to simplify the process of creating integrations support both programmers and non- programmers • Examples: • Let user draw the integration visually and use AI automate mapping with users having to fill configurations forms • Let user describes the integration in English, and a chatbot ask more questions to clarify and refine the integration Use Cases 5: Automatic and Self-Service Integration !23
- 24. Use cases 6: AI-based Security • Examples • Fraud Detection • Adaptive Authentication • Detection Session Hijacking • Detect sensitive data in messages and hold the message • Continuous authentication • AI can also be used to attack well as defence, and since it is hard to defend than attack, outcomes are unclear !24
- 25. Assessment of use cases !25
- 26. !26 Conclusion • Enabling AI creates many integration use cases • AI mistakes are more harmful than human mistakes due to: broad integration, cheap replication, and transparent application without any supervision. • Most AI use cases will be delivered as cloud APIs due to • Data provides competitive advantage and cloud enables vendors to collect data • Most custom AI models will be limited to large organizations due to a lack of expertise and data • Cloud enables organizations to concentrate expertise and simplify deployment • Use cases • Security, Data Integration, Enhancing Inputs, and AI support within existing data stores, already have systems available and are likely find more adoption. • Increasing Usability, Self Driving Operations (DevOps and Regulatory Compliance), Automatic and Self-Service Integration do not face critical challenges but will need 5-10 years of development for wide adoption • Business Automation will often be infeasible for SMEs if not already supported as a cloud API.
- 27. Analysis based on ETAC !27 This is analysis of blockchain for integration, based on the Emerging Technology Analysis (ETAC) framework
- 28. To receive updates to ETAC and ETAC-based emerging technology analysis, subscribe to our Newsletter. Thanks for reading!!