Testing AI Systems: Quality Characteristics and Cognitive Biases

Elena Treshcheva,
Researcher, Business Development Manager, Exactpro

AI Summit New York
11 December 2019

What recommendations for AI testing are in place and how well is the technology overseen? What abilities of AI systems should be tested? What challenges does the process pose from the software testing perspective? Exactpro's new white paper – Testing the Intelligence of Your AI – is now available at https://exactpro.com/ideas/white-papers/testing-intelligence-your-ai

Follow us on
LinkedIn https://www.linkedin.com/company/exactpro-systems-llc
Twitter https://twitter.com/exactpro
Facebook https://www.facebook.com/exactpro/
Instagram https://www.instagram.com/exactpro/

Subscribe to Exactpro Vimeo channel https://vimeo.com/exactproSubscribe to Exactpro YouTube channel https://www.youtube.com/c/exactprosystems

Transcript

1. Build Software to Test Software exactpro.com Testing AI Systems: Quality Characteristics and Cognitive Biases Elena Treshcheva, Researcher, Exactpro
2. Exactpro Overview • A specialist firm focused on functional and non-functional testing of exchanges, clearing houses, depositories, trade repositories and other financial market infrastructures. • Incorporated in 2009 with 10 people, our company has experienced significant growth and is now employing over 550 specialists. • We were part of the London Stock Exchange Group (LSEG) from May 2015 till January 2018. Exactpro management buyout from LSEG was successfully completed in January 2018. We are headquartered in the UK and have operations in the US, Georgia and Russia.
3. Exactpro Client Network
4. AI-based Systems’ Quality Characteristics: - Ability to learn: The capacity of the system to learn from use for the system itself, or data and events it is exposed to. - Ability to generalize: The ability of the system to apply to different and previously unseen scenarios. - Trustworthiness: The degree to which the system is trusted by stakeholders, for example a health diagnostic A4Q AI and Software Testing Foundation Syllabus https://www.gasq.org/en/exam-modules/a4q-ai-and-software-testing.html
5. Ability to Learn: https://www.deeplearning.ai/ • Training set — Which you run your learning algorithm on. • Development set — Which you use to tune parameters, select features, and make other decisions regarding the learning algorithm. Sometimes also called the hold-out cross validation set. • Test set — which you use to evaluate the performance of the algorithm, but not to make any decisions regarding what learning algorithm or parameters to use.
6. Trustworthiness: https://innovation.defense.gov/ai/ During the DIB’s quarterly public meeting on October 31, 2019, the DIB members voted to approve the proposed AI Principles.
7. Trustworthiness: https://www.mas.gov.sg/news/media-releases/2019/mas-partners-financial-industry -to-create-framework-for-responsible-use-of-ai
8. Trustworthiness: How can we persuade people to trust an algorithm? Some important techniques are: • Explainability • Testing • Boundary conditions • Gradual rollout • Auditing • Monitors and alarms https://blog.deeplearning.ai/blog/google-ai-explains-itself-neural-net-fights-bias-ai-demoralizes-champions-solar-power-heats-up
9. Ability to Generalize: Scope of End-to-End and Negative Testing
10. Congruence bias Confirmation bias Law of triviality Zero-risk bias Anthropocentric thinking Illusion of control Cognitive Biases Affecting Software Testing of AI-based Systems Automation bias
11. Mohanani, R., Salman, I., Turhan, B., Rodríguez, P., & Ralph, P. (2018). Cognitive Biases in Software Engineering: A Systematic Mapping Study. IEEE Transactions on Software Engineering.
12. Confirmation Bias
13. AI-based Systems: Machine-Readable News
14. Anthropocentric Bias: Why We Treat Robots Like Humans Darling, Kate and Nandy, Palash and Breazeal, Cynthia “Empathic Concern and the Effect of Stories in Human-Robot Interaction” (2015). Proceedings of the IEEE International Workshop on Robot and Human Communication (ROMAN), 2015. 6 p. https://www.ted.com/talks/kate_darling_why_we_ha ve_an_emotional_connection_to_robots
15. Anthropocentric Bias: Testing Chatbots Anaphora / Context Human: I bought 500 Company X shares two years ago. The stocks’ cost was 60,000 USD. What’s their today’s cost? Chatbot: What currency would you like to have for the rate? X Spelling / overall correctness Human: What is the settlement date of the tradeId XXX?? Chatbot: ???
16. AI-based Systems: Algo Trading
17. Congruence Bias Direct Testing Indirect Testing Indirect Testing
18. Applications of the Proposed Approach https://unsplash.com/search/photos/san-francisco The First IEEE International Conference on Artificial Intelligence Testing (IEEE AITest 2019), April 4-9 2019, San Francisco East Bay, CA, USA User-Assisted Log Analysis for Quality Control of Distributed Fintech Systems Iosif Itkin, Anna Gromova, Anton Sitnikov, Rostislav Yavorskiy, Evgenii Tsymbalov, Andrey Novikov and Kirill Rudakov.
19. Law of Triviality (the Bike-Shed Effect)
20. AI-based Systems: Pricing Calculator
21. Automation Bias
22. AI-based Systems: Fraud Detection and Market Surveillance
23. Build Software to Test Software Click to know more about Exactpro Test Tools
24. Zero-Risk Bias
25. Non-deterministic Systems: Financial Market Infrastructures
26. The Illusion of Control and Happiness Sherman, G. D., Lee, J. J., Cuddy, A. J. C., Renshon, J., Oveis, C., Gross, J. J., & Lerner, J. S. (2012). Leadership is associated with lower levels of stress. Proceedings of the National Academy of Sciences, 109(44), 17903–17907.
27. Fenton-O’Creevy, M., Nicholson, N., Soane, E., & Willman, P. (2003). “Trading on illusions: Unrealistic perceptions of control and trading performance”. Journal of Occupational and Organizational Psychology, 76(1), 53–68. The Illusion of Control and Performance
28. 28 Build Software to Test Software exactpro.com CONTACT: info@exactpro.com London: +44 (0) 203 319 1644 Thank you https://exactpro.com/ideas/white-papers/testing-intelligence-your-ai