AI Sensors and Dashboards: Gauging and Monitoring the Inferences Capabilities of AI
1. AI Sensors and Dashboards:
Gauging and Monitoring the
Inference Capabilities of AI
Huber Flores
Associate Professor, University of Tartu
E-mail: huber.flores@ut.ee
Tartu, Estonia, Data Science Seminar – March 11, 2024
This research is part of SPATIAL project that has received funding from the European Union's
Horizon 2020 research and innovation programme under grant agreement No.101021808.
2. ● Not only refinement of algorithms
● Better tools to construct machine learning models
(TensorFlow, FedN, Flower)
● Increased processing capabilities
(Distributed systems) – Geoffrey Hinton
Machine and deep learning (aka AI)
3. ● Not only refinement of algorithms
● Better tools to construct machine learning models
(TensorFlow, FedN, Flower)
● Increased processing capabilities
(Distributed systems) – Geoffrey Hinton
● Classic verification methods are not
applicable for AI models
Machine and deep learning (aka AI)
[source] https://futureoflife.org/open-letter/pause-giant-ai-experiments/,
Accessed March 2024
One year ago!
4. ● Trustworthy computing defines a set of trustworthy
properties required for computing programs to be
considered trustful.
○ Secure
○ Reliability
○ Available
○ Private
○ Resilience
○ Transparency…
Trustworty computing
[source] https://www.heise.de/hintergrund/Microsofts-
Kampf-gegen-Bugs-1429333.html , Accessed March 2024
Bill gates internal memo
5. ● Trustworthy computing defines a set of trustworthy
properties required for computing programs to be
considered trustful.
○ Secure
○ Reliability
○ Available
○ Private
○ Resilience
○ Transparency…
● Trustworthy AI extended properties
○ Free of biases
○ Fair
○ Explainable
○ Interpretable…
Trustworty computing
EU AI ACT
US AI ACT
Chine generative
AI regulations
[source] https://www.heise.de/hintergrund/Microsofts-
Kampf-gegen-Bugs-1429333.html , Accessed March 2024
Bill gates internal memo
6. ● Fundamental mechanisms for data collection and
measurements
● Gauging = quantifying/characterizing
● Monitoring = Tuning
Sensors
7. (A few) Techonologies that have
changed the world
Fire
[source] https://science4fun.info/fire/,
Accessed March 2024
8. (A few) Techonologies that have
changed the world
Fire Electricity
[source] https://science4fun.info/fire/,
Accessed March 2024
[source]
https://www.pinterest.fr/pin/428827195765
555273/ Accessed March 2024
9. (A few) Techonologies that have
changed the world
Fire Electricity Trains
[source] https://science4fun.info/fire/,
Accessed March 2024
[source]
https://www.pinterest.fr/pin/428827195765
555273/ Accessed March 2024
[source] https://www.pinterest.com/pin/ancient-
antique-historic-vehicle-iron-locomotive-metal-
monument-railroad-railway-steam-locomotive-
steel--603060206326098813/
Accessed March 2024
UbiComp 2023 -
From Victorian trains to chatbots:
exploring AI interface design
15. ● How AI is built?
● Where is located in applications?
Sensors for AI?
16. AI model construction in a nutshell
[source] Ottun, Abdul-Rasheed., &.. Flores, Huber. (2023). Detection mechanisms to identify data biases and exploratory studies about
different data quality trade-offs. Deliverable 3.1, H2020 EU SPATIAL, 2023
Standard Machine
Learning Pipeline
[source] https://courses.minnalearn.com/en/courses/trustworthy-ai/overview/
17. AI models in applications
[source] Ottun, Abdul-Rasheed., &.. Flores, Huber. (2023). Detection mechanisms to identify data biases and exploratory
studies about different data quality trade-offs. Deliverable 3.1, H2020 EU SPATIAL, 2023
18. AI models in applications
[source] Ottun, Abdul-Rasheed., &.. Flores, Huber. (2023). Detection mechanisms to identify data biases and exploratory
studies about different data quality trade-offs. Deliverable 3.1, H2020 EU SPATIAL, 2023
19. AI models in applications
[source] Muccini, H., & Vaidhyanathan, K.
(2021, May). Software architecture for ml-
based systems: what exists and what lies
ahead. In Proceedings of IEEE/ACM
WAIN@ICSE 2021 (pp. 121-128). IEEE.
Classical architecture
+ Machine learning
20. AI models in applications
[source] Muccini, H., & Vaidhyanathan, K.
(2021, May). Software architecture for ml-
based systems: what exists and what lies
ahead. In Proceedings of IEEE/ACM
WAIN@ICSE 2021 (pp. 121-128). IEEE.
Classical architecture
+ Machine learning
21. AI models in applications
Classical architecture
+ Federated learning
[source] Ottun, Abdul-Rasheed., &.. Flores, Huber.
(2023). Detection mechanisms to identify data
biases and exploratory studies about different
data quality trade-offs. Deliverable 3.1, H2020 EU
SPATIAL, 2023
22. AI models in applications
[source] Muccini, H., & Vaidhyanathan, K.
(2021, May). Software architecture for ml-
based systems: what exists and what lies
ahead. In Proceedings of IEEE/ACM
WAIN@ICSE 2021 (pp. 121-128). IEEE.
Classical architecture
+ Machine learning
23. AI models in applications
[source] https://www.thefountaininstitute.com/blog/chat-gpt-ux-design
[source]
https://media2.giphy.com/media/W1fFHj6LvyTgfB
Ndiz/200.gif
[source]
https://media1.tenor.com/m/khe_nqmAFJMAAAAC
/driverless-car-veritasium.gif
24. AI models in (existing) applications
Movie matching score
= 75%
AI recommender and
personal guidance
25. AI models in (existing) applications
[source]
https://www.pinterest.com/pin/511440
101439075063/
26. AI sensors and dashboards
AI Sensor: Fairness
Education
Basic
High school
College
Socioeconomic
status Low
Middle
High
Age
Children
Adolescent
Adult
More…
AI movie recommender
AI
Sensor
AI sensors:
• Fairness
• Explainability
• Performance
• Robustness
• Transparency
• …..
Each AI sensor is
linked to a
trustworthy
property
[source] H. Flores: “AI Sensors and
Dashboards”, Computer Magazine
28. AI sensors and dashboards
Monitoring and
adjusting
Trust
score?
Interest
Power
Users
System
Operators Testers
Developers
Auditors
low medium high
low
medium
high
[source] Boerger, Michell., &.. Nikolay Tcholtchev. (2022).
Requirements analysis for AI towards addressing security
risks and threats to system and network architectures
Deliverable 1.1, H2020 EU SPATIAL, 2022