Meeting Minutes
===
###### tags: `AI Proficient` `Meeting`
Nov 3rd, 2020, online

**Participants**
* WLOTZKO, CHRISTIAN (Continental)
* Christian Matejicek (Continental)
* ASTIASARAIN, PAUL (Continental)
* Aitor Arnaiz (Tekniker)
* Egoitz Konde (Tekniker)
* Santiago Fernandez (Tekniker)
* Alexandre Voisin (UL)
* Marc Anderson (UL, post-doc)
* Karën Fort (UL)
* Benoît Iung (UL)
* Christophe Cerisara (UL)
* Van Phuc Do (UL)
* Van Thai NGUYEN (UL, Ph. D. Student)
* Alaaeddine Chaoub (UL, PhD student)
* Lazar (IMP)
* Dea Pujic (IMP)
* Nikola Tomasevic (IMP)
* Uros Milosevic (Tenforce)

# Roundtable
See video?

# Presentation by A. Voisin
## WP1 overview and expected outcomes
WP1: use case provider

ethics by design (address ethics from the very beginning of the project)

KPI ==> to add to the project dictionary (Key performance indicators)

WP6: use case evaluation and ethical recommendations

:dart: First deliverables 6 months from now on 2 tasks

### 1.1:

1.1 show all the issues to be demonstrated
Wish lists of Conti and Ineos - should be open as much as possible
* Technical issues
What are the positions of the humans

show the issues/open wishlist: help the human to perform better, create better quality

T1.1. Consider several points of view with regards to current situation: conventional performance, not conventional one (e.g. sustainability), Human interaction with the machine, the components; the means of interactions (e.g. ergonomics) ... [Not only focused on technical considerations], organisational, culture(?), processes(?)
T1.1. Make a reference about "new needs and requirements" with the impact defined in the projet (see the proposal; expected impact, specific impact pp 32 to 36).

Conti interested in human interface experience of UL (1.2)

### 1.2: ethical recommendations
* Ethical point of view should be taken into account very quickly from the beginning, not sequentially
* the ethical teams needs to know more since we are not from manufacturing,
* ethical part will be bottom up
* how to bring value to the use cases
* it would be good for ethical team to discuss with operators (using questionaires)
* to insure the worker's position
* wishlists should correspond to impacts given in the proposal

- feedback and how to apply the recommendations
- help to add human interface things in the wishlist
- for all the different people involved, including the developers
- we look at each use case as it comes up and we discuss it: what kind of problems there are. We'll find solutions together. One hour meetings if possible
- the ethical team can bring confidence to the customer, etc.
- We have to build the future of ethics in manufacturing, as it's not very developed
- The ethical aspect of AI technologies should be explored as well as the operator/machine aspect
- we can't just pin the ethical aspect to the pilot, we must generalize and be able to apply to other beyond manufacturing (beyond one algorithm, etc.)

Ethical recommendations built bottom up, not top down, with the companies.
Law is not ethics: ethics is soft law and it brings value to process/products
Ethics as a value !
Ethics should impact not only the operation of the system but also all the life cycle phases before and after the operatio (e.g. AI algorithm design, algorithm development ... )

Conti: Explanatory video ([https://www.youtube.com/watch?v=HjViHN6xuwk](https://))

1.2: need to be able to interact with operators (questionnaire)
No disturb workers with interfaces (which might lead to dangerous situations)

Ethics not only focus on the use case applications, but also on the AI technologies and tools arising as a result of the project (Ethic by design?)

### 1.3 Follow-up specification of pilot demonst. scenarios
T1.3. Select some results (list of wishes) issued of T1.1 to define future and realistic scenarios in consitence with the impacts to be demonstrated.

* proactive maintenance, improved product quality
* identify which assets can adress potential problems
* deliver real world scenarios
* decision on whether each is useful
* WP2 not neccesary directly related to human
* focus should be to help the human increase quality and efficiency and to ease the operation and execution (Aitor)
* By the end some good impacts have to be shown
* The viewpoint of AI technologies is also important

- Identify all technologies, potential embedding into existing cases,potential impact...

### 1.4 Project requirements and KPIs
related to operators as well as technical matters
T1.4. should provide metrics on different considerations (as already highlighted for T1.1.) such as technical, human-interaction, ... to be used for assessing the added value of the AI-PROFICIENT solutions (gap between the current situation at the beginning of the project and those obtained at the end of the project. Metrics could be generic (not necessarily dedicated to the three pilots) but of course used at least for the 3 pilots.
These metrics must be of course representative of the impacts we advocated in the proposal.

:warning: BI: impact of COVID to consider with our PO (project officer)=> moving some deadlines? To discuss during Kick off (end of Nov)

### 1.5 System architecture (ethics by design)

* should consider all the inputs and recommendations of the previous tasks in WP1
* but main result should be the architecture of AI proficient platform and high-level description of AI services to be developed (both at edge and platform-side)

:book: RAMI 4.0 -Reference Architectural Model Industrie 4.0


* provide description of different AI and machine learning services to be incorporated in AI-PROFICIENT platform (service specification, what are the inputs/outputs)
* some of the services will be able to act stand alone, but interaction of different services should be also considered by platform design
* one specific service might tackle one specific segment of manufacturing process >> but a high level view is needed

# Continental use case presentation (PA)

* machine is producing tread and sidewall
* tires have different mixes of rubber and raw materials
* repeatability of process
* Combiline - a quintuplex extruder
* where the "raw" rubber is introduced through 5 different extrusion heads
* many use cases in the thread packing - very complex process and the sidewall packing
* Each of us will get a package in detail with an overview of the parts of the Combiline
* the information displayed will be at the security level that it can be shared
* one challenge related to variability of raw material feed to the combiline
* use case description format nice (synthetic) may be miss:
* data is already available (and data wished to be added)
* involment of operator
* Ethical issues to be tackled (should be added as an item as such, so it's taken into account in all use cases)
* which impact on the proposal is fulfilled by working on a particular use case
* split use case description info: problem & expectation from AI-PROFICIENT

Challenge areas (The Wish list)
1. in the mixing - with regard to the quality of the product
2. in the combiline - see figure slide 4
3. Make a link with the impact consideration defined in the proposal: What are the impacts referred to the challenges/new scenarios proposed?
Use case 2:: Every 20 min. there is a change in production (affecting all parameters) set-up chage time is very important

Use case 1: more focused on the predictive maintenance and failure/degradation detection
if e.g. a bearing breaks (or is breaking) it needs to be detected in advance, otherwise the speeds of the conveyors are pushed out of sync
- ethical issue - speed increase may affect/injure the operators (the AI just gives advices, but still, the interaction could slow down the reaction)
- ethical issue - loss of experience from less operator interaction
Approx. 12 different use cases to be presented in WP1 kick-off in this way.

Some developments done could be also not "fully implemented" on the pilots. Indeed AI-PROFICIENT is a RIA with R meaning Research.

* Human-in-command - AI is providing just a decision support, while operator executes the control manually;
* [Human-on-the-loop](https://itlaw.wikia.org/wiki/Human-on-the-loop_weapon) - operator with a supervisory role, while AI is suggesting the control strategies and executes after approval; and
* [Human-in-the loop](https://en.wikipedia.org/wiki/Human-in-the-loop) - implicit knowledge by the operators is fed into the AI optimization cycle
* Fully automated, semi-automated and manual control to be considered as part of the use cases; real-world limitation will be encountered (what is feasible or not to be demonstrated)
* we have to remember the research and innovation impact aspect of the project - which may result in 'we tried this interesting avenue but it was not feasible'

These three previous items about human commanded etc. should be integrated in the frame proposed by CONTI (how the use case is placed with regards to these 3 items?). The same feedback on the Impacts considered (to be added in the frame).

# 16th Nov Draft Agenda
Morning
* general presentation of work package 1
* two use cases presentation from Conti and Ineos and maybe virtual visit (presentation of product and production means)
* proposal of what the ethics team expects, presentation and characterization of ethics
* definition of ethics and of Ethics by Design (probably software oriented), a few proposals, and we get some feedback

Afternoon
* Discussion, etc.

# Actions?
- Pass to ineos continental info
- Review these notes