Forum

Input requested for First AI HLEG Workshop: "AI Infrastructure and Enablers"

  • Nathalie Smuha profile
    Nathalie Smuha
    29 August 2018 - updated 2 years ago
  • 7 comments
    Total votes: 0

Dear members of the European AI Alliance,

As the Chair of the AI HLEG set out in his blogpost, the AI HLEG is looking for your contribution on the topics to be discussed at the First Workshop of the AI HLEG. These topics will also be addressed in the two main deliverables of the group. 

The fourth topic of the workshop concerns: "AI Infrastructure and Enablers​"

These are the questions on which your input is sought:

  • What are the most important building blocks at EU level to enable European businesses to successfully apply AI?
  • How can such building blocks be ensured, and how can investments therein be stimulated?

Deadline for input: 13 September 2018

Tags: 
AI HLEG

Comments

Log in to post comments
Ilie GHEORGHE's picture
Ilie GHEORGHE,
04/11/2018 06:04

I wish this series of "Input requested for" would continue. It drives the discussions constructively.

Total votes: 1
Jola DERVISHAJ's picture
Jola DERVISHAJ,
16/11/2018 16:37

Dear Ilie, thank you for the suggestion.The idea was transfered to the AI HLEG Meeting of 8 October and some of the subgroups have already reacted with their respective requests for in input:

HLEG - Input Request: Red Lines

HLEG - Input Request: Business Impact

HLEG - Input Request: Skills and talents

HLEG - Input Request: Appropriate mechanisms to measure the broad societal impact of AI

HLEG - Input Request: World Class Research

Please feel free to become part of the discussions!

Total votes: 0
Loïc Bardon's picture
Loïc Bardon,
04/09/2018 10:17

In the AI economy, force generates force. The more data you have, the better your product; the better your product, the more data you can collect; the more data you can collect, the more talent you attract; the more talent you attract, the better your product.

It is economically a virtuous circle.

However, the United States and China have already sucked talent, market share and data volume into this circle.

Not Europe.

China, for example, has an abundant reservoir of data (demographic, transactional, behavioral) that serves as a giant test laboratory to train AI learning algorithms (700M Internet users). According to an unpublished Pentagon report cited by Reuters after some leaks, the U.S. government now considers Chinese investments in American start-up AI specialists as a potential threat to national security.

What to do about the "data" supremacy of the GAFA and BATX at the origin of their technological monopolistic power?

The first thing is to liberate Europe from the "Dataïst" yoke imposed by the platformists. Easier said than done? Not so sure. A small, forward-thinking European nation had its nose in the hollow almost 20 years ago.

Estonia has developed, and has already proven, the technology base for its citizens to benefit from dematerialised and ultra-secure access to 2,000 online services. This technology, a state platform created in 2000, is the X-road (https://e-estonia.com/solutions/interoperability-services/x-road/). Why is that important? Because the X-road allows all integrated organizations (almost 1000 to date) to share on the state blockchain interoperable databases through the APIs available.

The two main countries with a stake in the global digital economy, the United States and China, are relying on the critical size of their domestic market and their national champions to become world leaders in AI. To position itself in the race, Europe has no choice but to rely on its domestic market by offering :
◾an innovative compromise that would fight against our propensity to always choose free and ease of use rather than privacy
◾let's say another model reconciling both
◾voir to allow users to monetize their data while keeping control of them

European start-ups could therefore rely on a "European architecture" and thus offer services that tickle the moustaches of the GAFA by focusing on the BtoB market of AI, which remains the Achilles heel of digital giants, or even by designing overlayers built on the "digital" back of the GAFA by already focusing on the technologies of the day after tomorrow. What if, for example, a European startup offered you a system of recommendations that allowed you to increase the value of your data? This is the scenario of Marc Elsberg's book Zero, which raises immense ethical questions of course.

Total votes: 2
David Pereira's picture
David Pereira,
03/09/2018 17:37
  • What are the most important building blocks at EU level to enable European businesses to successfully apply AI?
    • ​DataOps tools & methodology
    • API-based AI consumption model
    • Functional monitoring
    • Explainable AI solutions
Total votes: 2
Kurt Van der Veken's picture
Kurt Van der Veken,
03/09/2018 08:45

In general Identity and Acces Management integration with local Identity Providers for enterprises and public IDPs.

For machine learning, EU can consider the following;

  • A BigData platform with open Data of the EU for all citizens in EU.
  • A Machine Learning toolkit (Python based) including versioning for development of algorythms.
  • An API management tool that publishes and documents API's for the developed algorythms (and provides a feedback loop).
  • A generic portal with sufficient documentation on how to use the above (example for Belgium: https://data.gov.be/en).
  • A GUI platform for EU citizens to experirment.

For conversational AI, EU can consider the following;

  • Natural Language Processing infrastructure for languages used in the EU.
  • Text2speech & speech2text components for languages used in the EU.
  • A platform, that lists the uterances (intents & data entities) used in different languages and use cases.
  • A data platform,that retains anonymous voice and chat history build on-top of these conversational AI platforms.
  • An API platform that can interface with Enterprises.
  • A GUI platform for EU citizens to experirment.
  • A generic portal with sufficient documentation on how to use the above.
Total votes: 1
Alexander LÖSER's picture
Alexander LÖSER,
12/09/2018 12:46

Dear Kurt,

I would love to envison contributions to a dataset search engine. https://toolbox.google.com/datasetsearch is a good first shot by Google. Even though, it's not made in Europe, we should consider contributing here, in particluar for datasets on conversational.ai

Basically, those companies which own traning data in Europe often lack resources in R&D (mostly it's still R these days) to come up with their own chatbots. Therefore, a habour for training data would be my prio 1. This could trigger a large set of hackers and enthusiasts which could apply existing research from good papers in Github to novel domains. Those models could be resused by startups or dedicated groups in companies, without being faced by the technical hurdle of training models on their own. I would expect that those companies will provide fancy interfaces to their existing customers to enrich their services and keep their customers happy.

 

Best's

Alexander 

Total votes: 0
Stefano Quintarelli's picture
Stefano Quintarelli,
30/08/2018 15:23

I would say not only "apply" but "develop and apply". I believe, in some sense, the regulatory framework is both an enabler and an infrastructure.

 

whilst many aspects (for example explanability, verification, etc) have already been discussed elsewhere, I think the following issue has not been addressed yet: when we think of systems that optimize a specific function, in some (many?) cases we should take into consideration the collection of those systems as a group, not an sequence of individual systems.

(e.g. when we think of the benefit of driving tehnologies, we speak of the potential number of lives collectively saved)

should we measure if those systems are behaving as expected optimizing a specific function, we could find that overall those systems are behaving as expected, improving some aspect of our lives, even though some specific individual system is not performing as expected.

consider for example some kind of life saving technology. if it saves 1000 lives and kills 10, we should celebrate as a great success the overall system saving 1000 lives rather than individual systems killing 10 (somewhat extreme example, trying to make the point).

in more general terms, this comes down to the definition of the function we are trying to optimize and to the responsability of the provider of the system with respect to the actual behaviour of the overall system.

if behaviopur requirements (responsibilites, liability, etc) are set for each specific instance of a system, in some cases, we could loose the benefit of the overall systems.

a new kind of regulatory environment could be beneficial to addressing this issue; IMHO, it could be an enabler to accelerate development and application of AI to some areas.

when I was in the Italian parliament in the last legislature, I proposed a bill to address this kind of situations. you can read it here (with a more thorough explanation)
 

Total votes: 2