Dr. Emilia Tantar, Chief Data and AI Officer at Black Swan LUX, discusses the company’s journey, its focus on responsible AI, and how businesses can prepare for the EU AI Act. A spin-off from the University of Luxembourg, Black Swan LUX is a leader in AI innovation and compliance.
What is Black Swan LUX’s primary mission?
Black Swan LUX spinned-off from the University of Luxembourg in 2015 as a software company, with proprietary AI based anomaly detection solution for the wearable market. The name of our product is safeliveapp, and you can find it on our website. This has originally been our primary mission, in which the National Research Fund and private investors have put their faith and their money. With time we enlarged our mission to be also that of an SME encompassing strategic consulting around the responsible AI advent.
I believe this is today’s focus. We are and have been since the beginning strong supporters of the EU AI Act and related regulations and providing concrete instruments for a responsible AI adoption. As an SME, we have little financial resources, but we have invested a lot of our means in the EU process we believe in.
We have delivered AI for the public and private market since more than 20 years and we quickly understood that the techniques underlying the umbrella of “artificial intelligence” are approximate ones and come inherently with risks, thus as for any innovation a responsible approach is not to stop its adoption, but set boundaries aligned with EU values, that of trustworthy AI.
Our missions remained always aligned with our values, to serve the society through our research and strategic AI and data leadership capabilities and it is only natural that since 2018 we started preparing for assessing the risks and quality of AI systems and actively engaged in standardization. We did so by first leading the assessment of the challenges AI systems conformity will bring operationally for us and the EU market, leveraging on the more than 280 experts of the CEN/CLC JTC 21 AI. The technical report that I lead as editor “TR AI Conformity assessment” has been approved by all EU member states and is under publication, being the first such study worldwide and we continue with providing standards and services to support conformity for EU AI systems.
Soon, AI systems will be audited as part of the AI Act. Which companies should be most concerned about this today?
The obligation of conformity assessment of AI systems performed by a third independent party impacts companies which develop AI systems that fall within the high-risk category of the EU AI Act.
“#BeWiseStandardise! Clear, operational standards can ensure a smooth and efficient process for compliance with the EU AI Act.”
Dr. Emilia Tantar, Chief Data and AI Officer, Black Swan LUX
The EU AI Act has been published on August 1st and it applies in a first stage to companies that fall under the ban. Are there any criteria in particular that they should be aware of?
In terms of technically preparing compliance, the compliance with the requirements of the EU AI Act will extend to the AI layer existing risk management and quality assurance obligations established for all software systems available on the EU single market. Furthermore, as the AI regulation applies on top of existing regulations, requirements relating with data quality and governance, already known as for the GDPR and data act obligations will apply. As all AI systems are t on top of data, cybersecurity related requirements will also apply.
What is specific for AI systems relates with the trustworthiness of the AI systems, which are stochastic by nature. Four main technical specifications were identified in the EU AI Act for AI systems under the trustworthiness umbrella, namely transparency, human oversight, robustness and accuracy.
It should be noted also that the European Commission already mandated the European standardization development organizations with developing technical standards to become candidates for harmonization for all these aspects of conformity assessment for AI systems. It needs to be noted that technical specifications in support of the EU AI Act were requested by the European Commission to CEN/CLC JTC 21 AI and according to regulation 1025/2012, CEN and CENELEC (in this case) are the only organizations mandated to provide standards in support of the regulation.
The exact requirements that I briefly summarized above are listed in Annex 1 of the Standardisation request and that will be a good starting point to start preparatory work, till the standards are made available on the market.
How ready are Luxembourgish companies to be audited?
The readiness level can be self-assessed by all following a two-dimensional approach, the needs in AI skills and existing and planned developments of AI systems. Only after passing this assessment exercise the companies will have a clear view of their readiness level.
In terms of AI auditability, for companies already subject to regulatory constraints that have auditable processes in place and where risk and quality management are available, not to mention the necessary roles of compliance officer and adjacent ones, the AI audit will be smoother and imply transitioning from existing practice to amended variants. This will imply considering AI specific risks and quality assessment, as well as the AI specific trustworthiness dimension, which require further new AI specific skills.
“[…] even the EU AI Act mentions AI literacy as an obligation for the stakeholders in the EU AI value chain, from provider to deployer or operator.
Dr. Emilia Tantar, Chief Data and AI Officer, Black Swan LUX
What can Luxembourgish companies do to better prepare themselves?
I always say #bewisestandardise! Having a small number of clear and operational standards that bridge well with existing management systems could ensure companies a smooth and efficient process for compliance with the EU AI act.
As to ensure standards, candidates for harmonization are aligned with existing practice, I would strongly advise getting involved in the standardization process or if that is not feasible, catalogue the existing management standards and skills available in the company and project them to the future.
As convenor of the CEN and CENELEC JTC 21 WG 2 Operational aspects and President of the National Mirror Committee in Artificial Intelligence Standardisation, I would strongly encourage experts to join the current efforts in AI standardization at CEN and CENELEC JTC 21 European level, while not ignoring also the international standardization landscape.
Since some years at Black Swan Lux S.A. we specialize also in providing support to our various partners towards AI system’s compliance (from SMEs to Fortune 100) and we are ready and eager to support the local Luxembourgish companies become forerunner in compliance with the EU AI Act and reinforce their responsible AI approaches, of course while continuing our efforts in leading the development of the European standards. As a spin-off of the University of Luxembourg our main interest is to support the Luxemburgish AI economy thrive as that allows us to give back on the trust provided to us, and also supports our own resilience. The advantage of being a small, trusted country as Luxembourg is that we support each other and even as a SME we can do our part in supporting the LU ecosystem resilience.
What about SMEs and startups with less capital?
A great initiative we find useful particularly for preparatory efforts for SMES in this direction is provided by the EIT AI Community through an online tool allowing to assess your risk level using the AI Act Conformity check – ai.eitcommunity.eu/services/ai-act-conformity-check. The initiative is available in English and German. While this might not be sufficient per se, it allows companies to assess the requirements they need to align to. Next step would be using standards as a tool to ensure compliance, as the ones developed by CEN/CLC JTC 21 under the mandate from the European Commission.
As an Expert representing Small Business Standards (SBS) interests in standardization and as a EU Digital SME Alliance ambassador -and as done since some years together with the other SBS experts- we will do our best to ensure that the standards supporting the EU AI Act are operational and SME friendly, using concrete assessments as for example through Guide 17 assessment. We consider our efforts are important in supporting the EU economical resilience given that SMEs represent more than 90% of companies in Europe and AI becomes, like electricity- an engine for navigating the digital economy.
The AI Act is not just about regulations but also about competencies. What are some of the competencies companies should start recruiting sooner rather than later?
This is indeed an important aspect, as even the EU AI Act mentions AI literacy as an obligation for the stakeholders in the EU AI value chain, from provider to deployer or operator. Knowing that AI system developments are based on data and data is subject to cybersecurity concerns, especially for organizations handling sensitive data or high-risk systems, data and cybersecurity skills need to be covered. Relating to AI specific skills as for any software developments the AI systems risk management and quality management aspects will need to be covered as well. More details on the AI required competencies will come from the EU AI Office and it is clear that the ability to technically perform assessment of AI systems for conformity with regulatory constraints in EU and internationally will be key.
This article was published in the special edition on artificial intelligence of Silicon Luxembourg magazine.
