Blog

Your input for the First Workshop of the AI HLEG

  • Pekka Ala-Pietilä profile
    Pekka Ala-Pietilä
    29 August 2018 - updated 2 years ago
  • 2 comments
    Total votes: 4

By Pekka Ala-Pietilä, Chair of the High-Level Expert Group on AI

/futurium/en/file/pekkapicturesmallerjpgpekka_picture_smaller.jpg

 

Dear members of the European AI Alliance,

First of all, let me thank you for your contributions to the discussions on the platform of the Alliance in my capacity as Chair of the High-Level Expert Group on Artificial Intelligence (AI HLEG).

The AI HLEG has been tasked with the delivery of two very challenging pieces of work: (i) draft AI Ethics Guidelines, to which companies, organisations, individuals, institutions, and all other entities will be able to voluntarily adhere; and (ii) Policy and Investment Recommendations, which will offer advice on how to deal with AI-related challenges, and on how to ensure that Europe will be a highly competitive and global player in AI. In order to make these deliverables a success, your input and your support will be crucial.

The summer is almost behind us, and the time has come to intensify our efforts. After a successful introductory meeting of the AI HLEG, held before the summer (minutes available here), it is now time for our first Workshop, which will take place on 20 September 2018. It will be an opportunity for the members of the AI HLEG to share ideas with and learn from each other, as well as to seek contributions from all of you willing to share your knowledge and opinions. 

One of the core tasks of the AI HLEG is to reach out to the entire AI community and the stakeholders involved, and seek your input on the deliverables that we have been tasked prepare. Our Workshops will serve as a tool to discuss more in depth the topics to be covered by the deliverables, in order to help us with the drafting process. I am therefore kindly asking your input on the topics that will be discussed at the Workshop of 20 September, which will be the first of three workshops in total. 

The draft agenda of this First Workshop can be found here. As you will see, five topics will be discussed in particular. Accordingly, we have now created five discussion treads on the forum, each of them aiming to collect your contributions on these particular topics, and your answers to the questions raised thereunder:

  1. Trusted (i.e. responsible and safe) AI
  2. Transparency and Accountability
  3. Industry and Ecosystem Uptake of AI
  4. AI Infrastructure and Enablers
  5. Use-Cases for the Guidelines

We will carefully go through your comments, and draw on your input during the workshop discussions and while drafting the deliverables. The deadline to submit your contributions is 13 September, i.e. one week before the Workshop, allowing us to process your input well ahead.

In addition to the feedback that you can provide via the Forum, a recording of the workshop will be made available.

All of you have the opportunity to contribute to the reflection process of the AI HLEG, and I am personally looking forward to receiving your valuable contributions.

Best regards,

Pekka Ala-Pietilä

Comments

Log in to post comments
Vanja Skoric's picture
Vanja Skoric,
13/09/2018 17:05

Dear Pekka, 

many thanks for soliciting inputs on these crucial topics. Please find a contribution from the European Center for Not-for-Profit Law (ECNL) on the topic of Transparency and Accountability. We urge the group to include more civil society voices in this important discussion. 

Best regards,

Vanja & Francesca, ECNL

---------

(1) Importance of the AI-context transparency and accountability grows exponentially when discussing potential impacts on everyday lives, human rights, fundamental freedoms and, in particular, people’s rights to seek effective redress in case of negative consequences. When discussing those concepts, we think of: a) transparency of the criteria used by the automated processing mechanisms to make decisions and b) accountability of the redress mechanisms available to challenge those decisions. This is the view-point of the rights-based approach, putting the human wellbeing at the center of the AI-related context and discussion. We are especially concerned over the most vulnerable of the stakeholders in this context, which include civil society, marginalised groups and individuals. 

 

(2) Rights-based approach requires governments and states to provide an enabling, positive environment for the exercise of all human rights and fundamental freedoms. In the AI-related context, the onus is still on the governments and states to ensure that the use of algorithmic processing does not impair the protection and enforcement of human rights, recognising that obligations to respect human rights apply equally online as offline. 

 

(3) Following the rights-based approach towards the AI-context, transparent, accountable and ultimately responsible AI is the only way to achieve adequate protection of  human rights, fundamental freedoms and, in particular, people’s rights to seek effective redress.

 

(4) Regarding transparency, it might be useful to recognise:  a) the need of creators of AI systems to describe and give access to information on how such systems work; and b) the need of users or recipients of such systems to have access to information. These two aspects should include describing, inspecting and reproducing the mechanisms through which AI systems make decisions and learn to adapt, and methods of using and managing the created data. 

 

(5) Additionally, we could distinguish several situations of transparency: a) so called “technical” transparency that includes technical details of the AI system (i.e. source codes), which should be available for inspection to some stakeholders as a minimum (e.g. the regulators). Full technical details might be excluded from the public exposure due to the possibility of misuse, hack, or similar. b) So called “data” transparency that includes immense quantity of data that AI systems use to learn from, which, even if published fully, would likely be difficult to process and comprehend by humans. 

 

(6) There is a clear need to strike an appropriate balance to find a meaningful level of transparency that will serve its purpose for different stakeholders. The public will surely not trust „black box“ models that are not transparent. 

 

(7) Regarding accountability, there is a need to explain and justify decisions and actions of the AI system to all interested stakeholders. That means that the responsible AI system should provide meaningful insights on how it works and what are the factors driving algorithmic decisions. Decisions should derive from the „explainable“ algorithms that can be examined and debated, and ultimately overridden in the case of redress mechanism.

 

(8) Additionally, we could distinguish situations of accountability when humans can understand and trace how AI system has reached a specific decision, and when that is not possible. Increasingly, algorithmic processes and the outputs of the AI system are inexplicable even to their creators. These cases should require stricter oversight, as accountability must remain even when creators claim they cannot understand what their systems can do, or how they do it. There is onus on creators to find and include mechanisms for meaningfully explaining how the decision-making in algorithms is made.

 

(9) Levels of transparency and accountability should be determined by a responsible governments and states, not commercial interests. As numerous international obligations and documents provide, governments have positive obligations to protect human rights against abuses, including by private sector actors. Considering the risks stemming from the lack of transparency and accountability, governments should ensure that private companies adopt a preemptive approach and develop, jointly with other stakeholders, preventive policies concerning the risks of the use of AI systems, and its impact on individuals and society. 

 

(10) Governments should also conduct human right’s impact assessment, including public oversight, with experts from different sectors and fields, to identify, prevent, and mitigate against human rights and freedoms risks, and publicly account for their efforts. This should be done in all phases, but specifically prior to development, acquisition or implementation of AI systems. It should include relevant civil society stakeholders to take part in the creative and adoption process in order to contribute to impact assessment. Governments should also evaluate the implementation throughout the entire life-cycle on the AI system. 

 

(11) The very minimum that governments must ensure is the “explainable AI” - that is, the system where humans are able to demand the data and explanations behind the algorithmic decisions made for them, when impacting any aspect of their lives.  That includes developing methods to track what is motivating the decisions of machine-learning algorithms, to identify relationships between inputs and outcomes, spot possible biases, and give routes into redressing potential negative impacts. Government regulators should keep or make private companies responsible for providing meaningful transparency and accountability that makes the „explainable AI“.

 

 (12) Developers and creators of AI systems should include relevant stakeholders into preventive measures and pilots to identify risks, such as data quality issues, bias issues, conducting testing methods and piloting trials, ensuring that potentially affected groups and relevant experts are included in all phases of design, testing, and evaluation.

 

(13) Governments should utilise the mechanism of procurement or engagement of private services in public service delivery to maintain direct oversight, ownership, and control over the use of AI systems in their jurisdiction. 

 

(14) Governments should ensure that all individuals and groups affected by AI systems have the right to redress - that is, to challenge relevant outputs and decisions of AI systems that impact their rights and freedoms before a competent authority / judiciary. This should include the right to contest, reconsider and change decisions by using alternative methods of reasoning (other than AI system). These rights should not be signed away by terms of services or any other means.   

 

(15) In addition to regulation, governments, in collaboration with private companies, academic sector and civil society, should develop and implement measures of raising public awareness, training and capacity building of the public, government regulators and experts, raising educational skills for understanding  the AI through dedicated educational and life-long learning programs. 

 

Total votes: 0
Paul NEMITZ's picture
Paul NEMITZ,
30/08/2018 16:14

Democracy and Technology in the age of Artificial Intelligence

Dear Pekka, dear colleauges,

I wanted to share with you some thoughts on Democracy and Technology in the age of Artificial Intelligence which I just put on paper. You can find them here:

Nemitz, Paul, Democracy and Technology in the Age of Artificial Intelligence (August 18, 2018). DOI 10.1098/RSTA.2018.0089 - Royal Society Philosophical Transactions A. Available at SSRN: https://ssrn.com/abstract=3234336

This is a preprint version, still under Press Embargo, publication if forthcoming. I hope your thoughts will be inspired by this piece and please let me have any comments you may have at paul.nemitz@ec.europa.eu.

 

Kind regards, Paul Nemitz

Total votes: 0