AI Day Ethics discussion

ChatGPT4 audio transcribe feature was used

https://chat.openai.com/share/4f38c0b8-4316-4134-b6a5-4c2309e6e47b

Anonymous

What is all about and what is the approach there? Any person you want to comment on? Yeah, okay, so one comment. One is, indeed, in terms of the overall conversation around regulation, getting more diverse and also integrated, one more aspect. Part of one thing that complicates the discussion, generally the discussion, is a claim so that it is focusing on certain rules, certain paradigms. This is an example, the kind of examples of typical things, and how, and concerns that are repeated in media and social media over and over and over again. So, this is a bit paradoxical situation. We are developing a more diverse, more rich regulatory approach and at the same time we are focusing on this very small culture. And this is the kind of challenge that I think is now being reflected everywhere. You have sometimes referred to this thing as a concern, but one is not really this. It’s obviously, over 30 percent social structure, and policy discussion. So, what constitutes an AI act? Well, as we’ve done already, there’s been this in a couple of sentences. So, one really critical part of the AI act is the definition of AI. It defines an individual as the source of the regulation. It is the thing that is used as a biological technique to separate the AI system from all the other social systems. And it is a bit misleading to talk about the definition of AI anymore as the draft of the regulation has been changing so that the focus now is on the AI system. And this doesn’t make the challenge of defining the thing any easier because now there is this challenge to define the AI and to define the AI system. And this is a bit, I think, probably a bit politically not correct thing to say, but if I wanted to know what an AI is, I wouldn’t ask from you as a human. Instead, I would ask from you, right? What do you think it is? So, the definition of AI is the first part, the regulatory framework. The second part is the risk-based structure of it. So, the idea is that the AI act will regulate the style of trade order in supply chains or value chains so that depending on the type of risk that is raised by the use of AI or if related to the use of AI system, the precise obligations and requirements will be different. But anyway, the risks are characterized so that they are always the transition of product-based legislation. So, the idea is that, okay, they can be handled like products, but they are only in terms of basic rights. This is another approach which partially explains why it has been so difficult to formulate the legislation in a non-structural way.ChatGPT

ChatGPT

The provided input appears to be a detailed commentary on the complexities of regulating artificial intelligence (AI) and the formulation of the AI Act. The speaker is noting that while there is a development towards a more diverse regulatory approach, there is a paradoxical focus on a narrow aspect of AI, often reiterated in media discussions. They mention that a key part of the AI Act is defining what constitutes an AI system, distinguishing it from other social systems, which is a challenging task.

The speaker also touches upon the risk-based structure of the regulatory framework, which suggests that the AI Act aims to regulate AI applications across supply and value chains by assigning specific obligations based on the level of risk posed by the AI system. They highlight that this characterization aligns with a product-based legislative approach, which further complicates the creation of the legislation due to the need to ensure basic rights are protected.

This person seems to be critiquing the current state of AI regulation discussions, pointing out the difficulties in defining AI, the narrow focus of the debate, and the challenge of drafting effective, comprehensive legislation. The comment on asking an AI about its own nature rather than a human might be a rhetorical device to underline the complexity of understanding AI systems.

Anonymous

First, the vision, which is both types of vision. They work in three core visions, categorically, including social, economic, and other examples. Then there is the category of hybrid AI systems, which is the bioclass of the uses of AI systems. That is, everything, most of the applications related to the AI app. And these different AI systems that fall into the broad category of hybrid systems are those that may have the behavior of the views of non-human or race, sometimes human-type of cases, like we may be doing. Then there are systems, or if you want to use a word that’s not yet a word that’s okay, then there are systems. There’s a certain system that’s way more transparent, and that’s where it is challenging. And then there is typically the idea that, okay, those systems are so-and-so that it’s difficult for a human to know whether or not it is the kind of product produced by these techniques or whatsoever. And the idea of the program is that transparency. Transparency is, for instance, there are a lot of people who know whether or not they are interacting with AI. And the broad category is mostly all the other AI systems, which are thought to raise a multi-dignity value. But in the current world, there is an additional layer, the so-called post-diplomacy, which covers most of the systems that are generated by these generative AI techniques. That layer is now being negotiated, and the details are not public or fully decided yet, but in a way, the problem is still there. There will be some form of this goal that there will be five different layers of possibilities, or at least that is the big vision right now. Thank you.ChatGPT

ChatGPT

The speaker is discussing various aspects and visions for AI systems within the context of the AI Act. They outline three core visions which include social, economic, and other types of AI applications. Additionally, they mention “hybrid AI systems” as a significant category, which seems to encompass a range of AI applications that might display behaviors or characteristics associated with both human and non-human entities.

The focus then shifts to transparency, an important concept within AI regulation. The speaker emphasizes the need for people to be aware of when they are interacting with AI. This transparency is particularly relevant in the context of systems developed through generative AI techniques, which can create outputs that are indistinguishable from human-created content. These systems are part of a so-called “post-diplomacy” layer that is currently under negotiation, with details not yet public.

The vision, as described, indicates a complex, multi-layered regulatory approach with at least five distinct layers or levels of regulation, each presumably tailored to different types of AI systems and the various risks they pose. The “big vision” seems to be a regulatory framework that addresses the multifaceted nature of AI, ensuring that ethical considerations, transparency, and societal values are upheld across all applications of AI technology.

Anonymous

If you do it with pen and paper, is it then allowed, or should it be forbidden also in that sense? It won’t be allowed in this sense, because we have other regulations that define us, forbidding some of the programs it uses. But of course, the idea is that it should apply only to AI systems. I don’t think that being a European Union would include using pen and paper, to be part of an AI system. But who knows, maybe that’s possible. Okay, now a round of comments.ChatGPT

ChatGPT

The speaker seems to be addressing a question about whether activities performed with pen and paper could fall under AI regulations, likely in the context of the European Union’s AI legislation. They suggest that while current regulations are designed to govern AI systems, not manual processes, it’s conceivable that future regulations could evolve to encompass a broader range of technologies or methods, potentially including pen and paper processes if they are part of or related to AI systems. The speaker seems to leave room for discussion and further commentary, indicating that the definition of what constitutes an AI system and what activities are regulated by such legislation is still open to interpretation and expansion.

Anonymous

Well, it’s unique, and there’s probably a lot to criticize your better approach to support, but I think I would like to expand, and what I find is interesting is how the foundation model turns out. I guess that’s the last topic that we’re talking about. So, what’s the EU approach to foundation models? I have to be really clear. I’ve been not focusing on the so-called existential risk that many other parties are interested in, and focusing in the governance discussion, but actually the interest is on what needs to happen so that the whole rest of the AI ecosystem can comply with the requirements. So, if they are using this system as a part of their high-risk system, what needs to happen so that they actually comply knowing that from their system? And I want to thank the European Parliament for the first proposal, the paper regarding the transparency requirements for foundation models, and they are transparency requirements not towards regulators, but transparency requirements towards everyone who is using this system. And the approach is very different compared to, for example, the White House approach that they came up with. They have also proposed a transparency for the foundation model provider, but that is for the US government. So, it’s really funny in a way that, yeah, helping government really support all the AI ecosystem, then make sure that the downstream obligations are, yeah, based on and in compliance with the regulations. So, I think that’s something really good and I’m looking forward to really the outcome of this, how the foundation models will be there. Thank you.ChatGPT

ChatGPT

The speaker is reflecting on the European Union’s regulatory approach to foundation models in AI, which are fundamental AI systems that underpin a variety of applications. They clarify that their focus is not on the existential risks of AI, which other parties might prioritize, but on governance and ensuring compliance across the AI ecosystem.

The EU’s stance, as mentioned by the speaker, involves establishing transparency requirements for foundation models. Notably, these requirements are aimed at all users of the system, not just regulators. This is contrasted with the approach of the White House, which has proposed transparency requirements for foundation model providers that are specifically directed towards the US government.

The speaker expresses approval for the European Parliament’s initial proposal on transparency requirements and seems to appreciate that the EU’s regulations are designed to support the entire AI ecosystem, ensuring that all parties using AI systems, particularly those classified as high-risk, are compliant with the necessary regulations. The commentary ends on a positive note, with the speaker expressing anticipation for the outcomes of these regulatory efforts on foundation models.

Anonymous

There is a lot of talk about it as if it is spoiling the AI world, as if it is imposing prohibition on the party. I sort of find that there are those kinds of assertions that are a little bit problematic in the sense that the AI is an innovation. It is also business friendly and it is not. Why? Essentially, if you look at the wordings of the article, you will find that most of the stuff that the AI will trigger would have been way through to the previous economy, but for the existence of paper results. Essentially, what will happen because of the AI is that you have to produce a work of documentation. That is the sense. I would just like to say that the real worst is that the AI is overgrown. But at the same time, what the app seems to be doing is imposing a sort of coordinated software development process for AI development. Essentially, re-engineering seems to be something that is coordinated. I would assume that software development would be doing anything but in the right coordination.ChatGPT

ChatGPT

The speaker is addressing criticisms of AI regulation, particularly the view that it stifles innovation in the AI sector by imposing excessive restrictions. They find such assertions problematic because the AI Act is meant to balance innovation with a business-friendly approach while ensuring compliance and documentation akin to regulatory practices in traditional economies.

The mention of “paper results” suggests that the regulatory framework might require thorough documentation, which some view as a burden. However, the speaker seems to argue that this isn’t necessarily negative; rather, it’s an evolution of the development process that adds a layer of coordination and re-engineering to ensure that AI development is aligned with certain standards and practices.

The underlying message is that while the AI Act introduces new compliance measures, the purpose is not to hinder AI development but to ensure it progresses in a responsible, coordinated manner that aligns with broader social, ethical, and economic goals.

Anonymous

I’d love to hear your questions, whether or not we’ve lost something to that kind of statement. I’m sure they’d also be really informative. Well, we have to think about what is the background of AI. Like, we all believe that it’s both supersized and dimensionalized forces that affect European bodies. Well, if we think about then democracy, about then a sort of movement, my question then comes to you. If that is what a pen means, whether it’s a doctrine of your freedom of movement, whether it’s a doctrine of your democratic rights, be elected, or vote, would you want it to be prohibited? Of course. So that’s why I’m asking also, why is it that in this context, if you do it with AI, it’s prohibited. But obviously it should be, or maybe it is prohibited, otherwise also. So this is a bit confusing for laymen in legal matters.ChatGPT

ChatGPT

The speaker is raising philosophical and ethical questions about the implications of AI on democracy and individual rights. They seem to be drawing a parallel between the freedoms we enjoy in democratic societies, such as freedom of movement and the right to vote or be elected, and the way we approach AI. The comparison with the pen—perhaps as a symbol of writing, expression, and legislation—suggests an inquiry into whether AI should be regulated if it potentially infringes upon these democratic freedoms.

The speaker is questioning the consistency of regulatory approaches: if certain actions are prohibited when done through AI, why would they not be prohibited when done without AI? This reflects a broader concern about the application of laws to AI and the confusion that may arise for those who are not well-versed in legal matters.

The core of their inquiry seems to be about the principles that guide our restrictions on AI: Are we applying our democratic values consistently when we decide what AI can and cannot do? And how do we explain these decisions in a way that is understandable to the general public?

Anonymous

Post-project is an ecosystem of value chains. You have technological giants producing these generative AI models or other tools, on top of which some maybe small or medium-sized IT houses build AI systems or subsystems, which is a part of a larger system used by other companies, maybe a bank or a machinery company. And then there is the end user. How is it balanced in this sense of responsibility? Where it lies? Who is responsible? Who bears the cost? And so on. What’s your take on that? Well, in terms of finance, I work in a department that deals with public sector digitalization. So I’m not just a big-building department. I’m a part of the business staff of some companies, and so on. But the previous version that was negotiated by commission and a member state, there were significant exceptions to the small and medium-sized enterprises. And I think if I partially refer to those exceptions, I know that the regulation was directed more towards the so-called big operators, and so on. But from the public sector point of view, it is concerning us a little bit. I’m not concerned a little bit. But part of the burden will be heavily carried by the public sector, which is at the same time a small workforce. We are trying to get our public sector so that we can rely on and utilize our operators. So this is the kind of person that is responsible. Thank you once again for asking us to work for the benefit of the public sector. Thank you.ChatGPT

ChatGPT

The speaker seems to be discussing the distribution of responsibility and the cost of compliance within the AI value chain, which extends from large tech companies producing generative AI models to the end users in various sectors. They note that previous negotiations for AI regulation at the EU level included significant exceptions for small and medium-sized enterprises (SMEs), suggesting that the regulatory burden was intended to fall more heavily on larger entities.

From their perspective in a department dealing with public sector digitalization, there is a concern that despite these exceptions for SMEs, a substantial part of the compliance burden might still be transferred to the public sector. The public sector often operates with smaller workforces and relies on partnerships with various operators to digitize and provide services efficiently.

The speaker seems to acknowledge the complexity of assigning responsibility across different layers of the AI ecosystem. They emphasize the need to balance regulatory compliance with the operational capacity of the public sector, ensuring it can continue to serve the public effectively while adapting to new AI regulations. The gratitude expressed for being asked about their work suggests they value the opportunity to discuss the impact of AI regulation on the public sector.

Anonymous

If I go from article to article, what’s the AI going to be like? I could be a reader-competent lawyer, I could be a navigator, but I mean, I think it’s just how it’s going to be, and how it’s going to pass the test. I don’t know whether I want to run a literal biopsy of anything, of things, I think. We actually have two sources of science, of science, of the environment and AI development. And so, most of these both have a patient involved. And, for me, it’s a little bit sceptical whether somebody who sees the worldwide development starts without knowing it in any way, and they’re employing it, without knowing what they’re employing. But, in a sense, if you’re trying to, for example, provide public services with algorithms, then you better know what those algorithms do, and what kind of algorithms they’re going to use it. And, to be able to do that, you need that documentation. So, essentially, if somebody buys something and puts it in a bag, I mean, come on. That’s not how things are supposed to be done. And I think that that’s what’s going to be impossible with the AI. Why would they put it in a bag? And I think that’s a good thing. Although, it is a good thing.ChatGPT

ChatGPT

The speaker seems to be discussing the importance of understanding and transparency in the use of AI, particularly in public services. They emphasize the need for those employing AI technologies to be fully aware of what the algorithms do and the implications of their use. The “biopsy” metaphor implies a deep examination and understanding of AI technologies, rather than a superficial or uninformed application.

The mention of needing documentation reflects a requirement for accountability and clarity in AI deployment. The speaker seems critical of the notion that one could use AI without this level of understanding, likening it to blindly putting something in a bag without knowing its contents.

The speaker appears to be advocating for informed use of AI, where those who implement AI systems, especially in the public sector, have a thorough grasp of the technology and its effects. This perspective supports the idea that with AI regulation, such as the AI Act, indiscriminate use of AI without proper documentation and understanding will become impossible, which they view as a positive development. They argue for responsible and informed AI deployment as a standard practice.

Anonymous

I’m not a really good programmer without a single line of documentation. That’s just the fact. If you do your risk management to document it, because no one wants to communicate with the rest of the team. So I think way too much of our focus is on documentation. Documentation is the evidence of your work on the requirements. But the AI is tuned into not creating unnecessary documentation, but actually complying with the requirements, doing that governance work, identifying their risks, understanding the mitigations, doing those mitigations, and documentation is evidence of it. I don’t know if I want to explain this, because there’s so much physics on the documentation requirements. So let’s think about the other requirements and documentation and evidence of how you know and how you do that. Yeah. Then on the burden, I think something really interesting to see in the final paper, final recordings, is also what will be the final report for the employers of high risk. Because there are versions where there was a lot of new burden for the employers. Not the ones who are building these products, but for example, companies who are using Workday or other human resources-related products that include features, AI-based features. So what are the requirements of these employers of high risk systems? And that is interesting, because if what Parliament suggested for the responsibility of these employers would go through, that would actually, I think, scale the impact hugely and be a really good additional burden for the employers. But one thing that I still want to also reinforce is that no one in the supply chain for the end users of these products can actually finally do judgments about the risk, whether there is a high risk in that context for the system of supply. So depending on that, I think that it’s really hard to push that burden completely to the people on the upstream, yes. And actually, what I’ve been discussing with some of the Frontier Models or Foundation Models developers, they really struggle having that visibility into the actual end-user application. And again, there is a transparency need from the end users to work with Frontier Models developers as well, to help them understand the risks that appear when we apply these systems in specific contexts. Thank you. To summarize, you see that for the developers, tech companies, it’s only there just to do it properly, to do the documentation, to do the risk analysis and certify that. So we’re a bit worried, or more than a bit worried, about the end-user, the burden on the end-user and how that works out. Yeah, I’m not sure if I have much to add to your opinion. It’s just, it would scale the impact and scale the impacted organizations and the volume of processes that would need to be established to totally defend the model. So it’s a definite cost factor. But I don’t want to underestimate or make it smaller the expected impact to the providers as well. Of course, it will require a lot of new work. And that’s actually why my company exists. We want to make that more efficient. And I hope in doing this, applying it in a societal activity at a societal quality. So that’s my view that, of course, there will be additional burden. And it’s going to be very difficult if you do it with pen and paper. Because there is so much dependency and information that you need to keep yourself up to date. So I think there are several more solutions to support it.ChatGPT

ChatGPT

The speaker is addressing the role of documentation in AI development and deployment, particularly in the context of risk management and compliance. They argue that while documentation is crucial, it should not be the primary focus. Instead, the emphasis should be on actual compliance with requirements, risk identification, understanding and implementing mitigations, and then using documentation as evidence of these processes.

The speaker expresses concern about the potential burden on end-users, particularly employers who use AI systems categorized as high-risk. They mention that legislative proposals could significantly increase responsibilities for these employers, which could scale the impact and create a substantial additional burden.

They also highlight the challenge for developers of foundational or frontier models to have visibility into how end-users apply their systems, emphasizing a mutual need for transparency to understand risks in specific contexts.

The speaker concludes by stressing the importance of efficient compliance processes, which their company aims to support, recognizing that the burden of new regulations is significant. They note the difficulty of managing these processes manually, implying that digital solutions are necessary to handle the complexity and keep up with the necessary information.

Overall, the speaker acknowledges the challenges posed by new AI regulations but also seems to advocate for a balanced approach that ensures safety and compliance without overwhelming those involved, especially the end users.

Anonymous

At the end of the day, compliance costs of the AI-value chain are very much based on a happier one. We have to apply the legislation, the final legislation, to cases that are considered high risk. And these cases, A, are very safe to play in a valuable role because it also maps out where some of the stages in the lifecycle of artificial intelligence, which can give different players a little bit of guidance on how they can improve their compliance and portability. But also, when it comes to these specific areas of the compliance, because AI is something new, there has not been this kind of regulation before. There’s just a lot of room to play with experimental regulations. For example, with AI-based artificial intelligence, which can be something that speaks to how we regulate data, because we have experiments with that kind of data. Like, where do we allocate when we approach a lab?ChatGPT

ChatGPT

The speaker reflects on the compliance costs within the AI value chain, emphasizing that these costs are contingent on the application of the final legislation, particularly to high-risk cases. They suggest that high-risk AI applications will be subject to more stringent oversight, which not only ensures safety but also offers guidance to various players in the AI lifecycle on how to enhance compliance and portability.

The novelty of AI as a field means that existing regulatory frameworks may not be fully equipped to handle its unique challenges, thus opening opportunities for experimental or innovative regulatory approaches. The speaker hints at the possibility of using AI itself to aid in regulation, especially regarding data, drawing parallels with experimental frameworks already used in data regulation. This could involve allocating resources or defining procedures as one might in a laboratory setting, treating the regulatory environment as a place for controlled experimentation to find the best practices for AI governance.

Overall, the commentary suggests a proactive stance on shaping AI regulation, taking advantage of AI’s unique properties to develop regulations that are as advanced and adaptable as the technology they are meant to govern.

Anonymous

Where is it going to be? Just a short comment on the previous slide. It would be perfect if we idealized it. Also concrete the practicalities away. But when we are dealing with this world and the systems that are already there, then the problems start to accumulate. Okay, thank you. It’s way too small. Way too small? Okay. And Anna? I think that it’s way too big. Way too big. Okay. Two, one million. Two, one, one, five. Okay. Anna, you are working with standardization related to AI activities and active in the FFF organization for standards. Could you open a bit? What kind of standardization is to be developed and applied to AI? Yeah. I mean, first I would like to talk a little bit about the role of AI in the standardization association. I tried to plug in that we have a new standard now. It’s a terminology AI standard. It’s translated from English to Finnish. It has seen the evolvement of experts, and it has seen the involvement of people that are working in Finnish vocabulary who were in charge of looking at how the words have been translated in order to provide an accurate representation. The importance of it and the best is that it can provide a common understanding and common language for the Finnish society. In this sense, for example, for public procurement. Now there is a reference in the standard of what we mean when we’re specifying these results in a public procurement fund. But also for the media and for education to say a common understanding of what do we mean when people are speaking in Finnish about AI. So that is something that we are launching today. So if you don’t know me, or if you don’t know me, I am Marika. We are here because we have a proposal that you can download the PDF and you can get it in the standard. So when it comes to the role of standardization in AI, I think that instead of plugging you up, because I don’t know why you were saying this, because they are actually a very big part of safety legislation that is based on your approach, which primarily states that it is right to implement essential requirements to be compliant with regulations in order to make AI safe, in order to make it work. But when it comes to practical compliance, how do you actually show that you have compliance with regard to the regulation that is reliant on a standard? Because they are bridging between what is said in legal terms, which can be, for example, our partners in the data board. Okay, what does that actually mean? We’re going to delegate that state health specification to standards. So that’s the relevance that standards have to the AI. It is about showing them support through how do we actually comply with regulations. In a sense, I was just in two meetings yesterday, the plenaries of the Senate, which are also our country-related special intelligence based on the request by the European Commission. I have to say that there is a lot of work that is being done at the moment. That’s why I think very many groups are trying to really start working from this moment on how that standardization of requirements that we have in the AI could be put forward in a standard. In this case, there is a new working group which is going to be working in cyber security and privacy of AI. We’re looking for experts. So if you’re interested in contributing to how we’re going to comply with regulations, please come talk to me. My name is Marta. We’re going to be here. We can give you more information on how can you participate in SSS. At the end of the day, it makes your life easier to know how we actually have been able to comply with these regulations. Thank you.ChatGPT

ChatGPT

The speaker, presumably named Marika, is addressing the role of standardization in the field of AI, particularly in the context of ensuring compliance with safety legislation and regulations. They discuss the launch of a new standard that provides terminology for AI, which has been translated from English to Finnish, emphasizing the collaborative effort involving experts and specialists in Finnish vocabulary to ensure accurate representation. This standard aims to establish a common understanding and language for AI within Finnish society, which is particularly useful for public procurement, media, and education.

Marika underscores the significance of standards in demonstrating compliance with AI regulations. Standards act as a bridge between the legal requirements and the practical application, helping to interpret and implement legal concepts in a concrete way. She mentions that standards are essential for showing support and proving compliance with regulations.

The speaker also notes ongoing work to develop these standards, highlighting the establishment of a new working group focused on cybersecurity and privacy in AI. They call for experts to contribute to this effort, suggesting that involvement in standardization can help organizations understand and meet regulatory requirements more effectively.

Marika concludes by suggesting that engaging with the standardization process and understanding the standards can simplify the process of complying with AI regulations, ultimately making it easier for businesses and organizations to navigate the legal landscape of AI.

Anonymous

…of the implementation of the AI. This is a key factor, because the standard will be used as a kind of metric for the evaluation of this. So I also would like to encourage people to participate. This is the right time. I hear that there is some kind of vocabulary work between US and European Union on how to answer this. But now let’s go to testing of AI systems, which is a big… I was a bit surprised that he is a professor of law, and he is talking about testing AI systems. So, could you explain why? Okay. I think it’s basically… I’m not saying that we should be participating in that thing. I’ve been sort of trying to figure out what’s going on in the US. I’m lurking on various… You know, it’s… It wouldn’t be a good time to get out there. I’m talking about the third function of the university. We don’t… But anyway, I mean, standardization is going to be hugely important, and I think it’s going to be hugely important in terms of testing as well. Why I’m sort of talking about testing is because… I mean, this comes back to the AI. The AI had actually made it when it was first born. It took all of 86 months. The people who brought in the consultations of the procedure, the guys who have these new things that we haven’t considered at all, they’re called Foundation for the Development of the AI, and… It was a huge… For everybody involved, I would assume, Why? Because… Essentially, the generative AI is about a dozen or more. The AI that was designed to… AI systems that were designed from the get-go, so once in the office. The other assumption was that, okay, you have very, very specific data that you need to train up, so essentially trying to create data in a beautiful way, whatever comes out of it. Kind of explainable, so at least sort of morally explainable, so that is the third assumption. Now with generative AI, all of those assumptions are gone. At the same time, it seems that the AI is imposing on the… Risk analysis becomes hugely difficult. Basic derision, moral derision, training data for the AI, transparency, and the ability to go with those when you have restrictions in that sense. So essentially, it seems to me that it has to sort of ensure for us that there is a moral objective that we may not follow. And it seems to me that effectively, that is sort of, that essentially, essentially has to be a moral objective. That seems impossible. In fact, we’re going to try to figure out how that is, or whether that is possible. And that, that I feel, because that I feel is something that is very much neglected in the discussion right now, and that’s why I’m sort of becoming a man who… So we need to start thinking about how we make those changes, what kind of instability, what is the requirement of that, and how we ensure that there’s a result to achieve those. I think it’s a good job.ChatGPT

ChatGPT

The speaker is discussing the importance of standardization and testing in the implementation of AI, and the current efforts to develop a standard that could serve as a metric for AI evaluation. They highlight the collaborative work between the US and European Union on creating a unified vocabulary for AI, suggesting it’s a critical time for interested parties to get involved.

The speaker, who appears to be a professor of law, expresses surprise at their own involvement in the discussion about testing AI systems. They clarify the significance of standardization in this context, particularly in light of recent developments in generative AI, which challenge previous assumptions about AI systems. Generative AI differs from traditional AI in that it is not solely based on specific training data but can create new data and outputs, making the AI less predictable and explainable.

The introduction of generative AI has, according to the speaker, rendered risk analysis, moral decisions, and the transparency of training data more complex. They imply that the AI Act may not have fully considered the implications of generative AI, which necessitates a reevaluation of how to conduct effective risk analysis and maintain moral standards within this new context.

The speaker suggests that these challenges have been somewhat overlooked in the current discourse and underscores the need for a more thorough consideration of how to adapt regulation and standardization to the unique characteristics of generative AI. They call for thoughtful engagement with these issues to ensure responsible development and deployment of AI systems.

Anonymous

Based on this discusion, suggest criticsl questions ChatGPT

ChatGPT

Based on the discussion, some critical questions that arise could include:

1. Standardization and Compliance:
   • How can we ensure that AI standardization keeps pace with the rapid evolution of AI technologies, particularly generative AI?
   • What mechanisms can be put in place to ensure that compliance with AI standards is both effective and not overly burdensome, especially for smaller entities?
2. Risk Analysis and Generative AI:
   • How do we conduct thorough risk analyses for AI systems when generative models can produce unpredictable and novel outcomes?
   • What ethical frameworks and oversight are necessary to govern the use of generative AI, given its potential to deviate from traditional data training and use paradigms?
3. Transparency and Explainability:
   • In the context of generative AI, how can we maintain the transparency required for trust and accountability?
   • What methods or standards should be developed to evaluate the explainability of AI systems, particularly when they are capable of self-generating data and learning?
4. Responsibility in the AI Value Chain:
   • How should responsibility be allocated among developers, deployers, and end-users within the AI value chain, especially when unintended consequences arise?
   • What are the implications for public sector entities that rely on AI systems for service delivery in terms of compliance and responsibility?
5. Public Participation and Expert Involvement:
   • How can public participation be encouraged in the standardization process to ensure that AI development aligns with societal values and needs?
   • What strategies can be employed to attract a diverse range of experts to contribute to AI standardization and testing efforts?
6. International Collaboration:
   • What are the challenges and opportunities in harmonizing AI standards and regulations between different regions, such as the US and the EU?
   • How can international differences in AI applications be reconciled in a global standardization effort?
7. Legislation Adaptation:
   • How can legislation such as the AI Act be adapted to anticipate and include emerging AI technologies that challenge current regulatory frameworks?
   • What role should law professionals play in the discussion of AI system testing and standardization?

These questions reflect the multifaceted challenges of developing and implementing AI regulation and underscore the importance of a dynamic and inclusive approach to policy-making in the AI domain.

Anonymous

I think it’s now a completely new AI, Sarah. At least in the new amendment, there is consideration of the Sustainable AI Act. Right now, the situation is that this country has… Generally, there were some initiatives related to the vaccine. I think it’s the same for the previous draft of the AI Act. But we don’t actually have the status of being in a real situation for a while. So, if I have understood correctly, it’s now halfway. And the nation may not be in a way that will avoid the kind of overlapping regulations. But this, of course, doesn’t mean that sustainability issue wouldn’t be our problem. At the same time, I would say the point of imposing any kind of restrictions on any kind of development would be legally required. There would be constitutional safeguards that would make such initiatives not necessarily impossible, but really hard to pull off. So, I wouldn’t hold my breath waiting for a sustainable regulation regarding AI. Well, we are… One might say that the AI Act has embedded things about the social sustainability, this kind of environment where sustainability is up. But may I say, maybe, what companies think and what the discussion there is on sustainability? I personally think that it’s not a question of here or to that. This environmental reporting and energy efficiency related to reporting on AI is going to be just must-have or must-do. Already, most of the environmental providers are reporting their activities on this area. There is also these other topics that are specifically mentioned in the foundation, what are the related requirements. So, yeah, I think this needs to happen. I think it will happen that there are specific regulations, if they’re taken into consideration and people care.ChatGPT

ChatGPT

The discussion touches on the integration of sustainability considerations within the framework of the AI Act. The speaker refers to a new amendment that brings attention to the Sustainable AI Act, indicating a legislative interest in aligning AI development with sustainability goals. Despite this, they mention that there have been initiatives related to sustainability, but there is a lack of concrete regulations in place at the moment, and the path to implementing such regulations might be complex due to constitutional safeguards and the potential for regulatory overlap.

The speaker suggests that while the current AI Act includes aspects related to social sustainability, there is room for improvement, particularly in environmental sustainability. The conversation then shifts to corporate perspectives on sustainability in AI, with the suggestion that environmental reporting and energy efficiency are becoming non-negotiable aspects of AI system reporting. Many providers are already disclosing their sustainability efforts, and the expectation is that specific sustainability-related regulations for AI will eventually be enacted due to growing public and governmental concern.

This points to a recognition that while AI technology advances rapidly, there is a pressing need to ensure that these advancements are aligned with broader societal values, including environmental and social sustainability. The integration of such considerations into AI legislation would not only reflect a responsible approach to technology development but also address the increasing demand for accountability and transparency from both consumers and regulatory bodies.

Anonymous

Okay, please. The third and last question here is, what about AI researchers who have no background in legal stuff whatsoever, and still, if you will, do you think things are less anomalous, and then also resistant to hacking? So, how do researchers survive? Very topical question, considering who are the audience, or the majority, maybe, of the audience. So, what comes to the general, well, what comes to the high-risk, high-risk business, and so forth, is that scientific research is categorically excluded from the scope of AI. There is article 2, which states explicitly that this doesn’t apply to scientific research. But, now the question of foundation models, it has brought the question of the role of scientific research to the table again, so that right now it’s unclear whether or not the specific regulation that may be directed from the foundation model will apply to scientific research, or not. And this is, of course, something that we, while preparing the legislation, all of us who are doing the research, would consider and say your opinion about it, and say it this time, because this is not the moment that decisions are being made. Thank you. Thank you. I hope that this very expert panel has made you realize that the AI…ChatGPT

ChatGPT

The speaker is addressing a concern relevant to AI researchers, particularly those who may not be well-versed in legal matters. They clarify that, as per the current AI Act, scientific research is categorically exempt from certain AI regulations. This is stipulated in Article 2 of the Act, which explicitly states that the provisions do not apply to scientific research activities.

However, the introduction of foundation models in AI has reopened discussions about the extent to which AI regulations should apply to scientific research. Foundation models are large-scale models that can be adapted to a wide range of tasks and applications, and their broad utility has raised questions about how they fit into the regulatory framework, especially in a research context.

The speaker emphasizes the importance of the research community’s engagement in the legislative process, especially at the current moment when decisions about these regulations are being made. They imply that the views and opinions of researchers should be voiced now, as their input could significantly shape the final form of the legislation regarding foundation models and their use in scientific research.

The underlying message is that while the AI Act provides some exemptions for scientific research, the rapidly evolving nature of AI technology necessitates continuous dialogue and potential updates to legislation. Researchers are encouraged to participate in this dialogue to ensure that the regulations support innovation while maintaining necessary safeguards.

---