DARIO DE ZOLT’s Post

Will AI help in medical tech development? What are your thoughts? Below are some notes from Sol-Millennium’s R&D head.

I had the opportunity to reflect on the role of human-centered AI thanks to a survey I participated in, Dr. Aung Pyae I appreciate the invitation! I consider that AI has the potential to transform industries and improve lives, but the real challenge lies in ensuring it aligns with human values and needs. Bias in data, lack of transparency, and the delicate balance between privacy and personalization remain significant barriers to truly human-centered AI systems. In my point of view, overcoming these challenges will require diverse, representative data, explainable models, and ethical frameworks that prioritize fairness, trust, and human autonomy. Collaboration, collaboration, and more collaboration among educators, technologists, ethicists, and policymakers! A human-centered AI system should be designed with the primary goal of enhancing human well-being, respecting human values, and being aligned with human needs and intentions. Here's what I believe makes an AI system human-centered: 1. User-Centric Design: The AI should prioritize the user's needs and experiences, ensuring that its interface and functions are intuitive, accessible, and easy to use for a broad range of individuals, regardless of their technical skills or background. 2. Ethical and Transparent Operation: A human-centered AI operates transparently, meaning its decision-making processes are explainable and interpretable to users. It avoids opaque "black-box" algorithms where users have no understanding or control over outcomes. Ethical considerations, like privacy, fairness, and avoidance of bias, should be central. 3. Supportive of Human Autonomy: Rather than replacing human judgment or action, a human-centered AI augments human decision-making. It empowers users by providing insights, guidance, or support, while allowing them to remain in control of the final decisions. 4. Emphasis on Collaboration: Human-centered AI systems are designed to work in partnership with humans, enabling collaboration where AI complements human strengths, such as creativity, empathy, or critical thinking. The goal is not to replace humans but to enhance their abilities. 5. Adaptability and Personalization: The AI should be able to adapt to individual preferences, learning from interactions to provide personalized experiences. This includes adjusting to various cultural, emotional, and contextual factors in a user's environment. 6. Trust and Reliability: Human-centered AI builds trust through consistent performance, safety features, and mechanisms that allow users to verify and understand the accuracy of the AI's actions. What do you think are the biggest hurdles to achieving human-centered AI? #HI4AI #human #HumanCenteredAI #AIethics #AIFuture #TrustworthyAI #AIInnovation #BiasInAI #AITransparency #AIForAll #education #studentfocus #AIliteracy #AIpolicies #education #highereducation

Use of AI. If the use of AI is a challenge for all because of the changes it causes, for Quality Assurance it is also questioning about the relevance and veracity of the information. Management traditionally rely on Quality Assurance to ensure the reliability of the data transmitted (e.g., to the authorities). For example if: - a report satisfies the acceptance criteria, - an action plan is relevant to the related non-compliance, - an instruction is consistent with its master procedure, - a measured value comply with a standard requirement, - an action plan is properly followed. AI can support these activities better than humans doing data comparison, bibliographic searchs or, for generative AI, documents production. But who is responsible if the AI is wrong, misinterprets, hallucinates, overreacts? When the police stop you in the wrong direction, you can’t blame it on your GPS . Similarly, you cannot justify yourself to the authorities saying : "we followed the recommendations of the AI to qualify a process that produces welds resistant over 0.5N/15mm (instead of the 1.2 N/15mm)" or "We do not understand that you reject our response to your inspection report: it was produced 100% by an AI, it is necessarily fair". AI will never be responsible, neither to management nor to the authorities. Would you have confidence in an organization that indicates "AI" as a management representative in the organizational chart and as an approver of the PSUR? Therefore, if AI can be used by QA to accelerate its activities, its use must be strongly mitigated: - Train staff to formulate requests correctly and be aware of risks, - Clearly define the scope of AI use for your activity, - Require the provision of references for each information communicated, - Check the relevance of references. It is the sharp control, for example: does the paragraph of the standard indicated by the AI to extract a value or information is relevant to the request made? - Check the relevance of the arguments. It is the extended control, for example: do the results indicated by the AI as meeting the objectives and acceptance criteria really meet it; will the action proposed by the AI really make it possible to correct the observed defect), - Monitor feelings about AI activities. First, your colleagues will not necessarily appreciate receiving daily automatic reminders from an AI in charge of monitoring an action plan, without being able to justify themselves against a real person. Second, management may be tempted to shorten all project timelines based on AI response times, without considering verification steps. Remember that not officially using AI in your organization (doing nothing) will not prevent your employees from using it on a daily basis to respond more quickly to your requests, without you necessarily knowing it, and sometimes without their awareness of the risks. AI is an incredible accelerator. Efficiency if used properly. Major problems if not.    

𝐓𝐡𝐞 𝐁𝐮𝐳𝐳 𝐀𝐫𝐨𝐮𝐧𝐝 𝐂𝐡𝐚𝐭𝐆𝐏𝐓 𝐢𝐬 𝐄𝐱𝐜𝐢𝐭𝐢𝐧𝐠, 𝐁𝐮𝐭 𝐃𝐨𝐧’𝐭 𝐎𝐯𝐞𝐫𝐥𝐨𝐨𝐤 𝐭𝐡𝐞 𝐑𝐢𝐬𝐤𝐬 𝐨𝐟 𝐆𝐞𝐧𝐞𝐫𝐚𝐭𝐢𝐯𝐞 𝐀𝐈 There’s no shortage of conversation about ChatGPT, but amid all the excitement, we must pay attention to the risks associated with generative AI, particularly large language models (LLMs). While LLMs have impressive capabilities, such as helping people communicate more effectively through writing, their ability to generate human-like text masks a significant flaw: they don’t truly understand the content they produce. This can lead to considerable risks, including: 1. 𝐇𝐚𝐥𝐥𝐮𝐜𝐢𝐧𝐚𝐭𝐢𝐨𝐧𝐬: LLMs can generate inaccurate information, a phenomenon referred to as "hallucinations." This happens because these models prioritize grammatically correct sentences over factual accuracy. Imagine a customer service chatbot providing incorrect information without any way to correct it this could severely damage your brand's reputation. 𝐒𝐨𝐥𝐮𝐭𝐢𝐨𝐧: Implement explainability. Integrate LLMs with knowledge graphs that clarify data sources and allow users to verify the accuracy of the information. 2. 𝐁𝐢𝐚𝐬: LLMs are trained on vast amounts of data, which can reflect and even amplify existing social biases. If left unchecked, their outputs can reinforce harmful stereotypes and discrimination. 𝐒𝐨𝐥𝐮𝐭𝐢𝐨𝐧: Foster a culture of awareness and regular reviews. Assemble diverse, multidisciplinary teams to work on AI development, and use these reviews to identify and correct biases in the models and organizational practices. 3. 𝐂𝐨𝐧𝐬𝐞𝐧𝐭: A significant portion of the data used to train LLMs lacks clear provenance or explicit consent. Were individuals aware their data was being used? Are there intellectual property violations at play? 𝐒𝐨𝐥𝐮𝐭𝐢𝐨𝐧: Emphasize review and accountability. Establish clear AI governance processes, ensure compliance with data privacy regulations, and provide mechanisms for individuals to understand and control how their data is used. 4. 𝐒𝐞𝐜𝐮𝐫𝐢𝐭𝐲: LLMs are vulnerable to malicious exploitation. Hackers could use them to spread misinformation, steal data, or promote harmful content. 𝐒𝐨𝐥𝐮𝐭𝐢𝐨𝐧: Education is key. We need to understand the strengths and weaknesses of this technology, including its potential for misuse. Train teams on responsible AI development and the security risks, as well as mitigation strategies. 𝐋𝐨𝐨𝐤𝐢𝐧𝐠 𝐭𝐨 𝐭𝐡𝐞 𝐅𝐮𝐭𝐮𝐫𝐞: The future of AI is bright, but caution is essential. We must promote a culture of responsible AI development that prioritizes: -𝐓𝐫𝐚𝐧𝐬𝐩𝐚𝐫𝐞𝐧𝐜𝐲: Clearly communicate the limitations of LLMs and the potential for errors. -𝐀𝐜𝐜𝐨𝐮𝐧𝐭𝐚𝐛𝐢𝐥𝐢𝐭𝐲: Define clear lines of responsibility for AI outputs and their consequences.

Are You on "𝗧𝗲𝗮𝗺 𝗛𝘂𝗺𝗮𝗻" or "𝗧𝗲𝗮𝗺 𝗨𝗻𝗰𝗼𝗻𝘁𝗿𝗼𝗹𝗹𝗲𝗱 𝗔𝗜 𝗙𝘂𝘁𝘂𝗿𝗲" ? 🤖❤️ If you're designing, building, or deploying AI, the question isn't just about what your AI can do. 𝗜𝘁’𝘀 𝗮𝗯𝗼𝘂𝘁 𝘄𝗵𝗼𝘀𝗲 𝘀𝗶𝗱𝗲 𝘆𝗼𝘂'𝗿𝗲 𝗼𝗻. AI’s core tech may seem neutral, but our choices shape its impact. Every decision you make about how your AI feature is design, built, tested, and deployed will reshape our future—for better or worse. Does your AI solution align with transparency? 🟢 𝗧𝗲𝗮𝗺 𝗛𝘂𝗺𝗮𝗻: Do you show users how AI is making decisions? Let them in on the process? 🔴 𝗧𝗲𝗮𝗺 𝗨𝗻𝗰𝗼𝗻𝘁𝗿𝗼𝗹𝗹𝗲𝗱 𝗔𝗜 𝗙𝘂𝘁𝘂𝗿𝗲I: Or, are you keeping them in the dark, with decisions locked in a black box of algorithms? (1 point Uncontrolled AI Future) What about fairness? 🟢 𝗧𝗲𝗮𝗺 𝗛𝘂𝗺𝗮𝗻: Do you actively audit your AI for bias, ensuring it treats everyone equally? 🔴 𝗧𝗲𝗮𝗺 𝗨𝗻𝗰𝗼𝗻𝘁𝗿𝗼𝗹𝗹𝗲𝗱 𝗔𝗜 𝗙𝘂𝘁𝘂𝗿𝗲: Or, do you let bias slide, hoping for the best? (1 point Uncontrolled AI Future) Then there's privacy. 🟢 𝗧𝗲𝗮𝗺 𝗛𝘂𝗺𝗮𝗻: Is user consent front and center? Are you safeguarding their data with the respect it deserves? 🔴 𝗧𝗲𝗮𝗺 𝗨𝗻𝗰𝗼𝗻𝘁𝗿𝗼𝗹𝗹𝗲𝗱 𝗔𝗜 𝗙𝘂𝘁𝘂𝗿𝗲 : Or are you playing fast and loose with their privacy? (1 point Uncontrolled AI Future) And finally, accountability. 🟢 𝗧𝗲𝗮𝗺 𝗛𝘂𝗺𝗮𝗻: Do you take responsibility for the decisions your AI makes? 🔴 𝗧𝗲𝗮𝗺 𝗨𝗻𝗰𝗼𝗻𝘁𝗿𝗼𝗹𝗹𝗲𝗱 𝗔𝗜 𝗙𝘂𝘁𝘂𝗿𝗲 : Or are you washing your hands, letting AI run without oversight? (1 point Uncontrolled AI Future) If your score isn't looking too great right now, it's not too late. Reassess. Redesign. Reimagine - NOW These are learning systems - we are on a short runway to when we wont be able to undo the learning As Cassie Kozyrkov, former Chief Decision Scientist at Google recently said "Trying to remove training data once it has been baked into a large language model is like trying to remove sugar once it has been baked into a cake. You’ll have to trash the cake and start over" We all have a choice. Will you be on 𝗧𝗘𝗔𝗠 𝗛𝗨𝗠𝗔𝗡 or part of an uncontrolled AI future? Let’s start designing AI for people, not just profit. Comment Below Are you on "Team Human!" 👇 Enjoy this? ♻️ Repost it to your network and follow Riley Coleman Signup for my newsletter https://lnkd.in/eTTHSK_K

ON POINT: AI & “the ‘RECOGNITIONS -- Addressed here is need of a “Middle Way,” a balance enabling new AI Growth-potential to exist protected from there being any opportunities for unregulated exploitations, and from there existing excessive restrictiveness. Easier said than done—, SO, what is it that could remain the most endangering obstacle? Surely it’s that part of humanity that’s eager for monopolization of power, and which includes persons enslaved to Greed, Fear, Hate, Vengeance, which for the rational AI user spells urgent need of, “SECURITY,” itself an evolving technology. How, then, can the Good Guys of the greater AI world stay ahead of the Bad Guys? Of course, for starters all Good Guy-players could become keenly aware of what a technologist might christen as, “the ‘RECOGNITIONS,’” largely a replay of earlier efforts but applied with lessons recently learned—, in effect, the unalterable truths to which attention must be paid to and that would serve as a launching pad for safest possible new AI apps going forward. Among examples: adherence to the fact that AI is a global phenom while it’s also a local phenom, the two avoiding ever being at severe odds. Surely wherever legal AI is under exploration or in operational mode the principles, goals, strategies and tactics employed need be similar so that broadest mutual understanding for operational harmony among legal AI users can continue to exist. Too, before launch of newly developed and regulated AI R&D systems the concerns re. Human and Robotics dominance“One-over-the-other” must be revisited and resolved. Too, governance and private sector AI developers ought be teamed from the Get-go for safe and secure information-sharing and R&D practices, with governance willing to take on various investment roles, but always according to accepted Protocol. Important, too, is acknowledging that ever-advancing AI could occur within numerous sub-sets, e.g., Higher education, Health Care, Business entities large and small, Banks, other finance realms, and national and global security and the roles enacted by intelligence agencies and military organizations. Within this alchemy of AI-hungry factions, hierarchies and rivalries could form, some attaining more value than other genre areas—, cliques could arise and dominate the regions that provide assets for AI R&D efforts. Here’s where governance/private sector teaming could assist AI Growth-potential by including in its “Middle Way”-Protocol constant prioritization, and re-prioritization, for who ought to get which piece of the pie and when, an irony that this process would be AI-generated. Yes, there’s probably more for “the ‘RECOGNITIONS,’” undoubtedly a platform of common sense imperatives for any new and rational AI take-offs that may not always defeat Bad Guy tactics but will, if set properly within applied SECURITY measures, manage to be steps ahead of them more times than not. . . ml

The Dark Side of AI 🤖 1. Autonomy and Control: 🏭 As AI systems become more autonomous, it is difficult ensuring that they remain under human control and are aligned with human values is a significant challenge. 2. Bias and Discrimination: 👨⚖️ AI systems can perpetuate and even exacerbate existing biases if they are trained on biased data sets. This is leading to unfair treatment and discrimination in critical areas like hiring, lending, and law enforcement. 3. Privacy Concerns: 👨💻 The vast amount of data required to train AI systems often includes personal information, raising significant privacy issues. 4. Job Displacement: 🏢 Automation driven by AI is also leading to job losses in certain sectors, creating economic and social challenges. As we stand on the cusp of a technological revolution, the allure of artificial intelligence (AI) is undeniable. The promise of AI to transform industries, enhance efficiencies, and create unprecedented opportunities is captivating. However, as we navigate this transformative landscape, it's crucial to acknowledge the shadows that accompany this bright new dawn. Imagine this: a company deploys an AI-driven hiring system to streamline their recruitment process. The system promises to identify the best candidates quickly and efficiently, saving time and resources. However, as the system starts making decisions, subtle patterns emerge. Candidates from certain demographic backgrounds consistently score lower. On investigation, it turns out that the AI was trained on historical hiring data that was inherently biased. Instead of eliminating human bias, the AI had unwittingly perpetuated it, leading to discrimination and unfair treatment. This scenario underscores a significant challenge with AI—bias and discrimination. AI systems, if not carefully designed and monitored, can amplify existing prejudices encoded in the data they learn from. It's a crude reminder that our AI systems are only as fair as the data and algorithms we create. Principles for Responsible AI 1. Fairness: ⚖️ We need to ensure that AI systems are unbiased and treat all individuals equitably. This involves using diverse and representative data sets and continuously monitoring for bias. 2. Human-Centric Design: 👥🤖 AI should augment human capabilities, not replace them. Design systems that support and enhance human decision-making. 3. Transparency: 🔍 AI decision-making processes should be explainable and understandable. Stakeholders should know how decisions are made and have the ability to question and challenge them. 4. Privacy and Security: 🔐 Protecting user data is paramount. Implementing robust data protection measures and ensure compliance with relevant regulations. #ResponsibleAI #EthicsInAI #AIFuture #TechForGood #AIethics #InnovationAndEthics

Companies are struggling with AI. Here are the challenges. As businesses begin to explore the potential of generative AI, they are confronted with a series of significant challenges. These obstacles are not just technical—they touch on data security, ethics, resources, and even company culture. Here’s what decision-makers should consider: 🟣 Safeguarding data privacy and security ↳ Generative AI can handle vast amounts of sensitive data, but this brings a heightened responsibility to ensure compliance with data protection regulations. ↳ Industries like healthcare and finance, where data confidentiality is paramount, must prioritize robust security measures to prevent breaches. 🟣 Ensuring data quality and minimizing bias ↳ The output of generative AI is only as good as the data it learns from. If historical data is biased or flawed, it can lead to biased results. ↳ Companies need to invest in high-quality, unbiased datasets, although achieving this remains a challenge. 🟣 Overcoming resource constraints ↳ The shortage of skilled personnel and the need for significant computational power are major hurdles, particularly for smaller organizations. ↳ Addressing these resource gaps might involve strategic partnerships or investing in talent development. 🟣 Achieving model interpretability and explainability ↳ Understanding how generative AI models arrive at their conclusions is crucial, especially in sectors that demand transparency and accountability. ↳ This calls for efforts to make AI systems more interpretable, which may also boost trust and adoption among employees. 🟣 Integrating AI seamlessly into workflows ↳ Embedding generative AI into existing business processes without disruption is easier said than done. ↳ Ensuring AI outputs are accurate, reliable, and align with business standards is essential, particularly for customer-facing applications. 🟣 Addressing ethical and societal implications ↳ The potential misuse of generative AI, like creating deepfakes or misinformation, raises serious ethical concerns. ↳ Leaders must also consider the broader societal impact, such as job displacement and the need for reskilling the workforce. 🟣 Fostering organizational readiness and AI literacy ↳ Resistance to AI adoption often stems from fear or misunderstanding. Building AI literacy within the organization can mitigate this. ↳ Establishing a clear strategic vision and governance framework is essential for guiding AI implementation and ensuring long-term success. ❓ Which of these challenges resonate with your experience? 👉 Follow me (Andreas Schwarzkopf) to stay ahead with AI.

88% of data leaders say their employees are using AI, regardless of whether or not their company has officially adopted genAI tools. If you can relate, it might be time for an AI acceptable use policy. As AI becomes more ubiquitous, staying ahead of how it's used – and ensuring that usage is responsible and secure – are critical for avoiding risks, penalties, and problems that can't be undone. Find out everything you need to know about AI acceptable use policies – and how to implement them – here: https://lnkd.in/e4D_55Mr #GenAI #DataSecurity #AcceptableUsePolicy #AUP

I’ve been thinking a lot about AI agents lately and all the hype surrounding them this year—there’s no denying their potential, but I can’t help feeling skeptical about just how much we should trust them. AI has been nothing short of a game changer. Over the last few years, we’ve seen tools made from large language models (LLMs) transform productivity—whether it’s generating content, analyzing data, or even writing code. As impressive as this is, I find myself questioning how much we can—and should—rely on agentic AI, especially when it comes to tasks that require action, like pushing code to production or making financial investments. In my role as a BI Engineer. Every day, I use AI tools to help with everything from writing scripts to analyzing datasets. They’re great for speeding up the process and offering new perspectives, but here’s the thing: I can’t imagine fully trusting them to implement anything on their own. These tools are helpful, sure—but what if the code they push has a bug? Or worse, what if it introduces a flaw that no one catches until it’s too late? The risks aren’t just technical. Imagine an AI placing the wrong trade in a high-stakes financial environment or applying the wrong logic in a critical system. Mistakes like these could cost companies millions—or worse, harm people directly. And the big question is: who’s accountable when that happens? AI systems don’t “own” their decisions, so it ultimately falls back on humans. But if humans weren’t involved in the process, how do you assign blame? That’s why, for me, human oversight is non-negotiable. AI can assist with routine tasks, but we still need someone to review and make sure things are actually correct. AI doesn’t understand business context, edge cases, or the nuance of ethical decision-making. It follows patterns and data—but we all know that data can be flawed. This doesn’t mean agentic AI is a bad idea. Far from it—it has the potential to supercharge productivity. But there’s a fine line between automation and autonomy, and right now, we need to be cautious about crossing it. Building in checks, reviews, and accountability structures is critical if we’re going to trust these tools with bigger responsibilities. So, where do we go from here? Should we push ahead and let AI take more control, or slow things down and make sure we have guardrails in place? It’s a conversation worth having—and one we need to figure out before agentic AI becomes the norm.