Kabir Singla’s Post

What Does AI Transparency Look Like? It Depends on Your Industry If you measured it today, how would your business score on AI transparency? Whether you’re using artificial intelligence (AI) to recommend products, develop drugs, or screen job applicants, you’re entrusting it with data without fully understanding how it makes decisions. How can you better understand the logic behind the AI black box, and do a better job of disclosing when — and how — AI is being used?   Lack of transparency is creating a trust gap between consumers and brands. The most recent State of the AI Connected Customer report found that just 42% of customers trust businesses to use AI ethically, down from 58% in 2023. What would make consumers trust AI? By far, it’s more transparency about how AI is used, and the benefits it can create. You can take steps today to make sure you’re being open about how your company is using AI. State of the AI Connected Customer Insights from more than 16,000 consumers and business buyers finds trust is at its lowest point in recent years. Advances in AI make earning that trust more critical than ever. Knowing where customer’s pain points are is key to succeeding with AI. Get the free report ‘Companies treat their LLMs as trade secrets’ Transparency in AI has always been important, but came to the fore with the October publication of the Foundation Model Transparency Index. The research, led by Stanford University, found all 10 major foundation model companies, including Google and OpenAI,  were “lacking” in transparency. And that’s being charitable. When evaluated across 100 aspects of transparency, including the use of copyrighted data and harmful data filtration, on a 100-point scale, the highest-scoring company received a 54. Large language models (LLMs) are, by their very nature, opaque. Their scale and complexity make it hard to completely understand their decision-making process. In fact, because these models calculate probabilities rather than certainties, you can ask an LLM the same question 10 times and get 10 different responses. At the same time, according to the Prompt Engineering Institute, “companies treat their LLMs as trade secrets, disclosing only limited information about model architecture, training data, and decision-making processes. This opacity prevents independent auditing of the systems for biases, flaws, or ethical issues.”  There’s not much independent auditing happening, anyway. It’s a new practice that, save for outliers like New York City, isn’t mandated by any local, state, or federal authority. But you don’t need to wait for a mandate to implement transparency protocols of your own. Here are some transparency in AI issues that various industries are facing, and tips on how to address them. Read more about transparent and trusted AI How the Atlas Reasoning Engine Powers Agentforce How ‘Human’ Should Your AI Agent Be? In Short, Not Very Ethical hacking practices...

What Does AI Transparency Look Like? It Depends on Your Industry If you measured it today, how would your business score on AI transparency? Whether you’re using artificial intelligence (AI) to recommend products, develop drugs, or screen job applicants, you’re entrusting it with data without fully understanding how it makes decisions. How can you better understand the logic behind the AI black box, and do a better job of disclosing when — and how — AI is being used?   Lack of transparency is creating a trust gap between consumers and brands. The most recent State of the AI Connected Customer report found that just 42% of customers trust businesses to use AI ethically, down from 58% in 2023. What would make consumers trust AI? By far, it’s more transparency about how AI is used, and the benefits it can create. You can take steps today to make sure you’re being open about how your company is using AI. State of the AI Connected Customer Insights from more than 16,000 consumers and business buyers finds trust is at its lowest point in recent years. Advances in AI make earning that trust more critical than ever. Knowing where customer’s pain points are is key to succeeding with AI. Get the free report ‘Companies treat their LLMs as trade secrets’ Transparency in AI has always been important, but came to the fore with the October publication of the Foundation Model Transparency Index. The research, led by Stanford University, found all 10 major foundation model companies, including Google and OpenAI,  were “lacking” in transparency. And that’s being charitable. When evaluated across 100 aspects of transparency, including the use of copyrighted data and harmful data filtration, on a 100-point scale, the highest-scoring company received a 54. Large language models (LLMs) are, by their very nature, opaque. Their scale and complexity make it hard to completely understand their decision-making process. In fact, because these models calculate probabilities rather than certainties, you can ask an LLM the same question 10 times and get 10 different responses. At the same time, according to the Prompt Engineering Institute, “companies treat their LLMs as trade secrets, disclosing only limited information about model architecture, training data, and decision-making processes. This opacity prevents independent auditing of the systems for biases, flaws, or ethical issues.”  There’s not much independent auditing happening, anyway. It’s a new practice that, save for outliers like New York City, isn’t mandated by any local, state, or federal authority. But you don’t need to wait for a mandate to implement transparency protocols of your own. Here are some transparency in AI issues that various industries are facing, and tips on how to address them. Read more about transparent and trusted AI How the Atlas Reasoning Engine Powers Agentforce How ‘Human’ Should Your AI Agent Be? In Short, Not Very Ethical hacking practices...

𝗚𝗮𝗿𝗯𝗮𝗴𝗲 𝗜𝗻, 𝗚𝗮𝗿𝗯𝗮𝗴𝗲 𝗢𝘂𝘁 𝗶𝗻 𝗔𝗜: 𝗔 𝗦𝘁𝗮𝗿𝗸 𝗪𝗮𝗿𝗻𝗶𝗻𝗴 𝗳𝗼𝗿 𝘁𝗵𝗲 𝗘𝗿𝗮 𝗼𝗳 𝗚𝗲𝗻𝗔𝗜 𝗮𝗻𝗱 𝗔𝗴𝗲𝗻𝘁𝗶𝗰 𝗔𝗜 The aphorism “Garbage In, Garbage Out” (GIGO) highlights the critical link between input quality and output reliability. In artificial intelligence (AI) and machine learning (ML), this principle defines system efficacy. As Generative AI (GenAI) rises and Agentic AI looms, GIGO’s risks intensify. 𝗧𝗵𝗲 𝗘𝘀𝘀𝗲𝗻𝗰𝗲 𝗼𝗳 𝗚𝗜𝗚𝗢 𝗶𝗻 𝗔𝗜 𝗮𝗻𝗱 𝗠𝗟 AI systems reflect the quality of their training data. Biases, inaccuracies, or irrelevancies in datasets lead to flawed outputs. For instance, biased hiring algorithms reinforce societal inequities, while flawed predictive analytics yield erroneous forecasts. Traditional ML relies on structured data, whereas GenAI uses vast unstructured datasets. This raises stakes as flawed data can lead to harder-to-detect creative misinterpretations. 𝗚𝗲𝗻𝗔𝗜 𝗮𝗻𝗱 𝗔𝗺𝗽𝗹𝗶𝗳𝗶𝗲𝗱 𝗥𝗶𝘀𝗸𝘀 Generative AI tools like ChatGPT and DALL-E democratize AI but inherit vulnerabilities from their training data. Internet-scale datasets may embed biases, misinformation, and harmful content, producing outputs that amplify such issues. Risks include:  - 𝗕𝗶𝗮𝘀 𝗔𝗺𝗽𝗹𝗶𝗳𝗶𝗰𝗮𝘁𝗶𝗼𝗻: Reinforcing stereotypes in creative outputs.  - 𝗠𝗶𝘀𝗶𝗻𝗳𝗼𝗿𝗺𝗮𝘁𝗶𝗼𝗻 𝗦𝗽𝗿𝗲𝗮𝗱: Generating convincing but false narratives.  - 𝗨𝗻𝗳𝗼𝗿𝗲𝘀𝗲𝗲𝗻 𝗖𝗼𝗻𝘀𝗲𝗾𝘂𝗲𝗻𝗰𝗲𝘀: Misuse due to unpredictable outputs. 𝗡𝗮𝘃𝗶𝗴𝗮𝘁𝗶𝗻𝗴 𝘁𝗵𝗲 𝗚𝗜𝗚𝗢 𝗖𝗵𝗮𝗹𝗹𝗲𝗻𝗴𝗲 Mitigating GIGO in AI demands:  - 𝗗𝗮𝘁𝗮 𝗤𝘂𝗮𝗹𝗶𝘁𝘆 𝗔𝘀𝘀𝘂𝗿𝗮𝗻𝗰𝗲: Curate diverse, accurate datasets and clean data rigorously.  - 𝗧𝗿𝗮𝗻𝘀𝗽𝗮𝗿𝗲𝗻𝗰𝘆 𝗮𝗻𝗱 𝗘𝘅𝗽𝗹𝗮𝗶𝗻𝗮𝗯𝗶𝗹𝗶𝘁𝘆: Build interpretable, auditable systems to identify errors.  - 𝗘𝘁𝗵𝗶𝗰𝗮𝗹 𝗢𝘃𝗲𝗿𝘀𝗶𝗴𝗵𝘁: Implement governance ensuring AI aligns with societal values.  - 𝗖𝗼𝗻𝘁𝗶𝗻𝘂𝗼𝘂𝘀 𝗠𝗼𝗻𝗶𝘁𝗼𝗿𝗶𝗻𝗴: Detect and correct flawed outputs in real time.  - 𝗖𝗿𝗼𝘀𝘀-𝗗𝗶𝘀𝗰𝗶𝗽𝗹𝗶𝗻𝗮𝗿𝘆 𝗖𝗼𝗹𝗹𝗮𝗯𝗼𝗿𝗮𝘁𝗶𝗼𝗻: Unite technologists, ethicists, and policymakers to tackle challenges. 𝗔 𝗖𝗮𝗹𝗹 𝘁𝗼 𝗔𝗰𝘁𝗶𝗼𝗻 As technological stewards, we must tackle GIGO’s heightened risks in AI. The rise of GenAI and Agentic AI demands vigilance. By prioritizing data integrity, fostering accountability, and upholding ethical practices, we can harness AI’s potential responsibly. What do you think? Let's share your experience in the comment section below. #dataquality #artificialintelligence #garbageingarbageout

10 Ways Deepseek AI is Outperforming the Competition (And Making Everyone Else Sweat a Little) The AI world is no stranger to bold claims, but Deepseek AI has quietly been out here doing its thing – and doing it better. Here’s how it’s giving other models a serious case of performance anxiety: 1. Semantic Search Precision: While others are still figuring out what we mean by “search this, not that,” Deepseek AI is already serving answers that feel like it read your mind. Magic? Nope, just better algorithms. 2. Multilingual Prowess: Struggling with low-resource languages? Not Deepseek. It speaks your language (and probably your neighbor’s), all with the finesse of a local. 3. Real-time Adaptation: While some models take forever to learn, Deepseek is the kid in class who always has their homework done early – and it’s flawless. 4. Data Efficiency: Forget the data gluttons of AI. Deepseek achieves stellar results on a diet, proving less is more (except for pizza, of course). 5. Energy Efficiency: Not only is Deepseek smart, but it’s also green – consuming less energy while making others look like gas guzzlers in comparison. 6. Multimodal Mastery: Text, image, video? Deepseek doesn’t just dabble; it excels. It’s like the multitasker who handles spreadsheets, presentations, and coffee runs without breaking a sweat. 7. Customizability: Deepseek AI lets businesses shape it to their unique needs. It’s the Lego of AI models – build it your way, and it always fits perfectly. 8. Error Detection & Recovery: Other models trip over errors and face-plant; Deepseek calmly fixes its mistakes like a seasoned professional who’s been there, done that. 9. Privacy Rockstar: In a world full of data breaches and sketchy practices, Deepseek is the saint of AI – compliant, secure, and trustworthy. 10. Scalability Extraordinaire: Whether it’s helping a startup or a corporate behemoth, Deepseek is the go-to model that doesn’t flinch, no matter the size of the task. In a landscape crowded with AI models making a lot of noise, Deepseek AI is quietly setting benchmarks. A little humility, a lot of capability – that’s the kind of competition we can all aspire to. What’s your take on the evolving AI race? Drop your thoughts below! #AIInnovation #DeepseekAI #SemanticSearch #MultilingualAI #DataEfficiency #EnergyEfficientAI #RealTimeAdaptation #ScalableAI #CustomizableAI #PrivacyFirstAI #MultimodalAI #NextGenAI #ArtificialIntelligence #FutureOfAI #SmartAI #AIForBusiness #AdvancedAI #SustainableAI #AITrends #AIExcellence

I hear a lot of talk about how AI just needs more data to improve. But that logic is flawed. The real problem isn’t the amount of data—it’s the lack of reasoning. That’s where AGI makes all the difference Data is the lifeblood of traditional AI, but it’s also its greatest limitation. Models like GPT-4 rely on massive datasets to learn patterns and generate outputs, but these datasets are inherently biased—they reflect the perspectives, gaps, and errors of the humans and systems that created them. This leads to several critical issues: 1️⃣ Bias In, Bias Out: If the training data contains bias (e.g., unequal representation of demographics, skewed legal precedents, or outdated medical guidelines), the AI will perpetuate and amplify these biases in its responses. 2️⃣ Context Blindness: Traditional AI can’t reason beyond its training. It’s bound by the scope of the data it was trained on, which means it struggles with nuance, edge cases, or novel scenarios. 3️⃣ Overfitting to Patterns: AI trained on large datasets often memorizes patterns without truly understanding them, making it vulnerable to errors when the input doesn’t match the expected structure. Why AGI is the Game-Changer AGI (Artificial General Intelligence) operates fundamentally differently. It doesn’t depend on memorizing vast amounts of data to perform well; instead, it can reason, adapt, and derive insights in real time, making it far less prone to bias. Here’s why: 1️⃣ Reasoning Over Memorization: AGI can analyze situations logically, even when data is incomplete or biased. It applies principles and context to arrive at solutions, much like a human would. 2️⃣ Dynamic Learning: Unlike static models, AGI learns and adapts on the fly. For example, if it encounters a new precedent in a legal case, it can integrate that into its reasoning without needing retraining. 3️⃣ Cross-Domain Understanding: AGI draws on knowledge across multiple fields, ensuring it’s not trapped within the limitations of one dataset or domain. This makes it incredibly effective for solving complex, interdisciplinary problems. 4️⃣ Ethical Decision-Making: By reasoning through ethical principles rather than relying on biased datasets, AGI can align better with human values and fairness standards. The Future: Moving Beyond Data Dependence Traditional AI systems are powerful but inherently constrained by the quality of their training data. AGI, with its reasoning-first approach, represents the next leap forward—one that reduces the impact of bias, adapts dynamically to new situations, and empowers humans to solve problems in ways that static models simply cannot. This shift isn’t just technological; it’s philosophical. It moves AI from being a tool that replicates past patterns to a partner that reasons alongside us, helping to shape a more fair and innovative future.

Here is an excellent post by my colleague, Kevin Welsh. For anyone interested in AI, this is a quick and worthwhile read.

AI's "Reviewer Paradox": Most discussions around AI use cases focus on what models can do, but it's just as important to consider what the users (aka humans in the loop) can do. The users reviewing the work need to have a good baseline expertise to judge whether the output is good, and this can create a paradox that limits AI's usefulness in some cases. For example, most people can tell how good a blog post or a graphic is without knowing how to create one or articulate "why" it's good. This explains why AI tools for generating simple content took off so quickly. But that's not the case for, say, legal documents. AI can probably draft a perfect contract, but you wouldn't know that unless you also have the required expertise. This is why AI makes more sense as a "co-pilot" for experts in use cases requiring deep reviewer knowledge. And on top of that, you also have to consider another dimension — the effort it takes to review the work. Sometimes AI adds a review step to existing workflow, making things more tedious overall. Take AI SDR for example: The promise is that you can send out personalized, well-researched cold emails at scale. But since every email is different and a mistake could tank a potential deal or get you fired, you better review the drafts before they get sent out. This isn't something sales people had to do previously, and it might end up canceling out the productivity gains. Finally, the toughest category is use cases that require both high reviewer expertise + high review effort. A good example here is data analysis: How do you know the answer is correct? You need someone with data expertise who understands the business context to review: - Whether the question itself is meaningful (often not the case) - Whether the query reflects the question (there could be more than 1 correct answer) - Whether the answer is correct (Sometimes the raw data is already wrong or biased) All things considered, it might be more efficient to just have the expert do the work. You could argue that these challenges will slowly disappear as AI models improve, and it's not inherently different from reviewing a human's work or hiring an expert, which is fair. But let's be honest — having a human to hold responsible (or blame 😅) matters way more than we admit. This part of human nature probably won't change anytime soon, if ever. And if you think your AI product or feature can take on that responsibility, just know that people will have a much, much higher bar for what "good" looks like. #product #ai #llm

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

𝗨𝗻𝗱𝗲𝗿𝘀𝘁𝗮𝗻𝗱𝗶𝗻𝗴 𝗘𝘅𝗽𝗹𝗮𝗶𝗻𝗮𝗯𝗹𝗲 𝗔𝗜 (𝗫𝗔𝗜): 𝗪𝗵𝘆 𝗜𝘁 𝗠𝗮𝘁𝘁𝗲𝗿𝘀 𝗶𝗻 𝗮 𝗪𝗼𝗿𝗹𝗱 𝗼𝗳 𝗔𝗜 Artificial Intelligence is rapidly transforming industries, from healthcare to finance, by making decisions that affect our lives in significant ways. But as AI becomes more complex, there's a growing need to understand how these decisions are made. This is where 𝗘𝘅𝗽𝗹𝗮𝗶𝗻𝗮𝗯𝗹𝗲 𝗔𝗜, 𝗼𝗿 𝗫𝗔𝗜, comes into play. 𝗪𝗵𝗮𝘁 𝗶𝘀 𝗘𝘅𝗽𝗹𝗮𝗶𝗻𝗮𝗯𝗹𝗲 𝗔𝗜 (𝗫𝗔𝗜)? Explainable AI (XAI) refers to methods and techniques that make the decision-making process of AI systems transparent and understandable to humans. Unlike traditional AI, where the "how" often remains a mystery, XAI aims to clarify the reasoning behind AI's outputs. 𝗪𝗵𝘆 𝗗𝗼𝗲𝘀 𝗫𝗔𝗜 𝗠𝗮𝘁𝘁𝗲𝗿? • 𝗕𝘂𝗶𝗹𝗱𝗶𝗻𝗴 𝗧𝗿𝘂𝘀𝘁 𝗮𝗻𝗱 𝗧𝗿𝗮𝗻𝘀𝗽𝗮𝗿𝗲𝗻𝗰𝘆: When AI can explain its reasoning, users are more likely to trust its decisions. This is crucial in sectors like healthcare and finance, where decisions can have significant consequences. • 𝗥𝗲𝗴𝘂𝗹𝗮𝘁𝗼𝗿𝘆 𝗖𝗼𝗺𝗽𝗹𝗶𝗮𝗻𝗰𝗲: Laws like the GDPR require organizations to provide explanations for automated decisions affecting individuals. XAI helps companies meet these legal obligations. • 𝗜𝗱𝗲𝗻𝘁𝗶𝗳𝘆𝗶𝗻𝗴 𝗮𝗻𝗱 𝗠𝗶𝘁𝗶𝗴𝗮𝘁𝗶𝗻𝗴 𝗕𝗶𝗮𝘀: XAI enables us to detect biases in AI models, promoting fairness and ethical use. For example, if a hiring algorithm is favoring certain demographics, XAI can help pinpoint the issue. • 𝗜𝗺𝗽𝗿𝗼𝘃𝗶𝗻𝗴 𝗔𝗜 𝗠𝗼𝗱𝗲𝗹𝘀: Understanding how AI makes decisions allows developers to fine-tune models for better accuracy and performance. 𝗛𝗼𝘄 𝗗𝗼𝗲𝘀 𝗫𝗔𝗜 𝗪𝗼𝗿𝗸? XAI employs various methods to make AI decisions interpretable: • 𝗙𝗲𝗮𝘁𝘂𝗿𝗲 𝗜𝗺𝗽𝗼𝗿𝘁𝗮𝗻𝗰𝗲: Highlights which input variables influenced the outcome the most. For instance, in a medical diagnosis model, patient age and specific symptoms might be identified as key factors. • 𝗠𝗼𝗱𝗲𝗹 𝗦𝗶𝗺𝗽𝗹𝗶𝗳𝗶𝗰𝗮𝘁𝗶𝗼𝗻: Simplifies complex models into more understandable versions without significantly losing accuracy. • 𝗩𝗶𝘀𝘂𝗮𝗹𝗶𝘇𝗮𝘁𝗶𝗼𝗻 𝗧𝗼𝗼𝗹𝘀: Uses graphs and charts to illustrate how inputs affect outputs, making complex data more digestible. • 𝗖𝗼𝘂𝗻𝘁𝗲𝗿𝗳𝗮𝗰𝘁𝘂𝗮𝗹 𝗘𝘅𝗽𝗹𝗮𝗻𝗮𝘁𝗶𝗼𝗻𝘀: Shows how slight changes in input could alter the decision, helping users understand the decision boundaries. 𝗧𝗵𝗲 𝗙𝘂𝘁𝘂𝗿𝗲 𝗼𝗳 𝗔𝗜 𝘄𝗶𝘁𝗵 𝗫𝗔𝗜: As AI continues to evolve, the importance of XAI will only grow. In a world where AI decisions impact everything from our health to our livelihoods, ensuring these decisions are not just accurate but also explainable is key to building trust. XAI bridges the gap between complex algorithms and human understanding, ensuring that we remain in control of the technologies shaping our future. #ExplainableAI #XAI #ArtificialIntelligence #AI #TechInnovation #Transparency #AIethics #TrustInTech

AI and ML are reshaping our world, but data bias poses a critical challenge. #DataBias #Ethics. Biased data can lead to unfair practices, misinformed decisions, and financial losses. From skewed hiring tools to inaccurate customer insights, the impact is far-reaching. Let's ensure our data is diverse, representative, and unbiased to drive innovation and growth. Detect, mitigate, and prevent bias to unlock the full potential of AI in a fair and ethical manner. #AI #DataEthics. https://lnkd.in/gd2c-eZE