The Importance of Human Input in AI

Human judgment is the necessary complement to AI. High performance Humans + AI organizations require healthy networks connecting those who can best complement AI in a specific domain to the point where value is created. We had a wonderful structured conversation in the Humans + AI Community today including Marshall Kirkpatrick Bryan Williams Kanella Salapatas Dan Bashaw Dennis D Draeger that brought out many really powerful insights. It started with me sharing my current work on accelerated judgment development and calibrating trust in AI and rapidly built from there to the practical implications and implementation. Just a few of the many highlights: ⭐ Build networks of judgment People need to be domain experts to assess for any deep work whether the AI is providing good outputs and should be trusted or not. This means we need to make it easier to find the right human judges for the domain. Building on well-established organizational network practices, it can be exceptionally valuable to map and activate networks of humans who can calibrate trust, challenge outputs, and help others improve their judgment over time. ⭐ Teach people to challenge AI AI literacy is not just learning prompts or tools, but more and more learning to question outputs. AI far too often behaves like a “yes person”. This means one of the most important workforce capabilities is the habit of probing, testing, and pushing back on what it produces. ⭐ Psychological safety with AI We usually think about psychological safety in relation to managers or teams, but there is now a new issue: people may also need the confidence to challenge AI. Because AI can appear authoritative and hyper-informed, there is a real risk that people defer to it too quickly. We need to make sure that people don't defer to the "authority" of the machine, and challenge what it produces. ⭐ Tacit knowledge is becoming strategic We repeatedly returned to tacit knowledge as the place where human value increasingly resides. If AI absorbs more explicit, codified, procedural work, then the human edge lies in what is harder to formalize: opinion, intuition, context, pattern recognition, lived experience, and judgment in motion. We need to surface that, but in a way that respects and reinforces the value of the individual. If you'd like to join these kinds of conversations where we dig into the potential and realities of Humans + AI in organizations, check out the community here 🙂 https://lnkd.in/gmhxvikq

Human in the Loop: the fifth ingredient of GenAI that works in practice   The more we work with GenAI, the clearer one principle becomes. The technology can scale effort, but judgment still rests with people. A system performs best when humans guide, review, and refine the output.   GPT models are ultimately pattern recognizers. They are not domain experts and they do not understand the deeper context, the stakes, or the nuance behind a decision. This matters because every model can hallucinate. When teams rely on models without human oversight, quality tends to deteriorate, errors compound and trust disappears. Ultimately, dissatisfaction grows and user adoption slows.   Human in the loop does not mean slowing everything down. It means placing expertise where it matters most. Define what good looks like, review outputs at the critical points, and make the final call on anything that carries risk or requires domain knowledge. Use human expertise to iterate and improve the AI-supported process.   Strong results come from combining human expertise, context, and judgment with the right model and a well designed workflow. This pairing lifts productivity while keeping standards high. GenAI becomes most valuable when it amplifies people rather than replaces them.

AI works best when human judgement is designed into the system, not added after the fact. That’s the idea behind this week’s The Data Science Decoder: “Human Judgment as Infrastructure: Why AI Works Best With Structured Escalation.” As AI moves into real decisioning, the question isn’t whether humans should stay involved. It’s how to embed their judgement intentionally. The strongest architectures don’t rely on ad-hoc oversight. They route decisions based on uncertainty, novelty, and impact, allowing automation to scale while human insight strengthens control. This approach turns escalation into a feature of the system. It improves resilience, supports governance, and builds confidence across stakeholders. Human judgment becomes part of the operating model rather than a safety mechanism. AI maturity isn’t defined by removing people from the loop. It’s defined by structuring how and where they add the most value. Read the full article in The Data Science Decoder:

A recent conversation with a tech executive revealed a crucial insight: As AI automates routine tasks, uniquely human skills become MORE valuable, not less. Here's how AI and humans complement each other: - While AI processes vast amounts of data and spots patterns, humans provide context and derive meaningful insights from those patterns - While AI makes predictions based on historical data, humans provide creative solutions and innovative approaches to unprecedented challenges - While AI automates interactions and processes, humans build genuine relationships and navigate complex emotional dynamics This shift creates what I call the "Cognitive Economy" - where human creativity, emotional intelligence, and complex problem-solving become the most prized assets. The evidence is clear: Companies aren't just hiring for technical skills anymore. They're seeking people who can: - Navigate complexity - Build relationships - Drive innovation Make ethical decisions The future belongs to those who develop these distinctly human capabilities. Are you investing in your cognitive capital? #Leadership #AI #FutureOfWork #Innovation

In the early days of AI, progress meant labelling more data. However, the next generation of AI systems isn’t built by adding more labels to models - it’s built by creating intelligent feedback loops between humans and models. The focus has shifted: - From labelling static datasets → to providing targeted human feedback on edge cases and model failures - From managing annotation queues → to prioritising the most valuable data for the next iteration - From manual ops → to closed-loop systems that guide what data to collect, where models break, and why The shift in focus isn’t just about efficiency—it’s about model performance. The best teams optimise not for data volume but for feedback quality and decision impact. Human feedback, routed at the right time and place through a controlled data layer, is becoming the most strategic asset in the AI development cycle.

Over the past few weeks, jointly with my fabolous colleagues at micro1, Ali Ansari, Ava Fitoussy and LIU ZHANG , we have been working on an burning question that sits at the heart of the Artificial Intelligence debate: What if the “last mile” of human work in AI does not disappear? Our new paper, “No Last Mile: A Theory of the Human Data Market” (https://lnkd.in/eaP2g-jh), develops a formal economic model of structured human data work and its long run role in AI systems. Much of the public narrative assumes that labeling, evaluation, auditing, and exception handling are temporary frictions on the road to full automation. We challenge that assumption. Using an econometric framework, we model structured human input as a persistent production factor that accumulates into a capability stock. This stock sustains reliability, expands the feasible task frontier, and enables deployment in real economic contexts. Because tasks, standards, and environments evolve, this capability depreciates and must be continuously renewed. The equilibrium implication is significant. Even as models improve, there remains a steady, non zero labor share dedicated to structured human data work. Our calibration suggests this may represent roughly 5 to 7 percent in the long run. For me, this is not just a theoretical contribution. It has direct implications for Artificial Intelligence governance, workforce strategy, and investment design. If human oversight is structurally embedded in AI systems, then it must be treated as core infrastructure rather than temporary scaffolding. Automation does not remove humans from the system. It reshapes the economic architecture of human contribution. We are honored to publish open source on arXiv, which is powered by Cornell University. https://lnkd.in/eaP2g-jh

Friends, while everyone races to implement the latest AI tools, there's one secret the top performers already understand: 𝗛𝘂𝗺𝗮𝗻 𝗮𝗴𝗲𝗻𝗰𝘆 𝗺𝗮𝗸𝗲𝘀 𝗔𝗜 𝘃𝗮𝗹𝘂𝗮𝗯𝗹𝗲—𝗻𝗼𝘁 𝘁𝗵𝗲 𝗼𝘁𝗵𝗲𝗿 𝘄𝗮𝘆 𝗮𝗿𝗼𝘂𝗻𝗱. After training thousands of professionals across higher education and industry, one pattern is crystal clear: People who bring purpose and ownership to their use of AI consistently figure out how to capture and create more value from AI. These individuals aren't passively using AI—they're actively engaging with it and even shaping it. Here's the thing: • AI systems don't have agency. They have capabilities, constraints, and contexts—all of which depend entirely on human expertise to shape.  • Your domain expertise is irreplaceable; it's precisely what makes AI effective. 𝙍𝙚𝙖𝙡 𝘼𝙄 𝙩𝙧𝙖𝙣𝙨𝙛𝙤𝙧𝙢𝙖𝙩𝙞𝙤𝙣 𝙙𝙚𝙥𝙚𝙣𝙙𝙨 𝙤𝙣 𝙥𝙚𝙤𝙥𝙡𝙚 𝙬𝙝𝙤: • Feel genuine purpose and ownership of outcomes • Collaborate intentionally, not passively, with AI • Know their unique human contribution can't be automated • Amplify their domain expertise using AI, instead of replacing it The companies leading the AI revolution aren't necessarily those with the flashiest technology. They're the ones investing equally in human potential and AI capabilities—creating environments where people flourish alongside AI. 𝗥𝗲𝗺𝗲𝗺𝗯𝗲𝗿: 𝗛𝘂𝗺𝗮𝗻 𝗔𝗴𝗲𝗻𝗰𝘆 > 𝗔𝗜 𝗔𝗴𝗲𝗻𝗰𝘆. Your AI doesn't care about outcomes. Your people do. Invest in both, and you’ll outpace others. I’d love to hear your thoughts—what role do you see human agency playing in your AI initiatives?

New Research Shows Where Humans Still Outperform AI — And Why It Matters for Trust Introduction A joint study from OpenAI and Anthropic reveals that while artificial intelligence excels at structured, repeatable tasks, humans still dominate when context, emotion, and trust are at stake. The findings underscore a growing behavioral divide between AI-driven efficiency and human-driven authenticity—a gap reshaping how people make decisions and how brands must engage them. Key Details Where AI Excels: According to the Anthropic Economic Index, users rely on AI for structured outputs—writing text, summarizing documents, generating images, or producing step-by-step “how-to” guides. These are tasks defined by clear inputs and predictable patterns. Where People Prevail: When choices require judgment, nuance, or emotion, AI’s role plummets. Only 2.1% of users consult AI for purchases, and even fewer for relationships or self-reflection. People still crave human perspective and lived experience before committing to decisions. The Human Validation Stage: Most users treat AI as a first draft tool, not a final authority. They turn to peers for trust and reassurance—a pattern reflected in platforms like Quora, where: 64% of users prefer human insights over AI summaries. 62% seek expert opinions in their feeds. 54% value firsthand, experience-based advice. Real-World Example: In decision-critical scenarios—career changes, software selection, education choices—users depend on stories from real people who’ve faced similar decisions. These insights convert 4.4x higher than traditional SEO traffic, according to Semrush data. Human-AI Symbiosis: AI systems increasingly cite and amplify trusted human content. Quora appears in 7% of Google AI Mode results, illustrating that human expertise fuels AI credibility. Why It Matters The data signals a powerful truth: in an era of algorithmic abundance, authenticity is the new currency of influence. AI may streamline information gathering, but humans still define meaning and trust. For brands and leaders, success lies in merging both—using AI for reach, but humans for resonance. I share daily insights with 30,000+ followers and 10,000+ professional contacts across defense, tech, and policy. If this topic resonates, I invite you to connect and continue the conversation. Keith King https://lnkd.in/gHPvUttw

💡 Human-centered AI isn't just a feel-good idea. 💡 Human-centered AI (HCAI) is a growing discipline committed to creating #AI systems that retain humans as a critical component. The premise is that AI should be human-controlled and augment human ability rather than replace humans in context. I've spoken about #HybridIntelligence and the idea that human + AI is better than either on its own. HCAI takes that a step further, recognizing that human control is necessary to ensure that AI operates ethically and transparently. HCAI core principles include: ⭐️ a focus on human needs ⭐️ human-AI collaboration ⭐️ user-centered design ⭐️ transparency and accountability ⭐️ positive social impact ⭐️ iterative improvement The idea is to ensure that AI benefits not only our bottom line but our society at large. 💡 And it's important to recognize that HCAI has clear business benefits. 💥 Informed decision-making: While the profound data analysis capabilities of AI are useful, combining that with human values and understanding provides more comprehensive strategies and solutions. 💥 Ethical efficiency and productivity: The computational strength of AI can scale the ideas and human insight of workers while retaining nuanced understanding and moral reasoning. 💥 Improved user experience: By focusing on user needs and preferences, HCAI can create more personalized products and engaging experiences for customers. 💥 Enhanced creativity and innovation: The collaboration of humans and AI can result in new ideas and solutions that would not be possible for either alone. 💥 Ethical considerations and trust: With increased transparency and explainability, and prioritization of human needs and values, HCAI helps to build trust with customers and partners. 💥 Continuous improvement: HCAI enables continuous refinement through iterative feedback loops, providing user feedback to make AI systems smarter and more effective over time. Can you think of other things humans bring to the equation that can benefit business?

User Prompt Is As Important As AI Model >> What Are You Asking For? New research from MIT Sloan School of Management shows that only half of performance gains seen after using a more advanced #AI model come from the model itself. The other half came from how users adapted their prompts — that is, the written instructions that tell an AI model what to do — to take advantage of the new system. The simple but powerful insight that user adaptation contributes as much to performance as the model upgrade itself highlights a critical reality for businesses: Investing in new AI tools won’t deliver their anticipated value unless employees also refine how they use them. This is intriguing, because prompting is a learnable skill that people can improve quickly, even without instruction. This suggests that improvements in AI deployed by companies could lead to productivity gains with less training and change management than in previous technology paradigms. >> Prompting is about communication, not coding The research showed that the ability to adapt prompts over time was not limited to tech-savvy users. “People often think that you need to be a software engineer to prompt well and benefit from AI,” Holtz said. “But our participants came from a wide range of jobs, education levels, and age groups — and even those without technical backgrounds were able to make the most of the new model’s capabilities.” The data suggests that prompting is more about communication than coding. “The best prompters weren’t software engineers,” according to David Holtz, Associate Professor at Columbia University. “They were people who knew how to express ideas clearly in everyday language, not necessarily in code,” he added. >> Intercepting and rewriting user prompts with a high quality LLM made results worse This strongly suggests that people communicating directly with GenAI is more effective than GenAI communicating with GenAI. This suggests, there could be significant risk of degradation of output if multiple Agents are chained together without human communication intervention (depending on the specific application and data sets). >> How can businesses use these insights to improve #EnterpriseAI initiatives? To build on the gains enabled by generative AI, the researchers offer several priorities for business leaders looking to make AI systems more effective in real-world settings: > Invest in training and experimentation. Technical upgrades alone are not enough. Giving employees time and support to refine how they interact with AI systems is essential to realizing full performance gains.   > Design for iteration. Interfaces that encourage users to test, revise, and learn — and display the results clearly — help drive better outcomes over time.   > Be cautious with automation. Automated prompt rewriting may be convenient, but if it obscures or overrides user intent, it can hinder performance rather than improve it. Follow me for Terzo #genai #innovation