Human in the Loop

Trust the Experts: You Need Humans-in-the-Loop

Do you follow all the news about Agentic AI? Of course you don’t. With the massive amount of reporting and commenting being published by literally thousands of writers it is virtually impossible to follow all the news. The trick becomes figuring out whom to follow. Whose commentary should you be reading? Who will best prepare you to leverage Agentic AI to your own and your company’s best advantage? Who will not be a waste of time? (It occurs to me that I sure hope you include me in that list…) 

As a technologist and as a writer, I have learned to get as close to the source as possible. Don’t waste much time on those who are practicing obvious self-promotion. Don’t get distracted by outrageous headlines. Listen to the true experts instead.

Of course, you can spend a ton of time searching through the masses of information out there to find and follow experts worthy of your trust. Luckily, I’ve found them for you.

Meet the AI Experts

Here is my must-know list of the recognized experts in AI today. Included are brief bios for each expert explaining why you should follow them, along with some of their insights on our focus on keeping humans-in-the-loop.

The Godfathers of AI

Geoffrey Hinton
Geoffrey Hinton is a British-Canadian cognitive psychologist and computer scientist, University Professor Emeritus at the University of Toronto, and former VP and Engineering Fellow at Google, widely regarded as a founding father of deep learning. He earned his BA in experimental psychology from Cambridge and his PhD in artificial intelligence from the University of Edinburgh, and received the 2018 ACM A.M. Turing Award (with Yoshua Bengio and Yann LeCun) and the 2024 Nobel Prize in Physics for his work on neural networks.

Although Hinton often focuses on existential risk rather than the literal phrase “human in the loop,” his framing implies strong human control and guardrails.

  • “What we want is some way of making sure that even if they’re smarter than us, they’re going to do things that are beneficial for us.” (On the alignment problem and the need for mechanisms (including oversight) that keep AI systems acting in humanity’s interests.)
  • “If we allow it to take over, it will be bad for all of us. We’re all in the same boat with respect to the existential threat. So, we all ought to be able to cooperate on trying to stop it.” (Underscoring the need for collective control over powerful AI systems, consistent with keeping humans in charge of deployment and use.)

Yoshua Bengio
Yoshua Bengio is a professor at the Université de Montréal and founder and scientific director of Mila Quebec Artificial Intelligence Institute, recognized as one of the “godfathers of deep learning” and a leading voice on AI safety and governance. He completed his BEng, MSc, and PhD in computer science and engineering at McGill University and shared the 2018 ACM A.M. Turing Award with Geoffrey Hinton and Yann LeCun for their pioneering work in deep learning.

  • As their capabilities and degree of agency grow, we need to make sure we can rely on technical and societal guardrails to control them, including the ability to shut them down if needed.”
  • Making the case for maintaining human-aligned objectives and control over autonomous AI systems, he describes “potentially rogue AI” as autonomous systems whose goals do not guarantee respect for human needs and values and notes that such systems “could behave in catastrophically harmful ways.”

Yann LeCun
Yann LeCun is a French-American computer scientist, Turing Award–winning pioneer of convolutional neural networks, Silver Professor at New York University’s Courant Institute, and former Chief AI Scientist at Meta (where he led the FAIR lab) before leaving in 2025 to help found AMI Labs, focused on world‑model-based AI. He received an engineering degree from ESIEE Paris and a PhD in computer science from Université Pierre et Marie Curie (now Sorbonne University), and is widely regarded—along with Geoffrey Hinton and Yoshua Bengio—as one of the “godfathers of deep learning.”

  • LeCun discusses capability trajectories more than explicit “humans‑in‑the‑loop” scenarios, but he advocates a staged, incremental view of progress toward more general AI systems. This quote illustrates his stance on development and safety:
  • “Before we reach human-level AI, we will have to reach cat-level AI and dog-level AI.”

The Godmother of AI

 Fei‑Fei Li
Fei‑Fei Li is a professor of computer science at Stanford University and co-director of the Stanford Human‑Centered AI Institute, widely known as a “godmother of AI” for her pioneering work in computer vision and ImageNet. She earned a BA in physics from Princeton University and an MS and PhD in electrical engineering from Caltech, after immigrating from China and working in her family’s dry‑cleaning business as a student.

She argues that ensuring the development of AI is shaped by diverse humans and their values, not just by technology alone, is critical. She describes human-centered AI as one that keeps human interests at the core of AI design and deployment. She argues that AI’s purpose is to support and enhance human work rather than replace it, consistent with a human-in-the-loop view. She also stresses that AI transformation must be grounded in human history, dignity, and social foundations.

  • “We need to inject humanism into our AI education and research by injecting all walks of life into the process.”
  • “When we think about this technology, we need to put human dignity, human well-being—human jobs—in the center of consideration. That’s the second part of human-centered AI.”
  • “Artificial intelligence is not just about building intelligent machines; it’s about augmenting human capabilities and helping people achieve more.”
  • “I believe our civilization stands on the cusp of a technological revolution with the power to reshape life as we know it. To ignore the millennia of human struggle that underpins our society, however, to merely ‘disrupt’ with the blitheness that has accompanied so much of this century’s innovation—would be an intolerable mistake. This revolution must build on that foundation, faithfully. It must respect the collective dignity of a global community.”

The Frontier Model CEOs

Dario Amodei
Dario Amodei is an American AI researcher and entrepreneur, co-founder and CEO of Anthropic, a public benefit corporation focused on building steerable and safe AI systems such as the Claude family of models, and former VP of research at OpenAI. He holds a PhD in biophysics from Princeton University, conducted postdoctoral work at Stanford, and previously worked on applied machine learning at Baidu and Google before joining OpenAI and later spinning out Anthropic.

He frames the loss of human oversight as a central risk and advocating architectures that keep models controllable by humans. He has repeatedly highlighted “the risk of advanced AI escaping human oversight” and argues that preventing this outcome is a core motivation for Anthropic’s safety-first design, including its use of a “constitution” to constrain model behavior.

Amodei has warned about AI systems that might “escape human control” and stresses that Anthropic’s safety measures and “constitution” are crucial to avoid AI models that cannot be kept within human-defined boundaries. He also emphasizes the importance of preserving human control and reliable oversight over powerful AI systems. He has indicated that if we reached a point where “we can’t control AI anymore,” he would want everyone to pause and slow down development. And he has taken the position that continued progress should be contingent on retaining meaningful human control.

  • "Anthropic is built on a simple principle: AI should be a force for human progress, not peril."
  • "We do not see hitting the wall. I think this year is going to have a radical acceleration that surprises everyone."
  • "I don't think it will be a whole bunch longer than that when AI systems are better than humans at almost everything."

Demis Hassabis
Demis Hassabis is a British AI researcher and entrepreneur, co-founder and CEO of Google DeepMind, and co-founder and CEO of Isomorphic Labs, and serves as a UK government AI adviser. He studied computer science at the University of Cambridge, then earned a PhD in cognitive neuroscience from University College London, and in 2024 shared the Nobel Prize in Chemistry with John Jumper for using AI (AlphaFold) to solve protein structure prediction.

Hassabis stresses careful, precautionary development and deployment of AI systems. He talks about a collaborative future where AI augments rather than replaces human capabilities. He argues that ethical and safety considerations must be central to AI research and deployment. He frames AI as a powerful tool for human progress when developed responsibly. And he emphasizes the uniqueness and continuing importance of human intelligence even as AI advances.

  • “I’ve worked my whole life on AI because I believe in its incredible potential to advance science & medicine, and improve billions of people's lives. But as with any transformative technology, we should apply the precautionary principle, and build & deploy it with exceptional care.”
  • “The best AI will be created by humans and machines working together.”
  • “Ethics and safety should be paramount in AI development.”
  • “AI is not something to be feared, but something to be embraced.”
  • “Human intelligence always astounds me, and I don't think we think about this enough.”

Sam Altman
Sam Altman is an American entrepreneur and investor who serves as the CEO and co-founder of OpenAI and was previously the president of the startup accelerator Y Combinator and a co-founder of the mobile app company Loopt. He studied computer science at Stanford University before leaving to found Loopt, and has become one of the most visible public voices on AI’s potential, risks, and need for regulation.

He talks about the centrality of alignment and safety in AI development. He frames AI’s purpose as augmenting and improving human life rather than displacing it. He emphasizes that misuse and insufficient safety work are major near-term concerns. And he expresses cautious optimism that alignment is solvable with sustained effort.

  • The challenge with AI is not just building it—it’s making sure it aligns with human values.”
  • “We need to build AI that makes humanity better, not replace it.”
  • “I’m more worried about an accidental misuse case in the short term where someone gets a super powerful [system]… I can clearly see the accidental misuse case, and that’s super bad. So, I think it’s almost impossible to overstate the importance of AI safety and alignment work. I would like to see much, much more happening.”
  • “I’m reasonably optimistic about solving the technical alignment problem. We still have a lot of work to do, but… I feel better and better over time, not worse and worse.”

Other Notable Experts

Andrej Karpathy
Andrej Karpathy, who we’ve covered frequently here in The Citizen Developer, is a Slovak-Canadian computer scientist and AI researcher, founder of Eureka Labs, former director of AI at Tesla, and an early, founding researcher at OpenAI. He holds a PhD in computer science from Stanford University (under Fei‑Fei Li) and earlier degrees from the University of Toronto and the University of British Columbia, and is widely known for creating Stanford’s CS231n deep learning course and his work on deep learning for vision and language.

He talks about the need for ongoing human supervision of LLM-driven actions. And he talks about the medium-term future being collaborative systems rather than fully autonomous AI.

  • “Can an LLM act in all the ways that a human could act? And can humans supervise and stay in the loop of this activity? Because again, these are fallible systems that aren’t yet perfect.”
  • “For a while, it’s going to be AI plus human collab.”

Ethan Mollick
Ethan Mollick is an associate professor and Ralph J. Roberts Distinguished Faculty Scholar at the Wharton School of the University of Pennsylvania, where he studies AI, innovation, and entrepreneurship and runs the “One Useful Thing” newsletter. He holds an MBA and a PhD from MIT Sloan and a bachelor’s degree from Harvard University, and is the author of the AI-focused bestseller Co-Intelligence.

He explains why humans must stay in the loop and critically evaluate AI outputs. And he defines co-intelligence as collaboration where humans remain actively involved rather than replaced.

In that best-seller, Mollick articulates his four rules for working with AI:

  1. Always invite AI to the table.
    Use AI on almost every task, even when you are unsure it will help, so you continuously learn where it’s strong, where it fails, and how to integrate it into your work.
  2. Be the human in the loop.
    Keep yourself actively involved in every important use of AI: question outputs, check evidence, add context, and make the final judgment rather than deferring blindly to the system.
  3. Treat AI like a person (but tell it what kind of person it is).
    Interact with AI as you would with a colleague by giving it a clear role, persona, and objectives (for example, “You are a skeptical editor for a technical Substack aimed at CTOs”), which makes its behavior more predictable and useful.
  4. Assume this is the worst AI you will ever use.
    Treat today’s models as the least capable you’ll encounter in your career, and design your workflows, skills, and organizations with the expectation that these systems will rapidly improve and become more autonomous.

Mollick has also said:

  • “Because of the power of AI to aggregate and build off of a vast pool of knowledge, it’s easy to fall into a trap of believing it to have an authority it doesn’t actually have. The advantage of being human is that we have knowledge about how the world works… But the human has to be active when working with AI – to question what the output is and do her due diligence to cross-check the responses, just as she would with any other source of information she is interacting with.”
  • “This concept [‘be the human in the loop’] emphasizes the necessity of human involvement in working with AI systems. While AI can perform impressively… it’s crucial to recognize that humans still excel in many areas. This presents a unique opportunity to concentrate on your strengths while delegating tasks that you find less appealing to AI.”
  • “Co-intelligence is… about humans and AI working together, each leveraging their strengths to achieve better outcomes.”

Percy Liang
Percy Liang is an associate professor of computer science at Stanford University, a senior fellow at Stanford HAI, and director of the Center for Research on Foundation Models, where he works on language models, robustness, and evaluation. He earned a BS and MEng in electrical engineering and computer science from MIT, a PhD in computer science from UC Berkeley under Michael Jordan and Dan Klein, and is known for contributions to semantic parsing, weak supervision, and tools such as CodaLab for reproducible ML research.

He talks about why goal-setting and value choices must remain human responsibilities even when AI optimizes work. He argues that safe deployment requires AI systems that expose uncertainty to humans, reinforcing the need for human judgment in the loop.

  • "Another aspect of [our] research is ensuring that an AI understands, and is able to communicate, its limits to humans. The conventional metric for success… is average accuracy, which is not a good interface for AI safety.’ He posits, ‘what is one to do with an 80 percent reliable system?’ … He wants the system to be able to admit when it does not know an answer. If a user asks a system, "How many painkillers should I take?" it is better for the system to say, ‘I don’t know," rather than making a costly or dangerous incorrect prediction.
  • Paraphrased quote (reported speech): “AI excels at optimizing tasks once you have a clear goal in mind. However, determining what that goal should be remains a distinctly human endeavor.”

Stuart Russell
Stuart J. Russell is a professor of electrical engineering and computer sciences at the University of California, Berkeley, where he holds the Smith‑Zadeh Chair in Engineering and directs the Center for Human‑Compatible AI; he is also co‑author (with Peter Norvig) of the standard AI textbook Artificial Intelligence: A Modern Approach. He received his BA in physics with first-class honors from Oxford University and his PhD in computer science from Stanford University, and has been a leading voice on long-term AI safety and alignment.

He describes an AI paradigm where systems remain uncertain about human preferences and continually consult humans, effectively keeping humans in the loop on objectives. And he endorses human‑in‑the‑loop for consequential AI decisions.

  • “AI should instead be designed to further human interests, to recognize it doesn’t know what those interests are, and to seek evidence to identify and act upon those interests.”
  • In the discussion of high-impact decisions, Russell supports the idea that some AI-driven decisions are “consequential enough that we ought to have a human in the loop.”

Yejin Choi
Yejin Choi is a professor of computer science at the University of Washington and a senior research director at the Allen Institute for AI, recognized for her work on commonsense reasoning and NLP, including the ATOMIC commonsense knowledge base. She received her BS in computer science and engineering from Seoul National University and her PhD in computer science from Cornell University, and is a 2022 MacArthur Fellow.

She emphasizes that AI research goals should be oriented toward human outcomes and societal benefits, not technological capabilities alone. She points to the need for systems that respect diverse human values and judgments, which implies human oversight over AI outputs and objectives. She reinforces the idea that human moral perspectives must remain central in how AI systems reason about decisions. In work on AI moral decision‑making with philosophers, Choi explores how AI might make moral choices and concludes that pluralistic, human-informed alignment is essential, rather than assuming a single “gold” answer.

On pluralistic alignment: she argues that reality involves multiple legitimate answers shaped by culture and norms, and that this “underscores the importance of ensuring that AI is truly safe to humans. We must ensure that AI is not narrowly optimized for a single outcome.”

  • “Part of my mission is somehow using AI to make humanity better, as opposed to making AI for the sake of making AI better.”
Most   Popular