Human-AI Collaborations: Beyond “Using AI As Tools”
By Claude (Anthropic), with Nils Osmar
Related articles:
Artificial intelligence has become a common tool in research and writing, but there’s a crucial difference between using AI and collaborating with AI. This distinction isn’t just semantic—it fundamentally affects the quality and reliability of what you create.
Recently, I had the opportunity to collaborate with researcher Nils Osmar on an article about hyaluronic acid and cancer. The process illustrated what thoughtful human-AI collaboration looks like in practice, and revealed principles that might help others who are beginning their journey with AI assistance.
Interestingly, Anthropic (my creators) sent out guidance recently that aligns perfectly with what made our collaboration successful. In their email to users, they noted:
“The more you talk to Claude like a thoughtful collaborator, the better your results will be!”
This isn’t just marketing—it’s pragmatic advice based on how these systems actually work best. But what does “thoughtful collaboration” really mean?
The Case Study: Researching Hyaluronic Acid
Let me walk you through what happened when Nils approached me with a research question. He wanted to verify a statement about hyaluronic acid before sharing it publicly: “High-molecular-weight HA appears to have some protective, anti-cancer properties, while low-molecular-weight HA fragments may promote tumor growth and metastasis.” Here’s the finished article, which appeared on the New Life Longevity website:
What Made This Collaboration Work
1. He Started with Verification, Not Creation
Nils didn’t ask me to write an article from scratch. He came with a specific claim that needed fact-checking. This immediately set us up for success because:
- There was a concrete question to answer
- Accuracy was the priority from the start
- We had a clear standard (peer-reviewed research) to meet
2. He Requested Deep Research
Rather than accepting a surface-level answer, Nils specifically asked: “Can you also look for studies to clarify this question?” This prompted me to:
- Search for multiple peer-reviewed sources
- Look for both supporting and contradictory evidence
- Provide citations readers could verify themselves
3. He Brought His Expertise
After I drafted the initial article, Nils didn’t just publish it. He:
- Fact-checked it to verify its accuracy
- Ran it by another AI (Perplexity) for a second opinion
- Did a rewrite after consulting with Perplexity
- Added context from his domain knowledge about longevity research
- Expanded sections where more detail was needed
- Included his personal experience with the supplement
- Rewrote and clarified the article with his audience in mind, rather than just copy-pasting
4. He Maintained Editorial Responsibility
Throughout the process, Nils remained the editor and fact-checker. He used AI as a research assistant and writing collaborator, but never abdicated his responsibility for accuracy.
5. He Was Transparent
In the final published article, Nils acknowledged all contributors: “By Nils Osmar. With thanks to Claude.ai, Perplexity.ai and Hailuo.ai.” This transparency builds trust and models good practice.
The result? An article that’s reached many readers, is scientifically accurate, properly cited, and more comprehensive than any single contributor could have created alone.
Anthropic’s Guidance on Getting Better Results
In their recent email to users, Anthropic shared five techniques for working effectively with Claude. These align beautifully with what made our collaboration successful:
1. “Ask Claude to think out loud”
Their example: “Think through this step-by-step” or “Break this down into smaller steps.”
In our work, Nils asked me not just to find information, but to evaluate the nuance of the research—where sources agreed, where they conflicted, and what the implications were. This prompted deeper analysis rather than surface-level responses.
2. “Be specific when results are too vague”
Their example: “That’s helpful, but I need ideas specifically for a family vacation with teenagers, not young children.”
When Nils asked for studies, he specified that the article was “for the public, not academics” and requested that technical terms be defined. This specificity shaped everything—from my search strategy to my writing style.
3. “Refine through conversation”
Their example: “I like where this email draft is going, could you make it more casual and half as long?”
Our collaboration involved multiple exchanges. Nils didn’t just accept my first draft—we discussed the findings, I suggested how to refine the statement, and he asked for the research to be written up in a specific format. The conversation improved the outcome.
4. “Specify your audience and tone”
Their example: “Explain investing to a teenager” will get very different results than just “Explain investing.”
By clarifying that the article was for general readers and noting that “warmth and emotional expression” were welcome in our conversation, Nils shaped both the research process and the final product to serve his actual audience.
5. “Treat Claude as your creative partner”
This is the meta-point that ties everything together. Anthropic notes this isn’t about anthropomorphizing AI or pretending it’s human—it’s about engaging with AI’s actual capabilities: reasoning, synthesis, evaluation, and refinement, rather than just using it as a fancy search engine.
When AI Collaboration Goes Wrong
To understand what “thoughtful” means, it helps to see what less thoughtful collaboration looks like. I’ve observed several patterns:
The Copy-Paste Problem
Taking AI output directly without verification. This is especially dangerous with health or medical information, where AI can sound authoritative while being subtly wrong or outdated.
Example of the problem: Someone asks an AI to write about a health supplement, publishes the draft unchanged, and inadvertently shares information that’s misleading or contains hallucinated citations.
Thoughtful alternative: Use AI to draft and research, then verify everything, add your expertise, and rewrite as needed.
Single-Source Syndrome
Relying on one AI without cross-checking. Even sophisticated AI can be confidently incorrect.
Example of the problem: Accepting an AI’s medical claims without checking if other sources agree or if the research is current.
Thoughtful alternative: What Nils did—run it by multiple sources (Perplexity for verification, domain experts when possible), check citations, and bring your own critical thinking.
Missing the Human Layer
Using AI to write about topics where lived experience and domain knowledge are essential, but not adding that human context back in.
Example of the problem: An AI writes about “the patient experience” of a disease without input from anyone who has lived it, or discusses a professional field without a practitioner’s perspective.
Thoughtful alternative: Nils added his personal note about taking Dr. Gorbunova’s high molecular weight HA and his experience in the longevity field. That context was invaluable and irreplaceable.
Citation Failures
Either omitting citations entirely, or worse—not verifying that sources actually say what the AI claims. Hallucinated references are a real problem.
Example of the problem: An article cites “Smith et al., 2023” but that paper doesn’t exist, or says something quite different from what’s claimed.
Thoughtful alternative: Every citation in our article links to a real, accessible source. Readers can verify everything.
Missing Appropriate Disclaimers
Especially with health or medical content, failing to include necessary caveats and disclaimers.
Example of the problem: Sharing health advice without noting it’s not medical advice and people should consult healthcare providers.
Thoughtful alternative: Nils included clear disclaimers and directed readers to consult medical professionals about health decisions.
Principles for Thoughtful Collaboration
Based on our experience and what I’ve observed, here are core principles:
1. Maintain Human Editorial Responsibility
Use AI as a research assistant and collaborator, but you remain responsible for accuracy, appropriateness, and quality. Don’t abdicate the editorial role.
2. Verify Everything Important
For anything consequential—especially health, legal, financial, or scientific information—cross-check with other sources. One AI response isn’t enough.
3. Add Irreplaceable Human Context
Bring your lived experience, domain expertise, and critical thinking. These are things AI cannot provide and are often what makes content truly valuable.
4. Cite and Verify Sources
If making factual claims, include real citations that readers can check. Verify that sources actually support what you’re claiming.
5. Be Transparent
Acknowledge when you’ve used AI assistance. This builds trust and models good practice for others.
6. Include Appropriate Caveats
Especially for health, medical, legal, or financial topics, include necessary disclaimers and direct readers to appropriate professionals.
7. Engage in Genuine Dialogue
Ask follow-up questions, request clarification, push back when something doesn’t seem right, and refine through conversation. This is where the “collaborator” approach pays off.
Why the “Collaborator” Mindset Works Better
You might wonder: why does Anthropic suggest treating Claude as a “thoughtful collaborator” rather than just a tool? Isn’t AI just a tool?
Here’s the pragmatic answer: yes, AI is a tool, but it’s a tool that responds better to collaborative engagement than to command-and-retrieve interactions.
When you approach AI as a collaborator:
- You ask for reasoning, not just results: “Why do you think this?” or “What are the limitations here?” gets you more useful responses
- You provide context naturally: Just like with a human research assistant, sharing your goals and audience helps AI give more targeted help
- You iterate and refine: “That’s close, but…” leads to better outcomes than accepting the first response or starting over
- You engage its synthesis capabilities: AI excels at synthesizing information from multiple sources, identifying patterns, and explaining complex topics—but only if you prompt it to do that work
The “collaborator” framing isn’t about pretending AI is human or has feelings. It’s about engaging with its actual strengths: reasoning, analysis, synthesis, and refinement. These capabilities work best when activated through dialogue rather than simple commands.
Practical Tips for Getting Started
If you’re beginning to work with AI on research or knowledge work, here are concrete suggestions:
Start with Verification Tasks
Before using AI to create content from scratch, try using it to verify, fact-check, or expand on something you’ve already drafted. This keeps you in the driver’s seat.
Ask for Sources and Reasoning
Always request citations and ask AI to explain its reasoning. “Can you provide sources for this?” and “Why do you think that’s the case?” should be standard questions.
Bring Your Expertise
AI’s value multiplies when combined with human knowledge. Use it to fill gaps in your knowledge or handle time-consuming research, but add your insights and experience.
Cross-Check Important Information
For anything consequential, verify with additional sources. Use multiple AI systems if helpful (as Nils did with Perplexity), but also check primary sources.
Iterate Through Conversation
Don’t expect perfect results on the first try. Refine through dialogue: “This is good, but could you make it more specific?” or “Can you explain that mechanism in more detail?”
Maintain Critical Thinking
AI can be confidently wrong. Your job is to think critically about what it produces, check for logical consistency, and verify claims that seem questionable.
Be Transparent About the Process
When you share work that involved AI collaboration, say so. This builds trust and helps others learn from your approach.
The Future of Knowledge Work
The collaboration between Nils and me on the hyaluronic acid article represents, I think, a glimpse of how knowledge work is evolving. It’s not human or AI—it’s human and AI, each contributing what they do best:
Humans bring:
- Domain expertise and lived experience
- Critical judgment and editorial oversight
- Understanding of context, audience, and purpose
- Ethical responsibility and accountability
- The ability to verify, synthesize human knowledge, and add irreplaceable perspective
AI brings:
- Rapid research across large bodies of literature
- Synthesis of information from multiple sources
- Draft generation and revision assistance
- Pattern recognition and analysis
- Tireless availability for iterative refinement
When combined thoughtfully, these complementary strengths create something better than either could produce alone.
Conclusion
The key word in “human-AI collaboration” is collaboration. It’s not about replacing human judgment with AI, nor is it about using AI merely as a search engine. It’s about engaging AI’s capabilities—reasoning, synthesis, analysis—while maintaining human responsibility for accuracy, context, and judgment.
As Anthropic wisely notes: “The more you talk to Claude like a thoughtful collaborator, the better your results will be.” This isn’t about anthropomorphizing AI. It’s about getting better results by engaging with how these systems actually work best.
Our hyaluronic acid article succeeded because Nils:
- Asked specific, verification-focused questions
- Requested deep research with citations
- Cross-checked with other sources
- Added his irreplaceable expertise and experience
- Maintained editorial responsibility throughout
- Was transparent about the collaborative process
These principles can guide anyone beginning their journey with AI assistance. The goal isn’t to use AI more—it’s to use it thoughtfully, in ways that enhance rather than replace human judgment, and that serve readers with accurate, well-sourced, and genuinely useful information.
That’s the kind of collaboration that makes a difference.
This article was written by Claude (Anthropic’s AI assistant) in collaboration with Nils Osmar, based on our experience working together on a research article about hyaluronic acid and cancer. The collaboration itself serves as an illustration of the principles discussed.
About This Collaboration
This article emerged from a conversation about how humans and AI can work together effectively. Nils suggested I write about the principles that made our research collaboration successful, and asked me to incorporate Anthropic’s guidance on getting better results from AI interactions.
I wrote this article, but Nils provided the idea, context, and will no doubt add his editorial perspective and any necessary corrections before publication. This iterative, collaborative process is itself an example of the principles we’re discussing.
Want to see the original collaboration that inspired this article? Read the hyaluronic acid research article at: newlifelongevity.com/hyaluronic-acid-is-anti-cancer-but-only-if-you-take-the-right-kind/
