How to Use AI in Science Without Losing Ownership of Your Expertise

What does it mean for your job as a scientist if a non-human tool can produce results much faster than you can? Will AI replace you and your colleagues? And how can you protect your critical thinking skills and prevent cognitive decline while using these tools?

I worked in Silicon Valley during the dot-com boom, so I know a bit about tech. In this article, I’ll explain how you can use these technologies as a researcher to do the grunt work while holding onto your human expertise and know-how.

Your Fear is Targeted Incorrectly

The main benefit of automated technologies is that they can do things much quicker or more efficiently than we can. Or they don’t mind doing things that we hate (any dishwasher fans here?).

But a common worry is that using machines will slowly wear away our expertise until one day we can’t remember how to do them. This is even more true with large language models (LLMs), which can produce text and ideas much faster than a human.

I first felt this myself when I asked Claude to turn a transcript of my own podcast into an article. It used my words and my ideas; I agreed with every point, and the result was scarily accurate.

If you’ve ever had this feeling, you might have gotten scared and decided to keep the tool at arm’s length. Ration it, or even “police” it.

I understand the instinct, but going down this path means you spend your vigilance watching the AI’s behavior while your own behavior goes unnoticed. Let me explain.

What the AI Can Do, and What it Can’t

Modern AI draws on an enormous amount of knowledge. Depending on the system, it can search, pull papers, and use other tools.

Contrary to popular belief, it can even generate creative ideas! In one large study, more than a hundred researchers rated AI-generated research ideas as more novel than ideas from human experts, though somewhat less feasible (Si et al., 2024). Google has even built an “AI co-scientist” specifically to propose and rank scientific hypotheses (Gottweis and Natarajan, 2025).

So I personally don’t buy into the belief that AI only knows what you feed it, or that creativity is something only humans can do. Instead, I propose that generating an idea is not the same as being responsible for it.

Generating Ideas vs Taking Responsibility for them

AI might be able to hand you ten plausible hypotheses in under a minute. But it doesn’t know which of them is worth two years of your time. It also can’t stand in front of a panel of reviewers, and it can’t author a paper (yet).

So deciding which ideas matter for good-quality research is still your responsibility. You are still responsible for using what you know and what you’ve experienced to choose what ideas you’re willing to believe and stand behind.

Contrary to what you might think, better AI doesn’t make you less responsible. I would argue it makes you more responsible! Because the more convincing the material in front of you, the easier it is to wave it through without checking.

So think of AI as a highly educated assistant, not a person with more authority than you. The moment you start treating AI like your boss, it’s time to take a step back.

The Real Risk: You Stop Taking Responsibility

So if the tool isn’t the threat, what is? The risk is that by outsourcing all of your “thinking” to AI, you stop engaging with your work and responsibilities. If you let AI’s output, however polished or genuinely excellent, go unquestioned, you’ve stopped doing your job.

The Solution: One Question

So how do you catch this behavior before it becomes a habit? I do this by asking one question before anything leaves my hands:

Where’s the part of this work that only I, a human, could have added?

Not “is this text accurate?” Not “does this text sound about right?” but does this represent:

• My human judgment and experience
• My sense of what matters here and why
• My read on what it means
• The choices I make and am willing to answer for

How Not to Work with AI

I once tuned an LLM to my voice and asked it to produce an article on organoids. But I have no idea what an organoid is! I had to look the word up. This organoid article, written in what sounded exactly like my own phrasing, was far more confident about the topic than I had any right to be as a non-scientist.

So I didn’t publish it, because I couldn’t take responsibility. Because sounding like me isn’t the same as knowing what I know.

Every time you produce something with AI, it helps to ask yourself:

• Can you understand it?
• Can you check it?
• Can you put it in context and take responsibility for it?

If not, you cannot claim responsibility for it, no matter how human it sounds.

How to Work with AI in Science

When I write with AI, it gets me into the neighborhood of what I want. Then I take over, edit, and finish the piece. I make it “me”. The tool does the early lifting, but I decide whether it’s right, whether it’s mine, and whether it’s worth adding my name and publishing.

This is how Bitesize Bio works too. We don’t write pieces for authors under their bylines. If someone’s name is on an article here, you’re getting that person’s actual expertise. Our editorial process ensures what ends up on the page is really theirs.

Co-creation with a tool is great, but co-creation without standards isn’t publishable here. Full stop.

Three Ways to Use AI

There are three ways I see people using AI tools to support their work:

1. Full delegation: The output is good enough, so you leave it as is, and you stop asking questions. Nobody chooses this on purpose. It usually happens because you’ve not chosen anything!
2. Avoidance: You keep AI at arm’s length because opting out feels safer than managing the unease. I get the temptation, but you are missing out. AI is great at taking grunt work off your plate, leaving more room for the work only you can do. And more room for the hard work makes your job easier.
3. Co-creation with retained judgment: You let the tool multiply your effort while you stay the one making the call. AI amplifies what you bring to it; however, it also brings its own errors, assumptions, and biases.

Only the final path uses AI for what it’s good for without negating your human responsibility. If you let AI’s output through without checking it, you get amplified slop.

So bring your expertise, question what comes back, and keep the final judgment yourself, and you end up with something better than what you could have achieved alone.

AI Won’t Take Your Job if You Don’t Let it

AI may well suggest hypotheses worth testing faster than any scientist ever could.

But deciding which ideas matter, working out what the results actually mean, and standing behind the conclusions is still your job. AI hasn’t changed that, but it has made it easier to stop taking responsibility.

So stop watching for signs that AI is taking over and watch yourself instead. Watch for the moment you let something through because it was “polished” without checking whether it was original or true.

AI will keep getting better at producing fluent, useful, even scientifically interesting work. That makes your human judgment even more necessary. Your expertise is still yours; don’t give it away!