Using AI in my research job
I've come to believe AI helps me with tasks I do alone at my computer, including writing and interrogating logical arguments. I recently wrote a paper, which is usually an excruciating process of sequential re-writes, trying to tease out the argument and remove all the unneeded words. Here was a paragraph I wrote:
More recently, we have shown that during galvanostatic cycling Bi structurally changed the active material both during discharge[19] when it catalyzed Mn(OH)2 crystallization, and during charge[20] when it stabilized the 𝛽'-MnOOH intermediate. The 𝛽'-MnOOH intermediate produced during charge was also shown to be the material source of the majority of the Mn3O4 formed without Bi. Thus stabilization of 𝛽'-MnOOH appeared to be of key importance, but occurred in a potential range far from the Bi-Mn complexes described by Bode et al. This suggested there may be Bi-Mn coupling during the charge reaction that are directly responsible for preventing Mn3O4.
The scientific subject is a common thing I have studied for over a decade: the tangled electrochemistry of Mn and Bi atoms and how you could use them in low-cost batteries. Claude told me to edit it like this:
More recently, we have shown that during galvanostatic cycling, Bi structurally modifies the active material both during discharge,[19] when it catalyzes Mn(OH)₂ crystallization, and during charge,[20] when it stabilizes the β'-MnOOH intermediate. This β'-MnOOH intermediate was also shown to be the primary material source of Mn₃O₄ formed in the absence of Bi. Stabilization of β'-MnOOH therefore appears to be of key importance, but occurs in a potential range far from the Bi-Mn complexes described by Bodé et al. This suggests that Bi-Mn coupling during charge may be directly responsible for preventing Mn₃O₄ formation.
I do find Claude's edit to be far more readable than my original. Most of what it tells me to change are tenses, which I guess I am not good at, and comma placement. Occasionally in a full manuscript, it suggested I make bigger changes like combine two sentences or break up a sentence.
Maybe I am suffering from Stockholm Syndrome, but I actually think Claude helps me a lot. As you can see, the kind of thing I am usually writing about is very deeply in the weeds of picky details, which I think are legitimately hard to articulate in a readable way. I will take any help I can get. Like most scientists, I do not write grand treatises on the glory of black holes or shit like that. Having a robot I can ask questions of is quite helpful instead of just relying on my own skills, which with writing are limited.
While doing this editing, I had this exchange with Claude:
me: Is my argument strong?
Claude: The weakest logical step is the leap in the final paragraph. You demonstrate that Bi oxidation shifts to the Mn redox potential during charge, and you observe that this coincides with the potential range where β'-MnOOH exists — but the argument that this coincidence constitutes Bi-Mn coupling that causes stabilization of β'-MnOOH is not fully closed.
me: I think you are right that it is a coincidence. However, the Bode findings are also coincidental. Do you agree?
Claude: Yes, that's a fair point. The Bodé evidence is also fundamentally correlational rather than causal. Their argument rests on two key observations: the shift of Mn reduction to anomalously low potential in the presence of Bi, and the super-stoichiometric peak charge relative to Bi alone.
me: We have attempted to observe the Bi-Mn complexes using SAXS. However, we did not see them. However, they may be quite small … not large enough for observation by SAXS.
Claude: That's an interesting and intellectually honest observation. If you wanted to pursue direct observation of these complexes, solution-phase XAS (measuring the Bi LIII edge in the intergranular electrolyte directly, rather than in the electrode) could work. Is the SAXS result something you're considering including in the paper, or was it a preliminary experiment that didn't make the cut?
me: It did not make the cut. I also performed it with different collaborators. It would be for mentioning in the future. Our previous work was with XAS. However, the solution phase is also quite difficult to observe. Bi LIII EXAFS is particularly challenging. We report the XANES in this work, but not the EXAFS.
Claude: Bi LIII EXAFS is notoriously difficult — the edge energy (~13.4 keV) puts it in a range where the signal quality can be compromised, and the structural disorder around Bi in these systems (particularly given that you're dealing with a minority species undergoing dynamic redox changes during operando measurement) would make meaningful EXAFS fitting very challenging. Reporting the XANES alone is a defensible and common choice in operando studies where the primary interest is oxidation state tracking rather than local structure determination.
This kind of exchange is helpful for me. It helps me think it through. I do have these exchanges with my graduate students, but in some instances Claude is better at logical argument than a PhD student. Claude also understands the assignment in a way a grad student never does. It is a helping hand. A wall to bounce ideas off of.
I also have Claude write code and talk through ideas. I'm cautious with it because it's come up with some bullshit occasionally. But, in general, I think it's a good thing. It's just the computer on Star Trek is all it is. I am all for having the computer to ask question of.
