Skip to main content

Chemical News

Robo-Chemist: The Latest Version

So let’s talk robotic chemistry experimentation – that always calms everyone right down, doesn’t it? This new paper in Nature from a group in Liverpool is (at heart) a pretty straightforward implementation of modern reaction optimization, with the added feature that it’s being done by a mobile robot, rolling around the lab in the dark for about 20 hours a day. The motorized chemist itself is shown at right – it’s battery operated (the four hours per day of down time is for charging) and can perform a wide variety of tasks and manipulations on both solid and liquid samples.

The reaction being investigated is the semiconductor-mediated photochemical production of hydrogen from water, which has of course been the subject of a massive amount of research. It’s an important topic – water is nothing more than burnt hydrogen, and “unburning” it in a catalytic manner like this would be very useful. We need hydrogen, among other things, to make ammonia to keep us all from starving. Right now we get most of it from steam reforming of hydrocarbons, especially methane, which is an energy-intensive process (has to be, thermodynamically) that runs at about 70% efficiency. About 4% of worldwide hydrogen is produced by electrolysis of water, and this photochemical route is basically a search for a better way to do that latter reaction as compared to sticking the electrodes in and throwing the power switch.

Semiconductor surfaces can react with light to produce electon/electron hole pairs, and those electrons can be used for the water-splitting. But you have to do something about those positively charged holes left in the semiconductor, so “hole scavengers” have been an active area of research. Tertiary amines can do the job, donating electrons back to the material, but that decomposes them in turn. Could there be a catalytic hole scavenger cycle running alongside the catalytic water-splitting one?

Finding such a thing requires a lot of experimentation; it’s very hard to model these processes from first principles, what with all the mixing, surface effects, multiple kinetic steps, and so on. So in this paper the authors picked a photocatalyst (the polymer P10) and first had the robotic system screen 30 candidate hole scavengers. This involves several steps – dispensing the solid polymer into vials, adding solutions of the candidate scavengers, capping the vials, sonicated them to disperse the contents,  photolysis of each vial (under a mixture of visible and UV light), and analysis afterwards by GC. The capping and photolysis stations were built with the robot in mind, but the others were all regular equipment.

Any promising scavenger candidates were set to repeat five times for comparison, and the only two that looked of any interest were ascorbic acid and cysteine, which was converted cleanly to the dimer cystine. The next round of experiments tried to optimize that system, with three photosensitizer dyes, varying ionic strength (addition of NaCl), changes in pH by NaOH addition, addition of surfactants, and addition of sodium disilicate. All of these variations could affect the others, and the group calculated that a full search of all the possibilities (over about 20 different concentrations for the components and the P10) could come out to 98 million experiments.

To deal with that, they set up a system where the system started from predictions of what combinations might be interesting (with bounds on how confident that prediction might be), and then uses a Bayesian framework with a “capitalist acquisition” strategy. The confidence bounds are recalculated as new data come in, and a whole portfolio of experimental conditions is generated with various levels of risk aversion/risk tolerance – greed, in the language of the capitalist algorithm. These portfolios are, in effect, searching for the highest return (maximum rate of hydrogen production), and the system ensures that some of the experiments will be very conservative while others will be comparatively wild shots into the unknown.

After about 150 experiments (roughly two days of running time) it seemed clear that neither the surfactants nor the extra dyes were bringing anything useful – all of them killed hydrogen yields every time they were added. So these were deselected. And after 8 days of total working time (688 experiments) the system had found a mixture of P10, NaCl, NaOH, sodium disilicate, and cysteine that increased the hydrogen yield by a factor of six. Interestingly, the beneficial effects of adding NaOH had earlier been masked by the addition of the dyes, and it made a comeback in the later rounds of experimentation. The sodium silicate variable had a lot of work put into it earlier, but was less important by the end. This might be a good place to note, since the paper makes no mention of it, that the addition of NaCl might not just be an ionic strength effect – chloride ion is also thought to be a useful hole scavenger itself in some systems.

Judging from the time needed to run these experiments by hand, the authors estimate that the mobile robot system is up to one thousand times faster in exploring such a large experimental space, and note that “It is unlikely that a human researcher would have persevered with this multivariate experiment using manual approaches given that it might have taken 50 experiments or 25 days to locate even a modest enhancement . . .” That’s certainly true, but the humans who set up the robots and watched their progress (or lack thereof) can be similarly impatient. I could easily imagine me (or some other human) looking at the results after a couple of days and saying “150 experiments and nothing to show for it! We’re doing something wrong, aren’t we?” Actually, by getting rid of the dyes and the surfactants, that’s more or less what these authors did. Keep in mind that the strategies for exploring experimental space are separate from the capabilities of the mobile robot itself – it’ll roll around and set things up according to whatever genius (or completely cockamamie) schemes you provide.

And that brings up some larger points. I like the idea of using automated and semi-autonomous systems to plow through big problems like this one. But it’s important to realize what the robot and the software don’t do. The search algorithm, for starters, doesn’t seem to have decided to ditch the dyes and the surfactants: the human experimenters did, and it turned out to be key to getting results in the end. Above that, it was of course the human experimenters who decided on the parameter space to search in the first place – the idea of using surfactants was based on another polymer catalyst where that seemed to help, and likewise the pH and ionic strength. These were human-generated hypotheses. The robots and software can run off and do experiments, but humans have to tell them what to do, or at least where to get started and what variable to consider. An even larger issue is the decision of what to turn the robots loose on in the first place – it is, after all, a human decision to try to look for photocatalytic hydrogen generation systems as opposed to spending your time and money somewhere else.

The authors address these points directly, to their credit:

This approach also has some limitations. For example, the Bayesian optimization is blind, in that all components have equal initial importance. This robotic search does not capture existing chemical knowledge, nor include theory or physical models: there is no computational brain. Also, this autonomous system does not at present generate and test scientific hypotheses by itself37. In the future, we propose to fuse theory and physical models with autonomous searches: for example, computed structures and properties1,2,3,4,5 could be used to bias searches towards components that have a higher likelihood of yielding the desired property. This will be important for search spaces with even larger numbers of components where purely combinatorial approaches may become inefficient. . .

So in the end, it comes back to what’s been said before about such automation: it does not get rid of the human element, so much as push the humans to work on the parts that only humans are good at. Those are the higher-level things: what experiments to run, what hypotheses to formulate and how to test them, what variables to introduce. And above all that, what entire types of experiments and projects should be running in the first place. No software would have told you that there was such a thing as the Diels-Alder reaction out there to be discovered and optimized, nor that photoredox synthetic chemistry was an underdeveloped field where effort would pay big dividends. No software would have said “Hey, go search in bacterial defense mechanisms for a tool to edit all the other genomes up to humans” or “Whoa, did you realize that there must be a phase-separation component in the transcriptional machinery?” Nope, that stuff is up to us to find, to know what we’re looking at when we see it, and to realize what it could be used for after that.

41 comments on “Robo-Chemist: The Latest Version”

  1. Lazy guy says:

    Meh, I was more impressed by the mobile molecule printers build by Tesla

    1. Tourettes of Chemistry says:

      It would be impressive if a Robo-Birch would be demonstrated for solvated free electrons to fill those pesky cationic holes. [See ItPL – 24 June 2020]

      Cut the metal (sodium, potassium), rinse, get the NH3(l) in the reaction vessel, mix it all and then find a way to make the concoction available for the system at hand. Would even settle for NaK (commercially available) as the alloy liquid might ease the logistics of material manipulations.

      Go Robo, Go…
      Schedule safety meeting.

  2. EJ says:

    Well in that case, we need more robot friendly labs.

    One issue I see is seriality: the robochemist arm can seemingly screw one cap on at a time, mix in one component at a time, and so on. Labs can’t be assembly lines, but theres definitely room to complete tasks in batches, or perhaps use several slower, cheaper robots in parallel.

  3. dearieme says:

    I’d love to see a summary of the minutes of the lab safety committee as it wrestled with the problems that might be presented. May a lab robot work out of hours on its own? May it sign an agreement to uphold safety policy? …

    1. Ola says:

      A safety meeting? In a university? 🙂

  4. bro-toredox says:

    Derek ‘robot’ is so un-PC. Grad students are people too.

    1. metaphysician says:

      Rossum’s Universal Grad Students?

    2. Andrew Molitor says:

      Precisely what I was thinking. “How is this different from a grad student?”

  5. NHR_GUY says:

    LOL!! 😂”water is nothing more than burnt hydrogen,” I’m going to have to remember that one

    1. JRD says:

      Or as Don Lancaster put it, “Water is an ash.”

  6. Anon says:

    I remember the “end of synthesis” post a while ago. You don’t sound as excited as you were about that one.

    1. Derek Lowe says:

      This is actually less groundbreaking than that one was, to me. The end-of-synthesis stuff is potentially another entire way to build molecules, geared explicitly towards machines. This work is more having a machine do the same thing that a human would do, just more relentlessly and with more patience and reproducibility. Which is what machines do, anyway – this is just a very good example of it. But the experimental algorithm that the robot follows is something that a human investigator could follow as well.

  7. Das says:

    It does get rid of the human element. Back ib the day these manual “simple” tasks where done by humans whk didnt switch to higher level tasks when automation was introduced.
    Is society better off when robots do their tasks and they are drawing unemployment checks?

    1. zero says:

      There are plenty of things people would do if they had time and resources, things like volunteering or aid work. A post-scarcity civilization where ~99% of people spend their time on hobbies and helping each other sounds like a worthy goal to me. It’s certainly better than the knife-edge wage slavery we have today.

  8. KazooChemist says:

    I can’t see the paper due to paywall. I get that the polymer is catalytic, but what is the stoichiometry of the other reagents? The cysteine to cystine is what I am wondering about. Is this also catalytic or is it one to one with hydrogen production? How do you get back to cysteine? By hydrogenation? (Ha-ha-ha).

  9. JasonP says:

    Maybe one could use robots on “the things I won’t work with” list”?

    KABOOM only destroys expensive equipment, not precious lives.

    1. Rich says:

      Nuclear reprocessing plants are fully remotely operated, but I think they use a human at the controls.

      1. Barry says:

        Those are still more “waldo” than “autonomous”

    2. Paul W says:

      Robots could be used to sort out the chemistry of thioacetone.

    3. I am SO reminded of The Skippy List. Note: do not read with beverages in hand or when you’re not supposed to be laughing.

      Original domain’s expired and squatted, so the archive is here:

  10. Stevo says:

    6 days to do the experiment. 2 weeks to tidy the lab and arrange glassware for the robot to fit. 2 years to learn to program the robot into doing everything successfully on a very small number of tasks.

    This sort of work can be useful. However, every new procedure would need lengthy optimization and learning. A student can learn a new technique such as vacuum distillation in an hour, work out what to do when all the B19 RBFs are used up, search the building for a photoreactor when they need it. Good luck with your robot.

    1. Pedwards says:

      A student can learn how to do a vacuum distillation in an hour, but then how long do they spend setting up, monitoring, and cleaning up vacuum distillations? How long does it take before they develop the intuition to be able to figure out the proper vacuum strength, flask temperature, and distillation path length on the first try of a new mixture? Multiplied by the number of people who go through the lab, and the hours start to add up. Meanwhile, one robot has to learn how to do a vacuum distillation and develop that intuition, and then the resulting program can be applied to hundreds or thousands of robots of that model with a software update.

      It’s a specific example (not to mention that no professor is going to waste money on a metallic distillation robot when they have access to a cheap protein-based distillation robot, but industry’s another story), but stuff like this is going to end up being standard in labs. It’s just a question of how long it takes for the machine to be good and cheap enough to be widespread.

    2. Anon says:

      I bet people said the same about High-throughput screening (HTS)…

    3. Scouse says:

      Taking 2 years to do 6 months work in 6 days. But is there any reason why they can’t just keep swapping out reagents and keep doing 6 months work every week ? ROI takes 2 months tops

  11. Rich says:

    If they get this working, the solar arrays will be humungous. My back of the envelope guess (I may of course have stuffed the sums up) based on a gas reforming plant in NZ (Ballance) that uses 5PJ of gas annually, which equates to 150tonnes of H2 a day, which would need maybe 1000km2 of area to catch. That’s just for fertilizer.

    1. Chris Phoenix says:

      I tried to reply but my comment ended up at top-level right below yours. I think you did mess up your numbers; I get about 3.2 km^2, about 30X the land area of the plant, and comparable cost for gas and solar.

      1. Rich says:

        Yes, my bad. I think your numbers are correct. It’s still a big thing though, but would fit on those fields I guess.

        (The plant is here: – I like how their neighbour is called “Blastways”)

  12. 5E15 J/yr of gas = 1.6E8 W. A quick search says solar panels have an average (round the clock) output of about 50 W/m^2. So replacing the energy embedded in the gas would require 3.2 million m^2. But there’s a thousand meters per km, so a million m^2 per km^2. So yes, I think you must have messed up your math.

    Gas is very roughly $3 per GJ so $15M per year. Solar is very roughly $3 per average watt so $500M for the installation which lasts 25(?) years, or $20M per year. But each of my numbers could be off by a factor of 2 or more – the values I found in quick online searching varied by more than that. So it looks like solar is probably comparable in price to gas.

    I think I found the plant you’re referring to in Google Maps, and it looks like it occupies about 1M square feet of land. So the solar required would be about 30 X that land area. (URL on my name.)

  13. ghyu says:

    Robots doing experiments for us has the risk of making us intellectually lazy. You can observe a similar behaviour when the PI tells the grad student to just do all the possible variations of an experiment in a systematic way. This will give you good, easy to interpret data, but it has the weakness of being very slow. This robot is giving me the dystopian vision of a lab full of people playing video games instead of trying to come up with new hypotheses while robots are doing their work.

    The human limitation of not being able to quickly do all possible experiments encourages creativity in finding better ways of doing things and taking risks for example by designing experiments based on a hypothesis not fully confirmed yet. And as in many areas of life, you have to take risks if you want to make an impact. Many of the greatest discoveries were not the result of the slow and steady approach to science, but rather curiosity and the willingness to pursue something like an odd fungus on your agar plates that happened to make penicillin (the robot would have discarded it).

    1. Das says:

      I agree with what youre trying to say. Being limited causes us to solve problems differently but its not as universal as you want to to be. Mendel bred peas for 3 decades before trying to publish his results. No Robot would have sped up that process (well artificial lightning and a greenhouse enables more than 1 crop per year). And sometimes you just need to develop that one transition metal catalyzed reaction to >90 % yield. No fancy redesign, just results in an economic manner.

    2. sgcox says:

      “Machines should work. People should think”
      Not much changed since 67 🙂

    3. James Millar says:

      I think you are underrating the slow-and-steady approach. Eureka is rare, but a very nice story. Plodding has always been important but that inevitably grows with maturity of the field.

      Your chances of earning a name reaction over a career have dropped, because the first hundreds covered a lot of ground.

    4. Before software was (commonly) multi-threaded and multi-process, I’d run into problems where it was genuinely not obvious why something was happening. You’d get the bug report and have no clear way of reproducing, and the crash info (such as it was) was a mystery.

      Even later on, things that *shouldn’t* be happening, were, and this wasn’t a biological system.

      So I can’t even imagine the additional complexity in a biological system analysis.

      (I seriously cannot make this up. My crash reporter just crashed while writing this.)

  14. Andy Cooper says:

    Thanks Derek – this is a nice summary, in some parts I think you go into the chemistry in more detail than we did in the paper! You make an interesting point about chloride acting as a scavenger – this is conceivable, but it would probably be much less effective than cysteine, and hence get ‘squeezed out’ by the algorithm. In fact, it is the algorithm itself that “ditches” the surfactants / dyes (we didn’t intervene), but you’re right that we set up the various hypotheses in the first place, not the robot. And yes, if the initial ideas are “cockamamie” then the robot will plough on until you tell it to stop…

    1. Nico says:

      Gosh it seems so absurd you have to even point that out to a bunch of scientists but from the look of the other comments and partly from the take of the article it seems like this is actually needed. I bet back in the days there were people saying “what’s the point in building a car Mr. Ford if it’s not able to avoid an obstacle like my good ol’ horse??? Not to mention my horse doesn’t even need a road!”

  15. Nile says:

    An interesting corollary to robotic lab assistants: how many lab workers do you know, who have significant mobility problems that make it necessary ror them to use a wheelchair for all or part of the working day?

    What, none?

    I know of one: they will shortly be submitting their thesis, on a topic tat required a lot of bench work and some very aggressive reagents.

    Examine your thoughts on that matter, and consider this: why do you think in that particular way?

    That had a supportive PI, a *mostly* supportive lab staff, a truly terrible building management team who required the threat of prosecution to keep the elevators working and the step-free routes unlocked, and a University administration who want this particular student to not exist.

    Doubtless, naming and shaming them would lead to legal action, followed by lengthy ‘brochure’ PR presentations telling us all how wonderfully caring, inclusive, and accessible their institution is: but they are not, and your Department or Division will contain a number of people like that, too, even if they are a better place than that one, overall.

    Count the wheelchair-marks in the corridor outside your office before you tell me otherwise; or even the number of less-obvious mobility aids that you have noticed during the working day.

    What, there’s none? Or maybe, one – just one – who’s very noticeable as a rare exception? That really doesn’t fit a statistical exectation based on the prevalence of mobility impairments in the graduate and working-age population.

    Whatever. Here’s my point; and yes, it is indeed about the robot in the lab:

    A large part of this student’s success has been well-laid-out laboratory space, well-organised – especially for reagent storage – kept scrupulously clean and unobstructed.

    In short, a laboratory that the robot lab technician can work in, too.

    Think it over: the work that you will have to do to make your lab a robot-friendly workspace is, in part, work that you probably should already be doing every day. And that would make your own life easier, too…

    Just like all the easy things that you and I could be doing, to make the everyday activities of our working lives accessible to people with a disability.

    But does it really take a robot to make us better people?

  16. Scriptwriter says:

    Can we make the robot into an animatronic Gene Wilder?
    Every so often, it would pause and intone “Give my creation life!”

  17. donorcure says:

    During the pandemic, scientists have been able to continue their research while social distancing and under lockdown through artificial intelligence using robots. Wow!

  18. SteveM says:

    Building a mobile robot makes no sense. The additional level of unneeded complexity is enormous. They should build a stationary robot with the lab process equipment directly integrated into it and/or positioned around it.

  19. AlloG says:

    I hear Dat Charles Shultz has rooms full of Robots in LaJolla next to a swimming pool with a full bar. Why does he need them for cartoons? This is a great paper I need one now.

Leave a Reply

Your email address will not be published. Required fields are marked *

Time limit is exhausted. Please reload CAPTCHA.

This site uses Akismet to reduce spam. Learn how your comment data is processed.