<100 subscribers
Share Dialog
Share Dialog
Originally published via martinholub.com.
Background: In the Netherlands, it is customary to include ‘propositions’, essentially opinionated statements defended alongside one’s doctoral work. In this post series, I am outlining the arguments supporting my propositions.
Proposition #5: The culture of academic research is not keeping pace with advancements in the for-profit sector.
I am sure I wrote the concentration somewhere… (feverishly looking through a mix of paper and electronic notes) …
I can’t get your environment to install on my machine… (I might if I try very hard, but didn’t you say it should just work?) …
You should send an email to Tracy if you’d like to get trained to use this instrument… (she works Monday, Tuesday, Thursday, but might be on vacation; anyway, she’ll send you a test, and after the training is over, you’ll fill out a form and be good to go!)…
In academic research, we are often in the business of trying to do something for the very first time. This fosters a culture where unique solutions and one-off tricks are common and celebrated. And honestly, that’s probably unavoidable and mostly alright. In many cases, that’s what I personally love. What’s not alright however, is our tendency to extend this artisanal approach to everything we do. The examples at the start of this post will be familiar to most academic scientists. There are countless others. At the university where I did my PhD, orders for laboratory supplies had to be manually reviewed by no less than four people, causing week-long delays. If you didn’t manually enter the item into the ordering system, it would never arrive. Automatic stock monitoring for common chemicals? Forget about it. Here are more examples: Want to keep your lab notebook old school, pen and paper, scribbling illegibly? Be my guest! Manually entering work hours for every day of 4+ years? Sure - great use of time! (And let’s not forget about somebody having to send reminders about it every month!). I’m far from the first to argue that research needs a productivity boost: 1
Increasing number of scientist, larger team size, and increasing complexity of scientific questions have led to decreasing per-scientist research output [source].
In the for-profit sector, efficiency is the mantra. If a solution can save even 5% of the time it takes to produce a deliverable, it’s likely to be adopted. Deployed across a team or company, such efficiency gains can have a tangible impact on the bottom line. Ignoring them means risking becoming irrelevant.
“The most important contribution management needs to make in the 21st century is […] to increase the productivity of knowledge work and knowledge workers. […] The most valuable asset of a 21st-century institution (whether business or non-business) will be its knowledge workers and their productivity.” — Peter F. Drucker (1999)
“Productivity isn’t everything, but, in the long run, it is almost everything. A country’s ability to improve its standard of living over time depends almost entirely on its ability to raise its output per worker.” – Paul Krugman (1994)
Consider this extreme, but illustrative, example of the impact of increasing productivity: In 2018, the CFDA (Chinese FDA) decided to do away with large bureaucratic burden and shifted to an ‘implied license’ system. As a consequence, the time needed for new drug application reviews decreased more than fivefold. [^2, ^3] Chinese biopharma increased candidate drug filings tenfold over the last decade as a result and has already overtaken Europe in clinical trial initiations. 2 The inability to innovate can clearly lead to a loss of competitiveness and risk of obsolescence.
More and more new medicines are being developed in China, mostly at the expense of Europe [source].
Another example: A recent experiment introduced an AI material discovery tool to a group of scientists at a large company. A 24-month follow-up compared their performance with a control group that continued without the tool. After 16 months, the treatment group discovered 44% more new materials, filed 39.5% more patents, and generated 17% more new product prototypes compared to the control group. As the authors did not neglect to remark, “these effects are large”. 3 Indeed, they are large, if not massive. Interestingly, despite increased productivity, scientists who adopted the tool reported lower job satisfaction. While it’s still early, one can easily imagine industry adopting such a tool despite its drawbacks - perhaps compensating for reduced satisfaction with bonuses tied to increased productivity. In academia, however, where individuals often act as their own boss, adopting a tool with a learning curve that initially increases productivity only gradually - and might decrease job satisfaction - is likely to face resistance. Adoption may either occur much later, or not happen at all. Some could argue that adoption of such a tool may not be necessary and insist that efficiencies of industry and academic scientists are anyway too different to compare reliably. The latter is true, to some extent, but I think it still gives enough information and can help find the right direction. An interesting and simple measure to use for the quantitatively inclined could be looking at individuals’ salary. In a market economy, higher productivity leads to higher salary. A valuable goal thus could be to try to drive up academic productivity so as the reduce the size of the pay gap between industry and academia. I believe that, maybe for the first time ever, this could become possible. Not by scientists adopting the exact same tools as the industry, but, if some scientific tasks prove more amenable to be enhanced with AI, by the development of academia-specific tooling.
Impact of AI on discovery [source].
It is valuable to look at practices in companies that are proritizing building efficient and productive work cultures. An at-hand example can be Google, which even offers certificates in project management. The company implements numerous tools and approaches to enhance productivity across teams. For example, Google Docs facilitates real-time collaborative document editing, including meeting notes and SOPs. Google Sheets supports project management tasks like sprint planning, resource allocation, and KPI tracking. Integrating AI and third-party products into these tools allows further automation and workflow orchestration. 4 For strategic planning, Google’s implementation of Objectives and Key Results (OKRs) framework aligns team goals, measures progress, and drives accountability.
Meanwhile, I have been sending documents titled paper_v16_AJMHCD.docxaround.
Clearly, Google is a fairly special place; it spent $45 billion on research in 2023 5 - nearly twice as much as the Dutch and UK governments combined. But one doesn’t need Google’s budget to prioritize efficiency, especially when opportunities for improvement in academic research are abundant.
Electronic records, including lab notebooks and shared documentation, are a must. Here’s a personal example illustrating their utility: I did my PhD in a group that had virtually no shared knowledge base, despite conducting research across several fields for more than 30 years. I do not doubt that this caused knowledge and expertise to get lost or having to be rediscovered from scratch. Contrast this with a lab where I worked on an EMBO fellowship. This lab had an internal online wiki (running on Atlassian’s Confluence). Shortly after arriving, I had a better overview of the lab’s projects than I’d had in my home lab after several years. I could immediately build on prior work rather than rediscovering the wheel. In fact, my project was only possible because of the work done by previous lab members. Without electronic records, continuity without an in-person overlap and in such a short time would not have been achievable.
And this is only a very straightforward application of electronic records. Looking into the future, I don’t see why it would not enable automatic material ordering. Similarly, there should be no reason to continue to write research papers entirely from scratch. Keeping an electronic lab notebook should allow for a draft - including figures, methods and SI - to be autogenerated. And while we are at that, alongside producing a paper version for humans, I expect future automations will produce one for machines as well, with rich metadata, and protocols in standardized language that could be directly implemented by automated labs.
Recall the quotes by Drucker and Krugman. If knowledge workers’ productivity matters to us, and we consider academics as knowledge workers, then these steps seem only natural. They will free scientists’ time for more creative work and represent an implementable solution towards addressing the reproducibility crisis.
Yet, these innovations will never come withing individual scientist’s reach if the whole academic community continues to consider ‘management’ a swear word and looking down on efficiency as something that is below their standards. Of course, efficiency does not always equal productivity. But often it does, even in academia, and when this is the case, eschewing efficiency is hurting science, contributing to research irreproducibility, and wasting taxpayers’ resources. And that’s not a mission I want to be subscribing to.
Unilever decided to be proactive and slash recruitment times by using machine learning algorithms and natural language processing to analyse speech patterns during virtual interviews. [source]
https://web.stanford.edu/~chadj/IdeaPF.pdf ↩
https://www.bloomberg.com/news/articles/2024-12-02/big-pharma-s-bet-on-china-biotech-is-a-rare-trade-bright-spot ↩
https://aidantr.github.io/files/AI_innovation.pdf ↩
https://www.itsdart.com/blog/what-does-google-use-for-project-management ↩
https://www.rdworldonline.com/top-30-rd-spending-leaders-2023-big-tech-firms-hit-new-heights/ ↩
Originally published via martinholub.com.
Background: In the Netherlands, it is customary to include ‘propositions’, essentially opinionated statements defended alongside one’s doctoral work. In this post series, I am outlining the arguments supporting my propositions.
Proposition #5: The culture of academic research is not keeping pace with advancements in the for-profit sector.
I am sure I wrote the concentration somewhere… (feverishly looking through a mix of paper and electronic notes) …
I can’t get your environment to install on my machine… (I might if I try very hard, but didn’t you say it should just work?) …
You should send an email to Tracy if you’d like to get trained to use this instrument… (she works Monday, Tuesday, Thursday, but might be on vacation; anyway, she’ll send you a test, and after the training is over, you’ll fill out a form and be good to go!)…
In academic research, we are often in the business of trying to do something for the very first time. This fosters a culture where unique solutions and one-off tricks are common and celebrated. And honestly, that’s probably unavoidable and mostly alright. In many cases, that’s what I personally love. What’s not alright however, is our tendency to extend this artisanal approach to everything we do. The examples at the start of this post will be familiar to most academic scientists. There are countless others. At the university where I did my PhD, orders for laboratory supplies had to be manually reviewed by no less than four people, causing week-long delays. If you didn’t manually enter the item into the ordering system, it would never arrive. Automatic stock monitoring for common chemicals? Forget about it. Here are more examples: Want to keep your lab notebook old school, pen and paper, scribbling illegibly? Be my guest! Manually entering work hours for every day of 4+ years? Sure - great use of time! (And let’s not forget about somebody having to send reminders about it every month!). I’m far from the first to argue that research needs a productivity boost: 1
Increasing number of scientist, larger team size, and increasing complexity of scientific questions have led to decreasing per-scientist research output [source].
In the for-profit sector, efficiency is the mantra. If a solution can save even 5% of the time it takes to produce a deliverable, it’s likely to be adopted. Deployed across a team or company, such efficiency gains can have a tangible impact on the bottom line. Ignoring them means risking becoming irrelevant.
“The most important contribution management needs to make in the 21st century is […] to increase the productivity of knowledge work and knowledge workers. […] The most valuable asset of a 21st-century institution (whether business or non-business) will be its knowledge workers and their productivity.” — Peter F. Drucker (1999)
“Productivity isn’t everything, but, in the long run, it is almost everything. A country’s ability to improve its standard of living over time depends almost entirely on its ability to raise its output per worker.” – Paul Krugman (1994)
Consider this extreme, but illustrative, example of the impact of increasing productivity: In 2018, the CFDA (Chinese FDA) decided to do away with large bureaucratic burden and shifted to an ‘implied license’ system. As a consequence, the time needed for new drug application reviews decreased more than fivefold. [^2, ^3] Chinese biopharma increased candidate drug filings tenfold over the last decade as a result and has already overtaken Europe in clinical trial initiations. 2 The inability to innovate can clearly lead to a loss of competitiveness and risk of obsolescence.
More and more new medicines are being developed in China, mostly at the expense of Europe [source].
Another example: A recent experiment introduced an AI material discovery tool to a group of scientists at a large company. A 24-month follow-up compared their performance with a control group that continued without the tool. After 16 months, the treatment group discovered 44% more new materials, filed 39.5% more patents, and generated 17% more new product prototypes compared to the control group. As the authors did not neglect to remark, “these effects are large”. 3 Indeed, they are large, if not massive. Interestingly, despite increased productivity, scientists who adopted the tool reported lower job satisfaction. While it’s still early, one can easily imagine industry adopting such a tool despite its drawbacks - perhaps compensating for reduced satisfaction with bonuses tied to increased productivity. In academia, however, where individuals often act as their own boss, adopting a tool with a learning curve that initially increases productivity only gradually - and might decrease job satisfaction - is likely to face resistance. Adoption may either occur much later, or not happen at all. Some could argue that adoption of such a tool may not be necessary and insist that efficiencies of industry and academic scientists are anyway too different to compare reliably. The latter is true, to some extent, but I think it still gives enough information and can help find the right direction. An interesting and simple measure to use for the quantitatively inclined could be looking at individuals’ salary. In a market economy, higher productivity leads to higher salary. A valuable goal thus could be to try to drive up academic productivity so as the reduce the size of the pay gap between industry and academia. I believe that, maybe for the first time ever, this could become possible. Not by scientists adopting the exact same tools as the industry, but, if some scientific tasks prove more amenable to be enhanced with AI, by the development of academia-specific tooling.
Impact of AI on discovery [source].
It is valuable to look at practices in companies that are proritizing building efficient and productive work cultures. An at-hand example can be Google, which even offers certificates in project management. The company implements numerous tools and approaches to enhance productivity across teams. For example, Google Docs facilitates real-time collaborative document editing, including meeting notes and SOPs. Google Sheets supports project management tasks like sprint planning, resource allocation, and KPI tracking. Integrating AI and third-party products into these tools allows further automation and workflow orchestration. 4 For strategic planning, Google’s implementation of Objectives and Key Results (OKRs) framework aligns team goals, measures progress, and drives accountability.
Meanwhile, I have been sending documents titled paper_v16_AJMHCD.docxaround.
Clearly, Google is a fairly special place; it spent $45 billion on research in 2023 5 - nearly twice as much as the Dutch and UK governments combined. But one doesn’t need Google’s budget to prioritize efficiency, especially when opportunities for improvement in academic research are abundant.
Electronic records, including lab notebooks and shared documentation, are a must. Here’s a personal example illustrating their utility: I did my PhD in a group that had virtually no shared knowledge base, despite conducting research across several fields for more than 30 years. I do not doubt that this caused knowledge and expertise to get lost or having to be rediscovered from scratch. Contrast this with a lab where I worked on an EMBO fellowship. This lab had an internal online wiki (running on Atlassian’s Confluence). Shortly after arriving, I had a better overview of the lab’s projects than I’d had in my home lab after several years. I could immediately build on prior work rather than rediscovering the wheel. In fact, my project was only possible because of the work done by previous lab members. Without electronic records, continuity without an in-person overlap and in such a short time would not have been achievable.
And this is only a very straightforward application of electronic records. Looking into the future, I don’t see why it would not enable automatic material ordering. Similarly, there should be no reason to continue to write research papers entirely from scratch. Keeping an electronic lab notebook should allow for a draft - including figures, methods and SI - to be autogenerated. And while we are at that, alongside producing a paper version for humans, I expect future automations will produce one for machines as well, with rich metadata, and protocols in standardized language that could be directly implemented by automated labs.
Recall the quotes by Drucker and Krugman. If knowledge workers’ productivity matters to us, and we consider academics as knowledge workers, then these steps seem only natural. They will free scientists’ time for more creative work and represent an implementable solution towards addressing the reproducibility crisis.
Yet, these innovations will never come withing individual scientist’s reach if the whole academic community continues to consider ‘management’ a swear word and looking down on efficiency as something that is below their standards. Of course, efficiency does not always equal productivity. But often it does, even in academia, and when this is the case, eschewing efficiency is hurting science, contributing to research irreproducibility, and wasting taxpayers’ resources. And that’s not a mission I want to be subscribing to.
Unilever decided to be proactive and slash recruitment times by using machine learning algorithms and natural language processing to analyse speech patterns during virtual interviews. [source]
https://web.stanford.edu/~chadj/IdeaPF.pdf ↩
https://www.bloomberg.com/news/articles/2024-12-02/big-pharma-s-bet-on-china-biotech-is-a-rare-trade-bright-spot ↩
https://aidantr.github.io/files/AI_innovation.pdf ↩
https://www.itsdart.com/blog/what-does-google-use-for-project-management ↩
https://www.rdworldonline.com/top-30-rd-spending-leaders-2023-big-tech-firms-hit-new-heights/ ↩
scimartin ('JJ')
scimartin ('JJ')
No comments yet