University on Michael’s Domain

I am focusing on the wrong things

Wed, 22 Apr 2026 00:00:00 +0000

This is a rant, but hear me out. Instead of working on how to utilize AI in my own field, I am:

The Finnish 1000 PhDs pilot

Tue, 28 Nov 2023 00:00:00 +0000

The Finnish government is going to finance 1000 additional new fully funded doctoral positions, which would start in the academic year 2024/2025.The money (85000€/year for each PhD student) will be competitively distributed between Finnish universities, and some of that money will be used to pay for the salary of the PhD students for three years. Perhaps one of the triggers for this initiative is that Finland has been falling behind in the OECD statistics in the number of highly educated workers. The question is whether this one-time expense will really increase the level of education or only boost the numbers.The average duration of a PhD education is around seven years in Finland. In order to pull off a 3-year PhD education at such a large scale, it is expected that the requirements for the PhD will be lowered. As a matter of fact, the requirements have already been lowered, but further “easings” might still come. Together with these, universities are aiming to streamline the kafkaesque absurd and medieval administration process, which can easily take half a year from the time of completing the last manuscript to receiving the actual PhD certificate.Many other outlets have reported about this pilot; here’s one that is available also in English: https://acatiimi.fi/2023/11/28/for-once/Although most people are happy about the additional money, almost everybody is sceptical to some degree. Good intentions are not enough, and there are many serious issues with this pilot:

Turing test with exam answers: Can I sniff out the AI?

Thu, 29 Dec 2022 00:00:00 +0000

In the Finnish daily newspaper Helsingin Sanomat, GPT-3 and Laura Ketonen from the University of Jyväskylä discuss how AI will affect student assessment ( Tekoäly ravistelee opiskelijoiden arviointia , English: Artificial intelligence shakes up student assessment). One of the obvious “applications” of AI would be exam cheating.Laura’s opinion piece provoked quite a few reactions. Some answered that AI produces hollow text . Others mentioned that AI does not have courage and initiative . It feels like we, as a human species, will soon be struggling with a generalized inferiority complex.In a natural science exam situation at the BSc or MSc level, the initiative is with the teacher, and the students are, by definition, reactive. How do you define hollow text when asking to explain glycoengineering? In life science, courage and initiative are essential features for any aspiring scientist. But without a solid body of knowledge as the foundation, courage and initiative rarely will lead to impactful new developments. At this moment, we cannot yet outsource this foundation to AI. Hence, we require from students a solid body of knowledge from which courage and initiative can draw to flourish. Perhaps, low-hanging fruits can be picked with courage and initiative alone. But solving today’s problems requires, in addition, a solid body of knowledge (“standing on the shoulders of giants”).When judging AI, we likely make the same mistakes as humans do with all new technologies: We overestimate its impact in the short run but underestimate its impact in the long run. Over and again, (narrow) AI has managed to break into fields that were previously reserved for humans. There is no reason to assume that this development will not continue. It may well be that the presently dominant approaches to AI (deep learning, machine learning, neural networks, statistical approaches) will not lead to major future breakthroughs toward human-like general AI, perhaps even self-aware AI. We don’t even know how human consciousness comes about. On the other hand, I agree with Daniel Dennett’s idea that there might be no hard problem of consciousness: Consciousness is what you get when you have solved all the easy problems. In any case, if a non-self-aware AI becomes indistinguishable from a self-aware AI, what’s the difference, and how would we be able to know? I am sure that at one point, we will also be able to outsource our knowledge completely. With Google, we have even started to make baby steps in this direction, but we need a better computer-brain interface to fully embrace the concept of knowledge outsourcing. Deep Blue beat international grandmaster Garry Kasparov in 1997. I did not want to accept this human defeat and declared the game unfair: Deep Blue crashed and needed to be rebooted once during the 6-game tournament, which - in my opinion - was equivalent to the human dying during the game (or at least being reanimated by CPR). In 2016, computers surpassed humans in the game Go, and today, they beat humans even at games like poker, requiring sophisticated psychological trickery like bluffing. To test how AI compares to students in an exam situation, Patrick added one AI-generated answer to the students’ responses in a recently written exam. This idea gave the grading an exciting spin! I think I knew what the AI answer was. However, I had played with GPT-3 before and knew that - untweaked - AI is overly correct and systematic in addressing questions.E.g., in a 3-part question, the AI normally systematically answers all three parts. Also, the AI did not know the exact content of my lectures, which made identifying the AI answer relatively easy. Patrick still needs to reveal whether I identified the AI correctly…The AI (or what I identified as the AI) performed worse than the best student, but overall it did really well. Clearly, our future exam questions need to focus even more on critical thinking and fictional examples, which the AI cannot know from its vast pool of training material.But given the increasing amount of information that AI will access in the future, AI can rely on other people’s critical thinking. The next generation of AI (ChatGPT-4 and Google’s LaMDA) is already waiting for public release… With some tweaking, Patrick could have easily fooled me. Maybe he did already… If he instructed the AI to break the answering pattern intentionally and if he did train GPT-3 with my lecture slides as reference material, all my bets are off.I’ll let you know whether I succeeded in identifying the AI correctly when Patrick reveals which answers were AI-generated. UPDATE: I (as well as the other teachers) did correctly identify the AI answers. The general opinion among us was that if we had not been alerted to the fact that one of the answers was AI-generated, we would not have noticed. The quality of the AI answers depended on the type of question. It appeared that (at this moment) a relatively easy way to throw off the AI is to give information needed to answer the question in a graphical format. But I am sure AI will learn to integrate images with text very soon.

University Pedagogy (UP2.2, 2019): Assessment of Learning and Giving Feedback

Fri, 02 Oct 2020 00:00:00 +0000

Becoming a qualified teacher in FinlandThe degree requirements for the 60-credit module in University Pedagogy (UP) consist of 25 credits of basic studies and 35 credits of intermediate studies. Although UP focuses on teaching at the university, the completed module qualifies you also for teaching at schools. The Pedagogy Center of the University of Helsinki organizes the courses of the basic study modules (25 ECTS credits) in English. However, if you want to complete the whole module, you will need to take courses which are only offered in Finnish or Swedish. That means you cannot become a qualified (“pätevä”) teacher without knowing Finnish or Swedish. I do not know whether my Finnish is meanwhile good enough to be able to do that. For the time being I stick to the courses taught in English.**University Pedagogy 2.2 (2019)**In the 2019 autumn term, I took the UP2.2 course (without taking UP2.1 first). We had a very nice teacher (Jokke Häsä), but I still don’t understand why he switched from doing Mathematics to teaching Pedagogy. Certainly a gain for pedagogy since he understands the statistics and we had some interesting forth-and-back about the statistics of the Johnston & Miles paper that we were reading. They paper is also attached to this post and the reporting about their statistical methods leaves much room for interpretation. I actually sent an email to the first author, but she had just retired a few months ago. Although I did not get an failure-of-delivery error, she perhaps did not anymore feel the need to respond.Blind flyingFor me, one of the biggest issues was, that I had nothing to compare too when I was preparing my assignments. When I am writing manuscripts for journals, I have thousands of examples which I can study and to which I can compare. So I decided during the course that I would make all my assignments public. Embarrassment included. But maybe some future participants of UP2.2 will have something to compare to. Sort of an exemplar. Not perfect, but passable. I only post my own assignments.**All my assignments available under CC0 (= no strings attached)**All this stuff is shared under the CC0 : Use it as you wish, no need to acknowledge me. The group assignment is the only thing missing here since I would need to get everybody’s green light, which is not going to happen. I should have done this 9 months ago, but I simply forgot. There was just too much other stuff happening.I did ok in the course, but I still don’t know anythingWhat grade did I get? Our group assignment got a 5 (on the scale of 1 to 5, where 5 is the best grade) and my own stuff a 4. Thanks to all the great members of team Carbonara! That averaged out as a 5 for the whole course. For some strange reason, the Moodle course page still shows that I only completed 66% of the course assignments, but WebOodi shows that I received the credits. So much for the wonders of our IT systems…

University rankings (and how to manipulate them)

Mon, 29 Aug 2016 00:00:00 +0000

Whow. Helsinki University was again able to improve its position in the Shanghai Ranking and is now number 56. However, even the University’s own press release had kind of a gloomy tune as if this was the last blossom before the final decay. It is true that the basis of the current success has patiently been built by the researchers over the last decades. Nevertheless, when politicians now draw the conclusion that the cuts of the educational budget are responsible for this success, they are probably right.Perhaps the cuts actually did contribute to the surge in productivity: All the scientists that unexpectedly were suddenly unemployed obviously continued to work furiously on their current projects to get their current manuscripts ready for publication. Papers are the overwhelmingly dominant currency in the academic job market. And these unemployed scientists might publish faster than before since unemployed scientists have no teaching duties…However, more importantly, the Helsinki numbers of the individual Shanghai Ranking indicators for 2015 and 2016 differ significantly only the per capita academic performance (publications/number of academic staff) and the number of highly cited scientists. Since the job cuts at Helsinki Universities have been going on already for a while even before the massive cuts of this spring, it appears reasonable to claim, that the 56th position is partly due to the job cuts, which did influence directly and rapidly the number of academic staff.The Shanghai Ranking emphasizes research, whereas in other rankings, Helsinki has even been dropping. E.g. in the QS World University Rankings , Helsinki dropped from position 67 (2014/15) to 96 (2015/16). Between these two rankings, the QS team made “improvements” to their ranking system and these were almost exclusively responsible for the drop if you look at the data closely. They “normalized” the citation ratio per faculty according to the subject, and life science got badly punished by being a “high citation field”. Very simply put, a citation to a paper in Arts and Humanities now counts about 40 times as much as a citation to a paper in Life Science/Medicine. Each of the 5 big fields (Arts & Humanities, Engineering & Technology, Life Sciences & Medicine, Natural Sciences, Social Sciences & Management) contributes now equally with 20% to the final rating in the citation per faculty category (detailed explanation of the normalization from here: http://content.qs.com/qsiu/Faculty_Area_Normalization_-_Technical_Explanation.pdf).Why on earth do they punish a field for having a bigger impact? Papers in the humanities are not cited as much as papers in life science. Maybe this could be due to the fact that their average publication is not as relevant for their own field as is a life science publication for the life science field? I agree that there are cultural differences between fields, but who is to say how much is due to such cultural differences versus low quality or irrelevant research? Exactly that’s what QS World University Ranking’s did by making up “normalization” factors essentially claiming equal relevance.However, all such rankings are easily manipulated. E.g. the citation to faculty (member) ratio can easily improved by outsourcing services that were previously performed by faculty scientists (and that’s what Helsinki University is just doing at the moment). Similar to the impact factor for journals, university rankings are losing their relevance if the goal is to score high and not to excel at research and teaching ( Goodhart’s law : When a measure becomes a target, it ceases to be a good measure).The Shanghai Ranking has been criticized much. It undervalues teaching since teaching performance doesn’t enter directly the calculation at all, and it favours big universities since 50% of the rating are based on sheer numbers which are not normalized to account for the size of the university. Nevertheless it it the only reproducible (and therefore scientific) dataset from the three big rankings ( https://link.springer.com/article/10.1007/s11192-012-0801-y ) since both the QS and the Times Ranking rely heavily on opinion and reputation surveys, which are notoriously difficult to reproduce.What is the take-home message? Helsinki University is probably as good as it used to be during the last decade and the big fluctuations are just artifacts resulting from changed ranking methodologies or resulting from administrative adjustments to the financial realities brought upon us by the Finnish voters in the 2015 parliamentary elections.

Moving from Paper to Electronic Lab Notebooks

Tue, 26 Jul 2016 00:00:00 +0000

Electronic lab notebooks (ELN) are the future. While this statement appears self-evident, it is not clear what software will establish itself and which ones will disappear. When I first checked the ELN market about three years ago, there were only a handful of solutions. Now there are already maybe about a hundred different software companies that try to cash in on the trend. And what is even more worrisome is the fact that the list of vendors contains already about 20 abandoned products…Thus, settling on a software is a serious decision for a lab and even more so for a whole institute. Many points need to be taken into consideration. Just to mention a few of them:- How sure is it, that the software is available and supported in 10 years? If the company disappears, your data might as well. All vendors will claim that your data is safe. You should be realistic given the widespread deception of customers and authorities by companies. Just think about VW…- What is the upfront investment and what are the running costs? If they charge you for storage space, the price might initially be low, but don’t underestimate how much data you will accumulate with all the new high res imaging and sequencing technologies. Once you are doing superresolution, you easily can gather a many GB each week. E.g. SciNote includes 100 GB of storage for 2 teams, which can fill up within a few weeks depending what type of research you are doing. I was trying to figure out how to buy storage beyond 100 GB, but I could not figure out how. I hope that the usability of their ELN is not on the same level… - How free is your data? Can you easily (and in an automated fashion in regular intervals) pull out your data if you need to? In what format can you get your data? If it is only PDF or Word files, just forget about it. You should get a real database dump that can be used to populate other systems. I would not trust any company to tell me ahead of time that they will go belly-up soon and that I should rescue my data…- Where are the data stored physically? This determines e.g. what law the data falls under and who has access to it. How much can you trust a small company overseas to keep your data private and secure? Let’s remind everybody of the data breaches that affected Sony, Linkedin, IRS, T Mobile, UPS, JP Morgan Chase… Some vendors offer in-house hosting, but the pricing seems to be too high for many academic labs to make this a viable alternative.- Open Source seems to be a solution to may of these problems. However, with diminishing computing infrastructure support from the university, the maintenance requirements of such a system need to be very low in order to make this feasible.- If it is marketed as “free”, what do they mean with free? Often, the “free” offering is not sufficient for any serious work and hence misleading. “Free” as in “free beer” is also not enough… Many labs need also the freedom to adapt the software to their specific needs, which excludes proprietary solutions without public APIs.In my lab we have been testing a few of the available solutions. Our emphasis was on open source solutions since only Open Source solutions make it easy to test the software extensively. So far we have tried:

FPLC Protein purification course

Mon, 04 Jan 2016 00:00:00 +0000

Eight postgraduate students registered for the FPLC protein purification course , which took place in December. If I learned anything, than that protein purification during a course should be done ALWAYS with a protein and a protocol, that has been used before successfully MANY times. Student-provided proteins are a great source for learning, but the time restraints of a course format did not allow us to finish the purification of these proteins during the course.Technical Problems with the new Äkta Avant 25In addition, the Äkta Avant 25 , that we newly purchased from GE Healthcare this summer, broke TWICE during the course. First the controlling computer broke (RAID failure). HP delivered the replacement drive within 24 hours and after the RAID had rebuilt itself, we could continue the course. However, during the first run after this incident, the Äkta ran into an overpressure problem. We identified a faulty flow restrictor as the cause, but we did not want to continue as we have had severe problems with air bubbles in previous runs. Therefore we performed the runs on the old Äkta Explorer 100 . GE Healthcare quickly had their service engineer check out the system, but since he did not have a spare with him, we had to wait until Monday 14.12. until the Äkta Avant was again fully functional.What we purified: soluble VEGFR-3 (VEGFR-3/Fc) and HepsinWe did purify soluble human VEGF receptor-3 (the first three domains of its extracellular domain connected to the constant Fc part of human IgG). This is a purification that we have done many times. It is equivalent to the purification of antibodies using Protein A sepharose.We had prepared in advance conditioned cell culture medium. We produce most of our proteins in insect cells (mostly Drosophila S2) and the VEGFR-3/Fc had been secreted by the S2 cells into the medium after induction of the metallothionein promoter with 1 mM Cu2+ for about 4.5 days. The preparation of the medium for purification consists only of 1) getting rid of the cells by centrifugation and 2) filtration to remove precipitates and other small particles that might clog the column. There is no need to adjust the pH.Rapid neutralization after low pH elution IS IMPORTANTWe ran the medium over a disposable 5-ml HiTrap recombinant Protein A column over night at about 1ml/min and eluted with a low pH buffer. The eluted 2-ml fractions were immediately neutralized with 400µl 1M Tris pH 8.5. Here we made a small mistake during one of the two purifications. GE Healthcare had not released the casettes for the 5-ml-collection tubes (they still have not done so even though they did promise them already for September 2015) and we used 15-ml Falcon tubes to collect 2-ml fractions, into which we had pre-aliqotted 400 µl of the neutralization solution. However, the mixing in these tubes was not efficient and the prolonged exposure to low pH resulted in a partial damage to our protein. This can be seen when comparing the size exclusion chromatograms of Group 2 versus Group 4: For Group 4 the first peak (aggregated protein) is much larger and more heterogenous compared to the same peak for Group 2.In fact, when VEGFR-3/Fc is eluted by low pH from protein A columns, it always precipitates at higher concentrations soon after elution, but dissolves again upon neutralization. This did not happen in the fractions 5.A.3 and 5.A.4 (Group 4) due to the inefficient mixing of elutate and neutralization buffer in the 15-ml-Falcon tube (the fraction size of 2 ml was probably to blame as well; 1 ml would have been better). During the run for Group 2, we removed the tubes immediately after the run had ended and thereby mixed the buffers, while for Group 4 the run finished during night time and the eluate remained largely unmixed until the morning.Alternatively, we could have eluted with a highly concentrated chaotropic salt at near-neutral pH (which we’ll do next time in case we have an automated run where the elution happens in the middle of the night). Pierce offers a (proprietary) “gentle” elution buffer of pH 6.6. I don’t know the Pierce buffer composiiton, but it is a highly concentrated solution of some chaotropic salt. 3M potassium/sodium thiocyanate or 4M magnesium chloride in buffered solutions around pH 7 are frequently used chaotropic salts for this purpose.HepsinThe student-provided proteins were challenging. First, their concentrations in the starting material was very low. While we can see clearly the protein when the VEGFR-3/Fc conditioned medium is run on a PAGE gel and stained with Coomassie, no such band is visible for the Hepsin. In addition, it appeared that a significant fraction of the protein seems not to contain the histag (anymore) and therefore is not captured with the first purification step. The fraction of Hepsin-H6 that does bind to the Ni2+ sepharose elutes already at an imidazole concentration of 20 mM, which makes washing the column challenging. The Hepsin with the longer histag (H10) survives the 20 mM imidazole wash, but it suffers also from low expression levels.Below are the chromatograms of the individual runs and the annoteded images of the Comassie-stained PAGE gels. The detailed protocol for the operation of the Äkta Avant 25 is still under preparation…

The new Finnish government cuts university funding by around 500 million € (or not?)

Thu, 11 Jun 2015 00:00:00 +0000

UPDATE: It appears that some of the party representatives that were the source of the information in this article have not been reflecting the official positions of their respective parties (in other words: they were talking BS without having any authorization to do so). In reality, the coming cuts and their sizes are going to be determined only during this coming fall season.