How can we udpate the "AAU"-model?

Tuesday evening, I had the chance to dine with two professors from Aalborg University, where we had a chance to discuss the subject of how a masters educations could be structured, if anything was possible!

Well, we quickly reached consensus that the current system is flawed, and is run on a "we know best, what the students and the industry wants". Well, do you now? I recall on my masters degree in Control and Automation, that we had courses, which seemed most of all like a mis-fit. Courses which didn't interest any of us, but we "had" to follow, cause it was part of our curriculum.

One idea, was to just liberalize the s*** out of this old-fashioned silo way of thinking, and instead of having a defined set of rules, let the students themselves define the rules. Instead of forcing everybody to follow the same path, just let people assemble their own education, like LEGO blocks.

The idea couldn't leave my head, which is why i'm now writing this piece. Not only would this create more motivated students (by removing the courses that you don't want to follow), but it could also create a much more dynamic curriculum. Courses on which less than 5/6 students had signed up, would not be run - which would aid in removing courses which didn't interest any students, and really interesting courses, could have as many as 100+ students per course! I know this is roughly how they run their universities elsewhere, but why are we not pursuing this idea at Aalborg University?

"But Rasmus, aren't you forgetting that at freshman on 7th semester won't be able to understand Lyapunov theory"? I haven't forgot it, i just haven't mentioned it yet. This could easily be overcome, by setting requirements to the individual courses as: "If you want course B, you need course A". In that sense, you still provide the students with the option, but you can also guide them in the direction you want. Toss in all the PhD courses, and I think we're moving towards a system, that could provide some really interesting candidates, not only for academia, but also for the industry! At least, if I was able to, my masters degree would have been a lovely combination of control systems theory, mixed with machine learning, AI and software development!

I think what i'm aiming at, is that its time to give up on hopeless conventional systems. Flip everything upside down - and give students the opportunity to create their own educations.

And on that note - I also have an idea for a cross-disciplinary free-study activity, but i'll keep you posted on that, once we get settled in Aalborg!

All the best,

Whats the fuzz with AI?

A lot of people blindly swallow the medias portrait of Artificial Intelligence and what it can be used for, and I get why. The mathematics involved in Neural Networks are so complex, that not even the people who invented it knows exactly what is going on. By all means, it is artificial.

But even if you haven't obtained an honorary doctorate in mathematics for your findings in set-theoretic definitions of natural numbers, I think it's important that you learn the basics, and then use this to filter the  bs-image the media portrays it as.

First, you need to remember the fact that a computer is able to do repetitive tasks over and over, for ever and ever. It will keep doing the same task, without getting tired, and without asking for a break. If you feed it the same inputs, it will produce the exact same outputs. Humans could essentially do the same, but as we all know, humans are prone to, at some point, make an error. If you're working as an archivist, your ability to file stuff, will on any given day be influenced by many factors. Did you sleep well, was your coffee to hot, is your girlfriend an idiot, your age, your vision, the sun, etc.. I think you get it, the list goes on! Whereas a computer isn't influenced by anything.

So, when someone states that an AI algorithm is going to solve everything, think bullshit. Two things:
1) The AI is only as good as the guy who developed it.
2) AI is currently not able to "learn".

I won't bother addressing the prior as this is obvious. But number 2 is interesting. People use the term "learn" about algorithms. What they fail to mention, is that it has nothing to do with learning. It has everything to do with finding the best response based on the inputs. Now keep in mind, all responses have to be pre-programmed into the algorithm. In that sense, it does not learn, it just searches through possible options, and then selects the "best" response. Now, if there is no "best" reponse, the algorithm can do anything. In the AlphaGo documentary on Netflix (seriously watch it, its really really good), the algorithm does something stupid in the 4th game against Lee Sedol. Here AlphaGo chose the wrong option. This move would have never been made by a human being.

What i'm trying to get at, is that AI is really good at solving typical repetitive tasks, and at it, it will beat any human, any day! But the inputs that cannot be fixed by the algorithm, still needs the human mind. And the last bit, is what the media (and a lot of the C-level executives that walk around) forgets. I hope this small piece of writing have shed a little light on the matter.

I'm really fond of AI, and it what i do for a living, but knowing what it cannot be used for, is more important than knowing what it can be used for! Because it surely isn't a tool for everything.

I'll finish this piece of with a quote.

"If you only have a hammer, everything become nails"

Have a good weekend!