Can GPT-3 write an academic paper on itself, with minimal human input?

Posted , updated on

Brilliant. The short answer is, of course, yes, and it doesn’t do a bad job of it. This is conceptual art of the highest order.

This is the preprint of a paper written by GPT-3 (as first author) about itself, submitted to “a well-known peer-reviewed journal in machine intelligence”. The second and third authors provided guidance about themes, datasets, weightings, etc, but that’s as far as it goes. They do provide commentary as the paper progresses, but they tried to keep that as minimal as needed, so that the paper could stand or fall on its own merits. The paper is not too bad. A bit repetitive, a bit shallow, but it’s just a 500 word paper- hardly even an extended abstract – so that’s about par for the course. The arguments and supporting references are no worse than many I have reviewed, and considerably better than some. The use of English is much better than that of the majority of papers I review.

In an article about it in Scientific American the co-authors describe some of the complexities in the submission process. They actually asked GPT-3 about its consent to publication (it said yes), but this just touches the surface of some of the huge ethical, legal, and social issues that emerge. Boy there are a lot of those! The second and third authors deserve a prize for this. But what about the first author? Well, clearly it does not, because its orchestration of phenomena is not for its own use, and it is not even aware that it is doing the orchestration. It has no purpose other than that of the people training it. In fact, despite having written a paper about itself, it doesn’t even know what ‘itself’ is in any meaningful way. But it raises a lot of really interesting questions.

It would be quite interesting to train GPT-3 with (good) student assignments to see what happens. I think it would potentially do rather well. If I were an ethically imperfect, extrinsically-driven student with access to this, I might even get it to write my assignments for me. The assignments might need a bit of tidying here and there, but the quality of prose and the general quality of the work would probably result in a good B and most likely an A, with very little extra tweaking. With a bit more training it could almost certainly mimic a particular student’s style, including all the quirks that would make it seem more human. Plagiarism detectors wouldn’t stand a chance, and I doubt that many (if any) humans would be able to say with any assurance that it was not the student’s own work.

If it’s not already happening, this is coming soon, so I’m wondering what to do about it. I think my own courses are slightly immune thanks to the personal and creative nature of the work and big emphasis on reflection in all of them (though those with essays would be vulnerable), but it would not take too much ingenuity to get GPT-3 to deal with that problem, too: at least, it could greatly reduce the effort needed. I guess we could train our own AIs to recognize the work of other AIs, but that’s an arms war we’d never be able to definitively win. I can see the exam-loving crowd loving this, but they are in another arms war that they stopped winning long ago – there’s a whole industry devoted to making cheating in exams pay, and it’s leaps ahead of the examiners, including those with both online and in-person proctors. Oral exams, perhaps? That would make it significantly more difficult (though far from impossible) to cheat. I rather like the notion that the only summative assessment model that stands a fair chance of working is the one with which academia began.

It seems to me that the only way educators can sensibly deal with the problem is to completely divorce credentialling from learning and teaching, so there is no incentive to cheat during the learning process. This would have the useful side-effect that our teaching would have to be pretty good and pretty relevant, because students would only come to learn, not to get credentials, so we would have to focus solely on supporting them, rather than controlling them with threats and rewards. That would not be such a bad thing, I reckon, and it is long overdue. Perhaps this will be the catalyst that makes it happen.

As for credentials, that’s someone else’s problem. I don’t say that because I want to wash my hands of it (though I do) but because credentialling has never had anything whatsoever to do with education apart from in its appalling inhibition of effective learning. It only happens at the moment because of historical happenstance, not because it ever made any pedagogical sense. I don’t see why educators should have anything to do with it. Assessment (by which I solely mean feedback from self or others that helps learners to learn – not grades!) is an essential part of the learning and teaching process, but credentials are positively antagonistic to it.

Originally posted at: https://landing.athabascau.ca/bookmarks/view/14216255/can-gpt-3-write-an-academic-paper-on-itself-with-minimal-human-input

Computer science students should learn to cheat, not be punished for it

Posted , updated on

This is a well thought-through response to a recent alarmist NYT article about cheating among programming students.

The original NYT article is full of holy pronouncements about the evils of plagiarism, horrified statistics about its extent, and discussions of the arms wars, typically involving sleuthing by markers and evermore ornate technological fixes that are always one step behind the most effective cheats (and one step ahead of the dumber ones). This is a lose-lose system. No one benefits. But that’s not the biggest issue with the article. Nowhere does the NYT article mention that it is largely caused by the fact that we in academia typically tell programming students to behave in ways that no programmer in their right mind would ever behave (disclaimer: the one programming course that I currently teach, very deliberately, does not do that, so I am speaking here as an atypical outlier).

As this article rightly notes, the essence of programming is re-use of code. Although there are certainly egregiously immoral and illegal ways to do that (even open source coders normally need to religiously cite their sources for significant uses of code written by others), applications are built on layer upon layer upon layer of re-used code, common subroutines and algorithms, snippets, chunks, libraries, classes, components, and a thousand different ways to assemble (in some cases literally) the code of others. We could not do programming at all without 99% of the code that does what we want it to do being written by others. Programmers knit such things together, often sharing their discoveries and improvements so that the whole profession benefits and the cycle continues. The solution to most problems is, more often than not, to be found in StackExchange forums, Reddit, or similar sites, or in open source repositories like Github, and it would be an idiotic programmer that chose not to (very critically and very carefully) use snippets provided there. That’s pretty much how programmers learn, a large part of how they solve problems, and certainly how they build stuff. The art of it is in choosing the right snippet, understanding it, fitting it into one’s own code, selecting between alternative solutions and knowing why one is better (in a given context) than another. In many cases, we have memorized ways of doing things so that, even if we don’t literally copy and paste, we repeat patterns (whole lines and blocks) that are often identical to those that we learned from others. It would likely be impossible to even remember where we learned such things, let alone to cite them.  We should not penalize that – we should celebrate it. Sure, if the chunks we use are particulary ingenious, or particularly original, or particularly long, or protected by a licence, we should definitely credit their authors. That’s just common sense and decency, as well as (typically) a legal requirement. But a program made using the code of others is no less plagiarism than Kurt Schwitters was a plagiarist of the myriad found objects that made up his collages, or a house builder is a plagiarist of its bricks.

And, as an aside, please stop calling it ‘Computer Science’. Programming is no more computer science than carpentry is woodworking science. It bugs me that ‘computer science’ is used so often as a drop-in synonym for programming in the popular press, reinforced by an increasing number of academics with science-envy, especially in North America. There are sciences used in computing, and a tiny percentage of those are quite unique to the discipline, but that’s a miniscule percentage of what is taught in universities and colleges, and a vanishingly small percentage of what nearly all programmers actually do. It’s also worth noting that computer science programs are not just about programming: there’s a whole bunch of stuff we teach (and that computing professionals do) about things like databases, networks, hardware, ethics, etc that has nothing whatsoever to do with programming (and little to do with science). Programming, though, especially in its design aspects, is a fundamentally human activity that is creative, situated, and inextricably entangled with its social and organizational context. Apart from in some research labs and esoteric applications, it is normally closer to fine art than it is to science, though it is an incredibly flexible activity that spans a gamut of creative pursuits analogous to a broad range of arts and crafts from poetry to music to interior design to engineering. Perhaps it is most akin to architecture in the ways it can (depending on context) blend art, craft, engineering, and (some) science but it can be analogous to pretty much any creative pursuit (universal machines and all that).

Address of the bookmark: https://thenextweb.com/dd/2017/05/30/lets-teach-computer-science-students-to-cheat/#.tnw_FTOVyGc4

Original page

Over two dozen people with ties to India’s $1-billion exam scam have died mysteriously in recent months

Posted , updated on

“… the scale of the scam in the central state of Madhya Pradesh is mind-boggling. Police say that since 2007, tens of thousands of students and job aspirants have paid hefty bribes to middlemen, bureaucrats and politicians to rig test results for medical schools and government jobs.

So far, 1,930 people have been arrested and more than 500 are on the run. Hundreds of medical students are in prison — along with several bureaucrats and the state’s education minister. Even the governor has been implicated.“

A billion-dollar fraud scheme, perhaps dozens murdered, nearly 2000 in jail and hundreds more on the run. How can we defend a system that does this to people? Though opportunities for corruption may be higher in India, it is not peculiar to the culture. It is worth remembering that more than two-thirds of high school Canadian students cheat (I have seen some estimates that are notably higher – this was just the first in the search results and illustrates the point well enough):

According to a survey of Canadian university & college students:

  • Cheated on written work in high school 73%
  • Cheated on tests in high school 58%
  • Cheated on a test as undergrads 18%
  • Helped someone else cheat on a test 8%

According to a survey of 43,000 U.S. high school students:

  • Used the internet to plagiarize 33%
  • Cheated on a test last year 59%
  • Did it more than twice 34%
  • Think you need to cheat to get ahead 39%

Source: http://www.cbc.ca/manitoba/features/universities/

When it is a majority phenomenon, this is the moral norm, not an aberration. The problem is a system that makes this a plausible and, for many, a preferable solution, despite knowing it is wrong. This means the system is flawed, far more than the people in it. The problems emerge primarily because, in the cause of teaching, we make people do things they do not want to do, and threaten them/reward them to enforce compliance. It’s not a problem with human nature, it’s a rational reaction to extrinsic motivation, especially when the threat is as great as we make it. Even my dog cheats under those conditions if she can get away with it.  When the point of learning is the reward, then there is no point to learning apart from the reward and, when it’s to avoid punishment, it’s even worse. The quality of learning is always orders of magnitude lower than when we learn something because we want to learn it, or as a side-effect of doing something that interests us, but the direct consequence of extrinsic motivation is to sap away intrinsic motivation, so even those with an interest mostly have at least some of it kicked or cajolled out of them. That’s a failure on a majestic scale. If tests given in schools and universities had some discriminatory value it might still be justifiable but perhaps the dumbest thing of all about the whole crazy mess is that a GPA has no predictive value at all when it comes to assessing competence.

Address of the bookmark: http://www.theprovince.com/health/Over+dozen+people+with+ties+India+billion+exam+scam+have+died/11191722/story.html