The potential of algorithms in education

[Picuino]

I think algorithms are going to come in strong in education. I am not sure if this change will be a good thing or not, but I am convinced that it is a revolution that will be upon us sooner rather than later.

The time has come to exploit the potential of algorithms in education:
https://medium.com/enrique-dans/the-time-has-come-to-exploit-the-potential-of-algorithms-in-education-c2615468d1eb
"""
Bill Gates hinted at a recent event on the relationship between technology and education that he sees a future for primary education based on using machine learning assistants to teach children basic skills such as reading and writing that can adapt to the level and progress of each child, making classes entertaining and improving the educational experience in the process.

The reality is that generative conversational assistants, hallucinations aside, could already be used to create personalized visual content to teach children to read. In fact, it could also be used for home schooling or to deliver private lessons in many other subjects for practically any child, especially for those from low-income households.

Given that generative algorithm technology is still at an early stage, many people will ask whether it’s a good idea to use it for something as sensitive as children’s education. Will we see children learning to read more or less unsupervised with an algorithmic teacher, and perhaps obtaining better results than its human counterpart? Obviously, the learning process isn’t only about content in a given format; social interaction and skill development are important, but it is worth looking more closely into whether we have reached the point where we can consider robotic teachers for our children. And why not at other levels? Wouldn’t participation, for example, key to children understanding concepts, be improved if they could interact with a teacher who answers their questions individually and who keeps detailed record of each student’s activity, affinities and interests, without them having to raise their hand and wait for the teacher to authorize them to take part in the class?

Generative algorithms trained with hundreds of millions of parameters and applied to language have already amply demonstrated their mastery of language and interaction. Those dedicated to image generation, trained with large repositories of labeled images, or those that can generate any type of voice, for example, make it possible to develop high quality educational or reinforcement content, where before we needed a whole set of infrastructure to create this type of material.

Are we already at a level where we can consider incorporating algorithms into something as sensitive as education? Would it improve results? Wouldn’t it make sense to explore it?
"""

[rstofer]

We have tried every half-assed scheme for education except 'algorithms' and nothing is working (in US) so why not give it a shot.  It couldn't produce worse results than what we are getting with modern thinking.

Better idea:  See what Singapore is doing and copy that.  Smartest people I ever worked with.

Another idea:  Go back to the early '50s when we educated the people who put us on the Moon.

[Picuino]

I don't really believe that algorithms are going to give us a better education. It's like other technologies, which make things easier but not necessarily better.
Algorithms can make obligatory education much cheaper and higher education much more affordable. That's an unanswerable argument for changing the system. If it's cheaper, it's done. And protests are ignored.
But just because algorithms are efficient and inexpensive doesn't mean that learning with them is better. I think they could give a good result if they were well tuned to achieve a good education in the broad sense, but the real motivation behind the development of such algorithms is going to be economic. It will serve to lower costs and make money for a few private companies.

[golden_labels]

We have tried every half-assed scheme for education except 'algorithms'

Algorithms were tested too. See teaching machines in first half of the 20th century. Reiterated each generation. My times being connecting holes with wires to match answers to questions, educational video games, and multimedia encycolpedias. The next revolution in teaching, praised by self-appointed experts, people running around with a new shiny hammer and searching for nails to drive, in the end addressing non-existent problems.

Do I have any reason to think the story will not repeat with ML algorithms, at least as they are today and can be in the near future? </rhetorical_question>

At some point in the future, with much more complex smortnet architectures, I believe automated teaching will be possible. But not with stuff like GPT, even if it improved tenfold. It’s not a matter of performance, but a mismatch between their operation and the problem to solve.

Where existing machine learning solutions could be used as auxiliary tools in teaching? Spotting grammatical or spelling errors. But… this is already solved with a much more appropriate algorithms. Indicating errors in solutions. But that would be of help to teachers, not to kids. Unless you want to teach children to never trust the machine, leading to a situation resembling that of when electronic calculators were introduced. The calculator ladies were using a device and then repeating calculations by hand… to ensure there was no mistake in the machine. Also note these are classifiers, not generators. For generators I see even less use. Perhaps providing hints to make learning more effective? Helping with pronunciation by emphasizing mistakes? Being a simple data retrieval engine with natural language interface (in 1/3 of cases answering “please ask your teacher”)?

The thing changes a bit, if talking about mature people. Ones that are conscious about their learning process and can be properly critical of feedback they receive. But not schoolchildren.

Coincidently: today I given to ChatGPT an example of an implication paradox and asked it to explain, why this is valid despite intuitively being wrong. It did beat 99% of the human population, because it did not answer “what the hell is an implication.” Other than that it simply started repeating that it is ok, because implication is defined like this in logic. So much for being helpful in teaching. And I say that, despite being aware that majority of maths teachers would not be able to give a better answer. But if we are talking about replacing teachers, the algorithm should do considerably better.

[John Nielsen]

I think algorithms are going to come in strong in education. I am not sure if this change will be a good thing or not, but I am convinced that it is a revolution that will be upon us sooner rather than later.

The time has come to exploit the potential of algorithms in education:
https://medium.com/enrique-dans/the-time-has-come-to-exploit-the-potential-of-algorithms-in-education-c2615468d1eb
"""
Bill Gates hinted at a recent event on the relationship between technology and education that he sees a future for primary education based on using machine learning assistants to teach children basic skills such as reading and writing that can adapt to the level and progress of each child, making classes entertaining and improving the educational experience in the process.

The reality is that generative conversational assistants, hallucinations aside, could already be used to create personalized visual content to teach children to read. In fact, it could also be used for home schooling or to deliver private lessons in many other subjects for practically any child, especially for those from low-income households.

Given that generative algorithm technology is still at an early stage, many people will ask whether it’s a good idea to use it for something as sensitive as children’s education. Will we see children learning to read more or less unsupervised with an algorithmic teacher, and perhaps obtaining better results than its human counterpart? Obviously, the learning process isn’t only about content in a given format; social interaction and skill development are important, but it is worth looking more closely into whether we have reached the point where we can consider robotic teachers for our children. And why not at other levels? Algorithms can adapt educational material and teaching methods to the individual needs of each student. Taking into account his pace of learning the material, learning preferences and level of knowledge. It was high time to change something in education, since for example I have been using an essay writing service for a long time, I found Canadian PapersOwl for this. I'm just bored, this is not what I expected from university to be honest. So I hope something will change dramatically very soon. Wouldn’t participation, for example, key to children understanding concepts, be improved if they could interact with a teacher who answers their questions individually and who keeps detailed record of each student’s activity, affinities and interests, without them having to raise their hand and wait for the teacher to authorize them to take part in the class?

Generative algorithms trained with hundreds of millions of parameters and applied to language have already amply demonstrated their mastery of language and interaction. Those dedicated to image generation, trained with large repositories of labeled images, or those that can generate any type of voice, for example, make it possible to develop high quality educational or reinforcement content, where before we needed a whole set of infrastructure to create this type of material.

Are we already at a level where we can consider incorporating algorithms into something as sensitive as education? Would it improve results? Wouldn’t it make sense to explore it?
"""

They already play an important role and their influence will only grow in the future.

[tggzzz]

LLMs are "statistical parrots". They construct sentences of words that are likely to go together.

LLMs don't have any understanding of the sentences they construct.

When people detect obvious nonsense, it is dismissed as "hallucinations" .


When people fail to spot that, the LLMs have - at best - emitted "word salad".


Now, is that better than a human teacher in a classroom setting? Discuss.


Alternatively, is it worth reading something that it isn't worth somebody writing?


Finally, one thing that scares the thoughtful LLM advocates is "what happens when LLM output is used to train LLMs?" .

[SiliconWizard]

All good points and questions.

And regardless of the technology involved, I think this one is particulary interesting: "is it worth reading something that it isn't worth somebody writing?"

As to LLMs training LLMs, that's, as I said, the equivalent of machine consanguinity in a way. Not something that will end up well IMHO.

[tggzzz]

As to LLMs training LLMs, that's, as I said, the equivalent of machine consanguinity in a way. Not something that will end up well IMHO.

I'm a little cruder: I think of it as eating your own sh1t.