{"id":9894,"date":"2025-03-12T03:00:00","date_gmt":"2025-03-12T07:00:00","guid":{"rendered":"https:\/\/www.both.org\/?p=9894"},"modified":"2025-03-06T11:31:16","modified_gmt":"2025-03-06T16:31:16","slug":"dont-skip-the-learning-process","status":"publish","type":"post","link":"https:\/\/www.both.org\/?p=9894","title":{"rendered":"Don\u2019t skip the learning process"},"content":{"rendered":"
\r\n
\r\n \r\n <\/i>\r\n <\/a>\r\n <\/span>\r\n<\/div><\/div>\n

Generative AI has the potential to streamline our work and make us more efficient. For example, we can use AI to translate old code<\/a> into more modern programming languages, helping with maintenance. AI is also usually good at summarizing things\u2014although we\u2019re finding instances where AI cannot summarize well<\/a>\u2014so an AI \u201cassistant\u201d might summarize meeting notes for you, or write an abstract for a long document, which you can review and edit.<\/p>\n\n\n\n

These advances are made possible because we already know how to do the work, but the AI takes care of the \u201ceveryday\u201d tasks. The human remains \u201cin the middle\u201d of the process, reviewing the work produced by the AI. That might mean integrating AI-generated source code into a larger project, or reviewing a summary before attaching it to a report. Using AI as a tool to do a thing<\/em> still relies on knowing how to do the thing<\/em> yourself.<\/p>\n\n\n\n

You need to learn it first<\/h2>\n\n\n\n

My concern is when I see people using AI to skip the learning process<\/em>. Aaron interviewed me last year about AI in education<\/a>. In that interview, I compared using AI as a tool to using a calculator to do math.<\/p>\n\n\n\n

My analogy runs like this: When you first learned about math in elementary school, you learned how to add and subtract. Later, you learned simple multiplication and division, then progressed to multiplying two-digit numbers, and long division. Throughout the process, you learned about math using pencil and paper.<\/p>\n\n\n\n

And now if someone asks you to multiply 7 \u00d7 4, you know the answer is 28, because you learned how to multiply<\/em>. You can take that a step further: you know that 14 \u00d7 4 is just 2 \u00d7 7 \u00d7 4<\/em>, or 2 \u00d7 28. That\u2019s 56. In the same way, you know that 14 \u00d7 8 is 112, because it\u2019s multiplying by 2 again. You can work out the answer because you learned how math works<\/em>.<\/p>\n\n\n\n

And now that you understand math, you can estimate: 8.8 \u00d7 3.14 is basically \u201calmost 9\u201d \u00d7 \u201cjust over 3\u201d so the answer should be \u201ca little over 27.\u201d But if you already understand how math works, it\u2019s okay to use a calculator to get the more precise result: 27.632.<\/p>\n\n\n\n

It\u2019s okay to use a tool if you already know how it works. If you understand the process, the tool makes you more efficient.<\/em> On the other hand, if you skipped the learning process in grade school and only used a calculator to do even the most basic arithmetic, you would find yourself at a serious disadvantage later on.<\/p>\n\n\n\n

AI can’t replace learning<\/h2>\n\n\n\n

An article in the ACM<\/a> last year demonstrated the danger of using AI to skip the learning process. In the article, a computer science professor challenged his students to use a programming language they didn\u2019t know (Fortran) to solve a problem. One group was allowed to use AI to write code for them, another group was allowed to use Llama to suggest code. A third group could only google topics about Fortran; they had to figure out the rest on their own.<\/p>\n\n\n\n

As the article highlights, the group using AI finished first, and the group that had to work things out took the longest:<\/p>\n\n\n\n

\n

One group was allowed to use ChatGPT to solve the problem, the second group was told to use Meta\u2019s Code Llama large language model (LLM), and the third group could only use Google. The group that used ChatGPT, predictably, solved the problem quickest, while it took the second group longer to solve it. It took the group using Google even longer, because they had to break the task down into components.<\/p>\n<\/blockquote>\n\n\n\n

A week later, the students were tested on it: the students who let AI do the job for them couldn\u2019t do it, but the students who had to figure it out by googling things passed.<\/p>\n\n\n\n

\n

Then, the students were tested on how they solved the problem from memory, and the tables turned. The ChatGPT group \u201cremembered nothing, and they all failed.\u201d \u2026 Meanwhile, half of the Code Llama group passed the test. The group that used Google? Every student passed.<\/p>\n<\/blockquote>\n\n\n\n

Learn it then do it<\/h2>\n\n\n\n

That\u2019s how I approach teaching my students: I teach undergraduate and graduate level technical writing. I want my students to learn how to do it first<\/em>, then it\u2019s okay to use a tool.<\/p>\n\n\n\n

For example, when my students learn how to use rhetoric to write a compelling proposal, I don\u2019t let them use AI for that. If my students skip the learning process, they will be underprepared for a career. One day, they will be asked to write a client proposal. If they skipped the learning process by using AI to do their work in the classroom, they won\u2019t be able to do it for their work. They may use an AI to write the proposal draft, because that\u2019s the trend anyway, but they won\u2019t recognize if the proposal is good or bad<\/em> or how to improve it.<\/p>\n\n\n\n

But if they learned how to write a compelling proposal<\/em>, then they will be able to evaluate an AI-generated proposal and make it better<\/em> by applying their own understanding of rhetoric<\/em> and how a client might respond to a proposal<\/em>.<\/p>\n","protected":false},"excerpt":{"rendered":"

Using AI as a tool to do a thing still relies on knowing how to do the thing yourself.<\/p>\n","protected":false},"author":33,"featured_media":3579,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_lmt_disableupdate":"","_lmt_disable":"","footnotes":""},"categories":[307,28],"tags":[123],"class_list":["post-9894","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai","category-opinion","tag-ai"],"modified_by":"Jim Hall","_links":{"self":[{"href":"https:\/\/www.both.org\/index.php?rest_route=\/wp\/v2\/posts\/9894","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.both.org\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.both.org\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.both.org\/index.php?rest_route=\/wp\/v2\/users\/33"}],"replies":[{"embeddable":true,"href":"https:\/\/www.both.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=9894"}],"version-history":[{"count":4,"href":"https:\/\/www.both.org\/index.php?rest_route=\/wp\/v2\/posts\/9894\/revisions"}],"predecessor-version":[{"id":9898,"href":"https:\/\/www.both.org\/index.php?rest_route=\/wp\/v2\/posts\/9894\/revisions\/9898"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.both.org\/index.php?rest_route=\/wp\/v2\/media\/3579"}],"wp:attachment":[{"href":"https:\/\/www.both.org\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=9894"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.both.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=9894"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.both.org\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=9894"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}