“If an elderly but distinguished scientist says that something is possible, he is almost certainly right; but if he says that it is impossible, he is very probably wrong.”
A couple days ago during tea time with my students of the joint program ADA University and George Washington University (Computer Science and Data Analytics) one of my students answering the question “what is your top concern about the future?” said “I worry that AI will take my job as a programmer”. I responded saying that probably it’ll not take place earlier than several decades.
Today I decided to check capabilities of the new arrival to AI landscape ChatGPT. Along with ability to be engaged in realistic conversations, chatbot can also generate computer programming codes. Today I’ve submitted the following query to the ChatGPT chatbot launched by OpenAI in November 2022 and built on top of the new version of well known large language model GPT 3.5.
My query was “generate code for web crawler with multithreading support that implements polymorphism in java”
As a response the model generated 50 lines of code in just a couple of seconds. I can’t say the code is ideal and ready to run, but it’s ideal as a template if you consider the time it took to generate, also it can be improved with further interactions with the model. The output is really amazing: it implemented the method for crawling, it implemented concurrency and it wrote the code in a truly object-oriented manner. After changing the last word of the query to “Python”, it generated the same task in Python language.
As a response to my another query “generate data model for HR system using ORM in java”, ChatGPT came with data model consisting of three classes: “Employee”, “Department” and “Holiday” supporting it with quite detailed description of relationship types between table (one-to-many, many-to-one) and primary keys.
I continued my interaction with “Having just three classes for full HR system is too simple. Can you offer data model for real HR system for the large business entity?”. ChatGPT responded with a very kind reply “Certainly! Here is a more complete data model for an HR system for a large business entity, using Object-Relational Mapping (ORM) in Java:” and according to the model’s point of view, it’s enough for HR system of large business entity to extend data model with several fields in each class and add one new class “PerformanceReview”.
The query “Please, generate data model for HR system using ORM in Go” results with the same data model, but in Go programming language. That means that ChatGPT is familiar with the syntax of all popular programming languages. This also proves another important truth: one can doubt in the ability of model to generate code, assuming that it just brings codes from code repositories (like Github, SourceForge, Codebase, Bitbucket, StackOverflow, etc.), but the last query disproves that assumption. ChatGPT really generates code, learning from billions of code snippets and projects it has access to.
Of course, it’s far away from being mature enough to generate a real software product, or data model for the real system, but certainly, it’s in the right direction and will improve over time.
How it will affect education?
What does it mean for educators and students? Some students may think that they will be able to utilize ChatGPT and other AI systems to delegate them their programming assignments. But, in reality it is not as easy as it can sound. To make the model to produce exactly what is required, one will need to write very accurate and precise requirements that include enough details for the model to solve the task. As you know, good requirements and specifications are up to 70-80% of the success. If students are able to write those requirements, why not spend the remaining 20% of the effort to write the code by himself, instead of risking being caught for plagiarism. At the same time, students can continue using the AI system for the satisfaction of ego and how well they perform as a human in comparison with AI.
It also will bring some challenges to professors as well. Probably, those who want to prevent students from feeding their assignments directly to AI-powered systems, will need to spend more time making assignment descriptions be more abstract and at the same time clear enough for students to generate requirements and write resulting code. Certainly, in light of AI advancements and derived opportunities, the education community will be forced to revise existing policies.
Conclusion
As I can’t refer to myself as distinguished or neither elderly, so even after my experience with ChatGPT, I’m still of the same opinion: programmers are not at the risk of losing their jobs, at least for another several decades. It’ll take time before the AI will be able to compete with human programmers (not coders). AI can produce the solution for the small coding task and quality will improve over the time.
But when it comes to programming of large and complex enterprise systems with sophisticated business models/logic, AI-powered code generators would not be helpful. This kind software development project is about many components including: project management, business intelligence, business analysis, quality assurance, platform/cloud engineering (DevOps), database management (ORM), data warehousing, backend/frontend development, UI/UX, security, scalability, testing, deployment. All these components should be in harmony with each other and strongly integrated.
Does it mean that AI will never be able to deal with that complexity? Of course, it doesn’t. But it’ll take a long before AI acquires enough knowledge about business and about the overall World. AI will also be prone to mistakes, as humans do, but those mistakes can’t be easily found and explained by humans. Possibly, AI will find a way to do it differently bypassing some steps and making the lifecycle less sophisticated. We will see what way AI chooses, probably, in a couple of decades, or sooner.
Abzetdin Adamov's IT Blog 