The Future of AI or Should Children Learn Computer Science
Friends from a large software development company asked me to answer a few questions about the future of AI. I found the answers interesting and decided to write them down in my blog as well.
Jensen Huang, Nvidia CEO, argued this week that today’s kids don’t need to learn Computer Science anymore. And that, thanks to AI advancement, everyone can be a programmer now.
To which extent do you agree with this statement? Why?
In the modern world, with its immense competition for skilled labor, it is incorrect, in my opinion, to talk about what one doesn’t need to study. It’s far more important to recognize that there is a need to study much more than before — children, to be valuable specialists in the future, need to learn much more than we did, including training their natural intelligence. This applies to fundamental sciences such as mathematics and physics, as well as applied sciences like Computer Science, and social sciences (which are important to keep societies free and capable of creating).
In IT, technology stacks used to change every 5 years, and undoubtedly, with the advent and development of AI, the creation of IT products will change significantly. However, skills in problem-solving, engineering, and the ability to create great products will remain important. Tools change, but more intelligent tools allow for the creation of more complex and valuable solutions, which are very important for humanity. Any knowledge that children acquire will be very useful in the future, including Computer Science, and, all else being equal, those with deeper knowledge will have an advantage in the labor market competition because they can utilize the symbiosis of human and artificial intelligence.
Programming is an essential part of Software Engineering, but there are also Quality Assurance, DevOps, Product Management and many other disciplines involved. Do you think AI can equally replace in all these disciplines? Can AI also be a software tester, devops or SRE engineer, product or project manager, etc?
Professions in IT are constantly changing, and these changes will continue — there will be much more professions, not fewer. Just like 20 years ago, and 10 years ago, in IT, one needed to constantly learn. Nothing changes. Only now, there is even more to learn. This field is not suitable for those who are not ready to learn new things as a student.
I am responding to this survey having set aside a book on modern development in Kubernetes. And I don’t like how it’s happening — it needs to be improved, including the use of AI. In my life, I have been a developer, team lead, IT evangelist, admin, devops, technical support, tester, people manager, a bit of HR, chief IT architect, development department director, development director in a huge corporation, and finally, a lead product manager in a global company. If tomorrow I have to work as a prompt engineer, developer or manager with AI, or someone else — when hungry, a person will master any profession :) It’s so interesting!
How do you see the Software Engineering is going to evolve through adoption of AI to all aspects of software production this year (2024), mid-term (2 years), long-term (5 years)?
I hope that within the next 2 years, local and on-prem AI models will start working well, with strong competition both between services and between local and open source models, as they become ubiquitous. Also, I hope that in the next 2 years, AI will penetrate deeper into specific fields, magically performing familiar tasks where help is needed. Additionally, I really wish that in the next 2 years, especially in the field of development, AI learns to understand the entire codebase as a whole and help by keeping its entire structure “in mind,” which is difficult for humans.
In 5 years, I hope there will be AI understanding of the deep structure of software interfaces and the ability to develop based on the description of what you want, no worse than a human (which will not simplify, but will change how IT products are created — they will be able to do much more, become much more complex, and will require even more powerful devices). But making predictions for 5 years is a thankless task.
Bonus question: do you know any good articles summarizing how co-pilot like models will impact software companies and what framework to apply when analyzing them?
I think that full line completion is a very intermediate step; it’s better with it, but it’s not the target vision of how I’d like AI to assist. I can’t point to specific articles, but a huge number of companies use full line completion, and they really do write code faster (and sometimes create very unusual bugs :) ).
What’s interesting is that both Chat and full line are used by senior+ developers when they venture into new subject areas for them — this help is not just for juniors and middles. But the question of impact for the company is much broader than just the use of convenient tools (convenient tools should just be there, as human labor is always more expensive).
Impact is more about doing what’s needed, and not doing what’s not needed. I’ll share my personal framework (developed by me while studying for an MBA) on how I understand which actions will lead to a greater impact in creating IT products:
