As the leader of a company that is investing heavily in theartificial intelligence, Nvidia CEO Jensen Huang came out with a controversial statement. NVidia’s number one, on the occasion of a meeting held in Dubai in the presence of many world leaders, said without mincing words that learning to program is useless. The new generations, again according to Huang, should no longer dedicate themselves to learning the various development languages but they should invest in other valuable skills such as biology, education, manufacturing or agriculture.
As can be seen in the short video clip published on X, the CEO of NVidia noted that for 10-15 years almost every person sitting on stage at a technology forum has insisted that it is “vital” for young people to learn to program computers and, in general, electronic devices . Huang said that, in her opinion, these considerations can be completely eliminated from now on.
“Our task is to create computer technology such that no one has to program it. And indeed, the programming language is the human one“Huang said. “Everyone in the world is a programmer now. This is the miracle of artificial intelligence“.
Artificial intelligence has killed programming, according to Jensen Huang
It is not the first time that Huang supports a thesis like the one advanced in recent hours. Artificial intelligence and, in particular, i generative models they allow any user to generate programming code that works in any language. The only language users need to write code, again according to NVidia’s “helmsman”, is the one they speak best. In short, enough is enough and more natural language to obtain code to transfer to the machines.
In our view, Huang’s statements are perhaps a little too optimistic. A certain one Steve Jobs he remembered at the time how crucial it was to learn programming. The founder of Apple remembered theimportance of programming as a tool to develop critical thinking and creative problem solving. His opening speech was historic Reed College in 2005: “I think everyone should learn to program a computer, because it teaches you to think differently. I think coding should be mandatory in schools. It is the language of logical thinking“. In Europe we have only recently been investing in it and, perhaps, not yet enough at an educational level. And now Huang arrives and tells us that it’s all useless because an artificial intelligence will and will take care of the development.
How much truth is there in Huang’s statements and what considerations do we feel we can make
If we think about GitHub Copilot and the data shared by the Microsoft subsidiary, itself deeply involved in the development and integration of new AI solutions based on OpenAI technology, we know that a good part of the published code on the hosting platform it is generated by developers using artificial intelligence.
Already at the end of 2022 we had given space to the vision of Matt Welshformer professor at Harvard University, co-founder of Fixie.ai, a company dedicated to rethinking traditional software development methods. Welsh also claims that computer programming is dead and that it is destined to evolve into something very different from how we had conceived it until today.
It’s difficult to ignore the sight of some of the most well-known bigwigs in the world of modern IT, as well as great experts in artificial intelligence.
What, in our small way, we feel like observing is that the conclusions reached by Huang seem excessively clear to us; they seem like judgments written in stone, almost unappealable. For many (too many?) years we have read or heard that something would kill the classical programming. And yet, even today, the harsh reality is that companies find themselves in serious difficulty because there are not enough developers with the appropriate skills.
How generative models produce programming code
Developers replaced by artificial intelligence? At present we still see this as unlikely. AI and generative models are not something “magical”: they are based on stochastic approach. As we saw in the article in which we explain how generative models can be at the service of business decisions, these are non-deterministic models that use probability distributions to describe the real world and attempt to correlate words, concepts and…, for example, programming code.
Everything comes from careful, in-depth and meticulous work training: the higher the quality and “specificity” of the data used, the better and more relevant the results produced by the generative model will be. In another article we tried to compare the difference between stochastic and deterministic approaches.
The artificial intelligence so strenuously supported by Huang works using generative models that can rely on computational resources offered, for example, by GPU NVidia. But the quality of the results, limiting ourselves to the programming code, largely depends on the data used in the training phase. And this data is often application sources e routine developed by real developers, then fed “en masse” to the LLM (Large Language Model) which will eventually be used by the generative model to create new code with a stochastic approach.
Poor quality source, programming code destined to not work
In short, if a hypothetical generative model could not count on one initial training phase well executed, the programming code it creates would be poor. Furthermore, without the availability of quality code written by “human developers”, how could we now achieve working code generated by AI?
But then, you tried asking the code generation slightly more complex to a generative model? In the vast majority of cases, that code still requires careful analysis and a series of corrections e di optimizations by a competent developer. Nothing but the uselessness of programming studies!
In our opinion, therefore, generative models are an excellent tool for speed up (also) the activities of software development (we also gave some advice for programming with ChatGPT and underlined how skilled Phind is in terms of programming), especially for fine-tuning routine applicants and get a template basis on which to then build something more complex. To date, they cannot offer high quality results in any application field, much less can they replace each other in full to a real developer.
If Huang had more prudently stated that AI will not kill programming, but will put it in the hands of more peoplewould certainly have been more objective and… down to earth in our opinion.
So yes, in our opinion it always remains fundamental for the new generations learn programming: otherwise, it would not be possible to develop that precious thing critical spirit and that wealth of skills useful for understanding what AI has really achieved. Otherwise, in the absence of these tools, we would almost transform ourselves into automata willing to passively accept what is proposed.
The opening image is taken from this NVidia press release.