A featured contribution from Leadership Perspectives: a curated forum reserved for leaders nominated by our subscribers and vetted by the Construction Tech Review Advisory Board.
Head of Tech Innovation and Product at Gruppo Mutui online spa
There is a storm coming in
I really hope that most of you will recognize the title of this article!
A cult movie that marked my generation (70 something) with the concept of evil AI and all the ideas that come with. It has been extracted from the last dialogue in Terminator, one of the movies that depicted the growth of machines as independent from humans.
I also remember the last sentence of one article that I wrote for this very journal: “Remember that a robot and AI are code— someone needs to write them… for now!”
I never thought that after just one year I would need to correct myself. I am of course speaking about the edge product called ChatGPT. A machine able to understand Natural Language and Process answers based on common and specific knowledge. It has been trained with literally tons of data taken from web, books, scientific articles and so on. The result, as you might have experienced on your own, is a machine able to discuss at the same level as humans do. It is able to generate text in response to questions regarding almost everything even tuning the language to be inclusive or pretending to be someone else.
And yes, it can read, understand, comment and correct code as it was a senior developer.
I have some colleagues that are already using code generation tools to help with unit tests and other routine operations, and it’s proven to be working smoothly in terms of time saving.Here we are at a different level of skill— ability to suggest and correct brand new code simple asking for it in natural language!
My opinion as former developer is still to be quite sceptic since there might be external factors to be considered. If you try ChatGPT with very specific questions about your company, it is highly possible that the answer is not entirely correct or even very wrong. This is due to the fact that there is no high-volume deep documentation about it, hence not enough training data. The same is for coding! I’m sure there are a lot of cases in which the same issue you’re facing has already been tackled and addressed by someone else in the world but having it ready for your system, your production environment and your security team still needs work. If you write code, you know what I mean: there is always a high level of specific code written for the hosting system and someone having an overview of the general picture must be on the human side.
In the long period they might not even need to write code but simply find the right language to explain some high-level specification to a machine and to test the result
So, are we all becoming ‘integrator’ of code generated by a machine? Well, my idea is that Citizen Developers might feel an impact and initially they will be very happy to have such help. In the long period they might not even need to write code but simply find the right language to explain some high-level specification to a machine and to test the result. For sure small and medium companies will need to study how to turn this new tool into a useful one in order not to fit the coding gap that they have but to give more bandwidth to existing structures.
On senior side, there are still major issues to be solved. I do not have the punctual information on training data and timings of ML models underneath ChatGPT but must be big numbers. This means that you can’t have real time data that would be needed to address security issues of just known bugs. This is the same issue you face in essay writing using AI. Training data stops at a certain date so you risk writing partial information that could also be wrong. Not thinking about kind of adversarial attacks to instruct machines to write unsecure code.
Technology cannot be stopped, so what I do and suggest is to invest time to find applicable corporate scenarios in which to test those solutions. We all need to be prepared to control the raise of machines that still cannot reach human performance… for now!
