Thursday, March 28, 2024

Microsoft Moves Toward AI That Creates Offspring

Datamation content and product recommendations are editorially independent. We may make money when you click on links to our partners. Learn More.

Microsoft Build is often a fascinating conference because if you look at several things they announce and then project those efforts into the future, you can get a better sense of when future revolutionary offerings might arrive.

A case in point is that at Build this year, Sam Altman, the OpenAI CEO, spoke about a massive language model that can do natural language tasks and was able to do initially unexpected things. One of those things was it could generate code from a comment or description. They also spoke of their new Supercomputer that should be ranked 5th in the world.

The presentation on this was right before they introduced Project Bonsai, which is a project for rapid, autonomous robots and right after their talk on Microsoft’s Azure Cognitive Services and Deep Speed, a massive resource library and training model that helps the user build models 20x bigger and 20x faster than before. 

I think they have many of the components needed for eventual AI procreation but, initially, these tools should massively increase the productivity of coders, particularly those working on some of the more advanced products. 

Self-Writing Code – To Procreating AI

Like a lot of you, I spent considerable time in Computer Science in college, but there were so many annoying things about coding that I gave that part up pretty fast.  I particularly hated the repetitive stuff you had to do.

It turns out, with the proper training, AIs can get good at repetitive tasks in that they can both learn from what you do repetitively and learn from what others do as well. And this is what Microsoft spoke about, by allowing their AI to scan the massive code repository on GitHub, the AI was able to learn both how to write certain repetitive lines of code and connect that skill to the comments, even across a wide variety of coders, on that code. 

Right now, and what they demonstrated, you’d have to write some kind of comment or description so that the system could determine what you wanted it to write, but then it would write what it learned. But, I can easily imagine, given AIs also do this all the time, that as the AI got smarter, it would be able to use something similar to autocomplete to increase speed even more.

And this wouldn’t just be looking at what you are doing and intuiting what was next needed and writing it but also correcting coding errors it sees while you are coding, much like a spell checker does not. And, of course, as this technology evolved, you should be able to increase just define what you want an app to do, and the system will take that definition and use it to autonomously code the app you asked for, leaving you to fine-tune the result. 

And once you get to the point where you can turn description into an app automatically by providing this capability to an AI, the AI could then write its code from the reports it generates. And that is how you get to the AI’s potential future capability to procreate. 

This idea of increasingly automated coding across a blended ecosystem of cloud resources and endpoints, which increasingly will include robots, is a potentially massive game-changer in terms of coding speed and quality. And it is also a fascinating application for AI because it will have a direct impact on how fast you could create a new AI and stand it up.  

Wrapping Up: Self-Writing Code

Microsoft’s Build was unusually forward-looking this year, but what got my attention was the ability for their new Azure Cognitive Services capabilities to significantly reduce the aggravation and time it takes to create an app.  While initially, this effort will be focused on improving the speed at which people can code, it will evolve. 

This evolution, coupled to Microsoft’s ever more capable AI, will increasingly allow developers to create apps from descriptions and eventually make it so AIs can do their coding.  And when an AI can completely code, it will be able to develop different variant AIs for independent, focused tasks. 

And if some of these AIs are in Robotic bodies, well, I think this means they’ll need to learn how to write Mother’s Day Cards because the Microsoft Azure Cognitive Services will likely be their mother. 

The next decade promises to be very interesting. 

 

Subscribe to Data Insider

Learn the latest news and best practices about data science, big data analytics, artificial intelligence, data security, and more.

Similar articles

Get the Free Newsletter!

Subscribe to Data Insider for top news, trends & analysis

Latest Articles