Do we need a new OS for AI?
I stumbled across the following article that contends industry needs a new OS to meet demands of AI and space systems.
https://medium.com/@theo/ai-and-space-show-a-need-for-new-operating-systems-0ea0494de4ad
The author’s requirements for a space OS include:
* Decentralised and Modular
* Customisability
* Real-Time Capabilities
* Community-Driven and Open Source
* Compatibility and Transition
Yawn. These are available today. But the author goes on to list some more novel requirements of an AI OS:
* Resource Management and Scheduling: traditional operating systems are not primarily designed for the complexities of AI workloads. Redesigning the kernel could allow for more efficient management of resources (like CPU, GPU, and memory) for AI processes, optimising performance and energy consumption.
* Security and Privacy: AI introduces new security and privacy challenges. A kernel redesigned with AI in mind could incorporate more advanced security protocols to handle these challenges, especially in processing large volumes of sensitive data.
* Real-Time Processing and Edge Computing: AI applications, particularly those involving machine learning and real-time data processing, can benefit from low-latency and high-speed processing. A kernel-level redesign could optimise these processes, especially for edge computing scenarios.
* Autonomous Operation and Self-Healing: an AI-driven kernel could enable the operating system to perform autonomous optimisation and self-healing tasks, predicting and preventing system failures, and optimising performance without human intervention.
* Hardware Acceleration: modern AI applications often rely on specialised hardware like GPUs and TPUs. A kernel designed with these in mind could provide better support and optimisation for such hardware, enhancing AI application performance. Much like what Graphcore set out to do with its IPU but has fallen foul of product market fit and high capital investment requirements to continue.
* Backward Compatibility and Transition: a significant challenge in redesigning the kernel for AI is maintaining compatibility with existing applications and systems. This transition would require careful planning and gradual implementation.
* Adaptive Behaviour: the system could adapt its behaviour based on the environment and usage patterns. For instance, it could optimise itself for energy efficiency, performance, or security, depending on the context.
What do you think? Does industry need to work on a new AI OS or can we make due with enhancements to what already exists?
Medium
AI and Space Show A Need For New Operating Systems
Are current operating systems, designed in the late 80s and 90s fit for purpose anymore? Space and Artificial Intelligence point the way.
1 Reply
Great article, very thoughtful and in-depth. It makes me feel like the Author is very passionate about AI development in Space. But in general the use of OS in products from consumer, automotive, medical, or even military hardware incl. space. The idea of developing AI OS is inevitable as we progress.
In my opinion, there will be players who are and will try ground-up to build OS for AI. And for sure existing products in the OS market are also trying to release their OS version with advanced AI capabilities. From my observations, I think existing OS will evolve as OS for AI, and these companies will acquire new players in the game (just like what happened to ChatGPT by Microsoft). The cost of running an AI show is huge, and companies know that very well. The new player will develop AI OS but then it will ultimately find its way in our existing OS. This story will continue for AI products like Operating systems.
Let's see how the ecosystem evolves for space/defense. I'm sure there will be a lot of proprietary OS that will show up, just like the invention of FPGAs that happened in the 80's and 90's for defense. Similarly, I see now OS for AI/AI in general will evolve for high-end and niche markets. I'd not be surprised to see some launch of OS for AI from players like Wind River, QNX Embedded OS etc.