Maybe you’ve heard the old African proverb, “it takes a village to raise a child.” The idea is that, raising a child is a group effort with the whole “village” contributing to raising the kid. We might say something similar can be said about AI (or DL or ML). 

In a sense AI is a kind of rudimentary form of intelligence. It certainly needs guidance, tools and adaptation to be productive. (Side note: It’s important to remember that this is just an analogy to show all the different components that make up an AI model. We at MDCS.ai in no way consider AI to have any sort of true consciousness. It may have intelligence, but that is not the same as consciousness.) 

The first thing needed in creating AI is good hardware (like a person needs healthy food, parents and elders).  

For AI these are: 

• GPUs (Graphic Processing Units) / TPUs (Tensor Processing Units) 
• High speed (InfiniBand) networks 
• Blazing fast storage  

GPUs and TPUs are the main powerhouses. They speed up that training and inference processes AI needs to learn. They are the ones behind the scenes doing massive amounts of calculations for the deep learning the AI needs. But they also equip AI with the tools it needs to store knowledge. The networks and storage brings all the data there when the GPU/TPU expects it. At this point, our AI has a good foundation and environment to grow into a healthy functioning model. But that’s just the beginning. Now, we have to interact with it and foster the next stage of AI growth.  

Software 

Software provides AI with Deep Learning Framework. It’s a curriculum for learning and decision-making. AI Middleware is like the bridge to the outside world, connecting AI to broader knowledge. It’s like kids learning through their interactions with society. Data Processing Tools take that raw data and transform it into valuable insights. There are always mistakes and problems (whether you’re raising kids or training AI). Just like you don’t let your kids run all over the place and do whatever they want, we don’t let AI just go however it wants. We have Monitoring and Management tools that keep the AI trained and functioning. Even more than anything with consciousness, AI will continue to drift in its model (often to a point of complete nonsense). All AI models need hyper vigilant monitoring and management. The reasons why is because AI cannot live a bubble (or sandbox). 

In the sandbox or out of it?  

Kids like to play in a sandbox. It’s how they interact and learn from others. For AI it’s getting out of the sandbox (a sandbox is a controlled environment). We can train, experiment, and test our AI model within a closed environment. But at some point, that model AI is going to have to face the big bad real world. Then, and only then do we get to see if our model really holds up. (Lots of the time it doesn’t – to the frustration of many.) We also get to see our skills (or lack of them) in action.  

• Are we monitoring effectively?  
• Is the data that we’re getting at all useful? 

Basically, we’re still training the AI – just in a much larger scope. Once we’ve got a good grasp on the model, now it’s time for it to grow up and get out of the sandbox.  

Our AI won’t just stay doing the same thing it always does. The outside world is not static. It’s changing, developing and growing. Our AI model has to grow and develop right along with it. (Or what’s the point?) Not only that, but if AI is in an environment where the village (or company) is growing, then that means new demands and requirements.  

Often AI isn’t out there in a happy environment. In fact, for the most part, it’s downright hostile. There are viruses, attacks and data breaches happening all the time. That’s where the security measures and protection come in. Lots of people want to interrogate and mine all that knowledge from the AI models (think malware and data breaches). The better the protection, the less likely our AI model is going to get attacked. That’s one part, the other is to make sure it’s protected against viruses and infections.  

I could go on and talk about financial advisors, compliance and ethics, lifecycle management, and more. But I think you’ve got the point. Running an AI model isn’t a one-and-done action.  Just like it takes a village of support and nurture to raise a child; it takes a village of support and effort to make AI models functional and effective. It takes collective, concerted effort to build and support AI. If you’re looking for a company that will be with you through every step of the way – look no further.  

MDCS.AI we’re your business for raising AI.

Michel Cosman – CTO MDCS.AI