AI the Magic Wand ???

Make your live easier with AI as your companion

We all know in the past decade humanity has taken leaps into the future in terms of the technological advancements that we have achieved and at the top of it is the fans favourite — Artificial Intelligence.

In today’s world where everyone is on a race to leverage AI based solutions to create highly accurate, flexible and optimised solutions, one also needs to consider whether the problem requires an AI solution. Basically implying — using AI to solve to solve “A Problem” is one part of the story but identifying the problem and is it the right fit for AI is another story altogether.

In this article we are going to revisit our world with a different perspective and understand what kind of problems can bring about the best of AI to revolutionise the society.

Let’s first try and understand the capabilities of AI —

How about now we take a look at some of the most compelling questions in the field of AI?

Can it Multi-Task?

As it is evident from the table above that in order to resolve complex or multi-stage problems, a single AI model will clearly be insufficient.

What we need to understand is even though we have made significant developments in the field of AI, still our technology doesn’t have the capacity to mimic a human. The highest levels of AI technology are only capable of performing very primitive and specific tasks.

Hence in order to create a fully sound solution, several models need to be integrated to play the music in harmony.

How important is Data?

Imagine teaching Calculus to a Kindergarten student. For someone who has just learnt the difference between Addition and Subtraction, it’s practically impossible to perform complex tasks such as Integration or Derivation.

In the same manner if you were to expose the neural models to minute volumes of data and expect it to create scalable and high performing results, I think we know what the answer is going to be.

What is the impact of Probabilistic output?

Having probabilistic output has it’s own pros and cons.

For instance when we have predictive modelling in place (Eg: Customer churn, Image recognition, Speech detection, Semantic segmentation etc.), we expect the system to imitate a human’s response and revert back with a confidence score of the likeliness of a result.

On the opposite hand when one has template based engines or rule based outputs, AI can turn out to be an overkill in such circumstance. Some of the main hurdles that one might face would include —
Probabilistic output would never be certain of it’s results making the system unreliable as compared to prior solutions as they were rule based requirements
Or there might be instances of not having a clear winner in terms of the predictions which would lead to extensive post processing

At the end of the day based on what the problem is it would make sense to make your choice.

How does Bias impact the results?

As AI is developed to imitate the functioning of Human Brain, it has a tendency to learn and recognise patterns at every given capacity. Hence of the many varieties of biases such as —
Bias in Dataset
Bias in Testing
Bias in Feedback
Bias in Training
The system tends to tilt its results in preference towards the majority.

This a topic of further discussion but if we were to point out some the trivial concepts involved with the exhibition of such behaviour, those would be Explainable AI, Ethical AI and so on.

Based on the discussion we have had thus far, we have a rough idea of what kind of problems would be a better fit for AI as a solution as compared to the ones that won’t.

Also we were able to answer some of the most crucial questions into understanding the limitations and scope of the technology. Let’s dive a little deeper and understand the full extent of this topic by picking up some examples.

One of the most mind blowing applications of AI has been Wake-Up.

Imagine the huge impact this could create in the world of —
Interactive Learning
Animations
Product Marketing
and so much more

Now such complex problems require multi stage modelling, that encompasses various components of automatic semantic segmentation, 3D rendering, automated animation, pose-detections and lot more.
As we had discussed multi-tasking is not one of the strong suits of AI but like it’s said “Strength in Unity”

Document AI — Another outstanding example of power of AI

If you observe closely you can identify the phenomenal accuracy at which AI could identify the various relevant information from within the invoice and transform it into a structured data.

In cases such as document parsing where over the years the variations in the document forms and templates is enormous, systems leveraging AI solutions are outperforming any existing solutions in the market.

Conclusion

As a realistic concluding statement, the applications of AI are only limited by your imagination, but the approach, scope, and feasibility in today’s world is a thought to ponder upon.

Based on our discussion, AI is clearly magical but even it has certain limitations which are only limited by time and research. By the time you go through this article some of the brightest minds of our planet would have made ground breaking discoveries in this field and the status quo would have changed.

Nevertheless there are time when AI based solution outperform our conventional approaches and sometimes not. Reasons and hurdles are something for us to overcome.

I hope this article was helpful in giving a brief about AI and its capabilities. 😁