As we know, the field of AI is becoming the most rapidly advancing one. The technology is slowly encapsulating more and more diverse fields. It makes it necessary to learn about technology and start adapting to it. But what is AI? How to develop a mindset to understand if a company requires it at all? This AI talk by Mr. Manish Singhal touches on the questions mentioned above. But before we dive into the real questions, let us understand what AI means.
What is AI?
Technologies that mimic human perception.
Mr. Singhal made the definition quite clear with this quote. This one line explains it all. Artificial intelligence is just trying to match human perception. There are two kinds of systems. Mr. Singhal talks about the same to touch upon the topic of why AI should be used.
Rule-based vs Pattern-based
Classical programming takes data and rules as input and produces an appropriate answer. A typical AI/ML paradigm takes data and answers as input and produces the rules it learned.
There are many cases in systems, where the answer is either 0 or 1, but in real life, there are a lot of cases in the grey area, i.e., somewhere in between. So, it is better to use a pattern-based system rather than rule-based. Because there is no way that all the rules can be provided to the system to get all the possible outcomes.
As explained by Mr. Singhal in the talk, there are many uses and advantages of AI, but why should one use it? One answer could be to get a pattern-based system to cover the grey area cases more effectively. But is that the only reason to adapt to an AI mindset? Actually not. The talk further discusses this issue with many companies. But to understand when to adopt an AI mindset, Mr. Singhal introduces some of the inside-lingo in the AI industry.
Concept of AI-First and AI Second
There are two types of companies dealing with AI technology. AI-first and AI second. AI-first companies mean that their primary product is using AI or data-driven products. AI second means that the data is either generated as a by-product of their primary product.
For example, if a company is dealing with voice technology or chatbots, they are considered an AI-first company because the primary product is based on AI, here which is natural language processing. But if a company is dealing with efficient routings for logistics, the primary product is not AI dependent. Rather they can use AI to analyse their routing data and make it somewhat better. This type of company is considered as an AI second company.
Depending upon this, a company needs to understand if at all, the adaptation is necessary or not. If yes, up to what extent. To answer this particular question, it was necessary to ask some fundamental questions about data.
Here are some of the questions posed by Mr. Singhal, which are necessary to answer while adapting to an AI mindset.
– Are the datasets easily available, or you need to generate your own?
– What is the cost of data?
– Is data a by-product of the business?
– Is the model data-hungry or data optimal?
As mentioned above, it is necessary to know what will be the data like, and from which sources it can be acquired. But one of the most important ones is the last question. Is the model data-hungry or data optimal? If it is optimal, then the need for data can be reduced, or else it is necessary to rethink the cost of data.
Deep Learning vs Feature Model Engineering
The final discussion was to understand when to use deep learning and feature engineering. Deep learning deals with giving a model a lot of data and letting it find out the features necessary to produce the desired output (detection or classification). In feature model engineering, along with data, the model is also provided with some manually engineered features to train upon. It makes the task quite simpler as well as optimises the use of data.
Mr. Singhal explained it with a simple example of identifying circles and squares. In the deep learning approach, many pictures of circles and squares will be provided to the model, and it will learn what to look for in both the objects. While in feature model engineering, we can give the model a hint to look at the angles, and based on that, let it decide if the object is square or a circle.
Mr. Singhal introduced the basis of adapting an AI mindset and building a company around it. The questions posed can help a person in the journey of making an effective AI product for the current market. And as Mr. Singhal said, the only thing to remember is that the use of AI should be justified, it should not be adopted just because others are doing it. Make sure that the technology is really required.