10 Ekim 2019 Perşembe

The Essential AI Handbook for Leaders

Dear Friends,

I have tried to summarize a very nice book about AI for you. It is a brief summary of what AI is and how we should benefit from it. The author is Marcus Wallenberg.

The book starts with a nice quote "the move from old to the new is the only tradition worth keeping". Yes, this is true. The mankind have always been in the improvement of their life and to do so they are always on a quest. AI is the latest output of their continous quest.

AI which is the short form of Artificial Intelligence, representing the set of computer science techs that allow computer software to learn from experience, adapt to the new inputs and complete the tasks that resemble human intelligence. In short, these algortihms mimic the human behaivour at its best. We are already in touch with AI, but maybe we are not aware. Mapping applications, autocorrect in MS Word, e-mail span filters are all working under AI principles.  

What humas would consider relatively simple tasks like making coffee or pointing the one dog in a photo full of cats are actually extremely difficult for computers to achieve. While tasks considered impressively intelligent for humans like regressions or driving the map between two cities are fairly easy for the computers.

To understand the concept better, we need to address what machine learning, deep learning (DL) and reinforcement learning mean.

Most of the recent triumphs of AI were made possible thanks to a group of AI techniques collectively referred to DL, or more scientifically DL Neural Networks. With DL you teach a neural network by exposing it to data and information about that data. For example, if you want to identify the cats, you show the network many cat pictures and tell it, these are cats. If the connections inside neural network increase, the accuracy of AI increases also.  

DL models can be constructed using supervised (SL) or unsupervised (USL) learning. SL uses labelled training data to predict or aim towards specific outcomes of new data while USL can reason about data without any need for predefined labeled training data. Reinforcement learning uses feedback algorithms to reward and punish the model to achieve the best possible outcomes.

In SL, if you want the model to learn elephant, you teach AI model to identify every picture of an elephant on the internet. In SL you teach the model with thousands of pictures and tell the model which ones have the elephant inside. An USL model would not be great for determining elephant or no elephant. But, it is very helpful in identifying the patterns in images to form seperate clusters of pictures. Then, you will be able to extract elephants out of the group of animals. In the reinforcement learning you start without any data and teach the model on trial and error base.  

What are the interesting use cases for AI in our life? AI could grade the tests for an entire class for the teacher in seconds. It can identify or predict the tumors faster and more accurate than any of the radiologist or cancer doctors. Once trained, AI weather forecasts can be accurately made in seconds which will be cheaper, quicker and more reliable. AI can predict the number of clients to be served at dinner in a restaurant for a specific night. AI can translate or transcribe an audio or voice message into a written text in seconds. AI can help the parents what to be cooked for dinner depending on the stuff in the fridge.

Despite its big uses described above, there are plenty of limitations of AI. We need to make a categorization here to explain these limitations better. The first category is the Artificial Narrow Intelligence. It refers to use of AI in specific cases. Beating the opponents in chess, diagnosing the illness, understanding the language and converting it to text are the specific fields where AI is very powerful. Our smart phones are also full of AI pieces that are designed for very specific tasks. There is another category called as Artificial General Intelligence where AI is not so much powerful. It refers to AI that is completely able to reason and understand like humans. This is not so much powerful because AI has some requirements to be fulfilled. These are;
- The problem should not be vague. You can not be succesful if you assign a "make my company profitable" job to AI.
- Lack of data. The fuel of AI is data. If you are out of correct data, the AI model will not succeed.
- Unorganized data. The AI models are more succesful in organized or categorized data.

At the end of the day it is not technology that creates success, its people. Its the leaders that take the right decisions based on the the most accurate data, insights and their experiences. It is the ones who do faster than the competition that will succeed.  

Better to read the book, thank you for sharing your time to read this article.

Hiç yorum yok:

Yorum Gönder