Can Machines Think?
The Secret to AI Success: Asking the Right Questions and Defining the Problem
Throughout my life, I have come to appreciate the immense power of asking great questions. At first glance, one might assume that asking a great question is a simple task. However, upon further reflection, it becomes clear that formulating a truly great question is far from easy. A well-crafted and impactful question leads to a new world of learning, providing valuable insights into a wide range of subjects. I believe that the art of asking great questions is one that requires dedication and refinement over time.
In 1950, Alan Turing posed a great question that continues to resonate today: can machines think? Turing's work on this topic remains relevant, and his Turing Test (originally known as the imitation game) is well known within the field of IT. Alan Turing not only posed the question of whether machines could think, he provided a detailed explanation in his paper on how to approach answering this question, which he called the imitation game. I believe this is what made his paper so famous. Turing's great question still holds great potential for uncovering valuable insights into the realm of Artificial Intelligence.
While the term Artificial Intelligence (AI) was first coined in 1955, its roots stretch back even further through numerous publications and technological advances. Over the years, AI has continued to evolve, but it is only in recent times that we have witnessed an explosion in its development. The driving force behind this growth lies in the ever-increasing computing power and capacity to store and process vast amounts of data.
In the realm of Artificial Intelligence, we distinguish two primary branches: Artificial Narrow Intelligence (ANI) and Artificial General Intelligence (AGI). While ANI has experienced rapid growth in recent years, the same cannot be said for AGI.
ANI is a type of learning algorithm specifically designed to perform a single task. This technology finds use in a wide range of applications, including image recognition, chatbots, conversational assistants, speech recognition, self-driving cars, machine translation, and more. The goal of ANI is to address well-defined problems or questions.
There is a lot of hype surrounding Artificial Intelligence (AI) these days, which has unfortunately led to instances of its misuse and overuse. It's important to keep in mind that AI is not a cure-all solution for every problem. Before attempting to implement your next project using AI, take the time to write down the great question that defines the problem you're trying to solve. Be as specific as possible, outlining the inputs and outputs involved. Asking great questions is key. Keep in mind that while some questions are best solved by using AI, the vast majority aren't.
Initially the great question you're trying to answer may not be feasible to implement using AI. Or it may require huge volumes of data or processing power that could make it not feasible. However, by refining and targeting the question, it may be possible to leverage AI to solve the problem with some success. The definition of the question/problem not only helps you confirm if you should use AI, but it also helps you identify the data you need to train the model (using a learning algorithm).
One example of a question that may not be feasible to answer using AI initially, but that can potentially be addressed with some success after refinement is:
Q: Can we identify all the objects from any given image? Given input image A, generate B, a list of all the possible objects found in A.
Although AI has shown success in image recognition, identifying a potentially large number of different objects is a highly complex problem that requires a massive amount of labeled images, compute power, and time. However, by refining the question to identify a limited set of objects, it may be feasible to leverage AI. This assumes you have enough labeled data to train the model effectively.
As a general guideline, AI tends to work well when there is a large amount of data available and the problem to solve is relatively simple (something that can be solved in less than a second of mental thought). Conversely, AI often doesn’t perform well with complex problems that have limited amounts of data available to train the model.
I just introduced new concepts such as data, learning algorithms, and models. If you're not familiar with these concepts, don't worry. I plan to explore them in greater detail in future posts so that you can have a better understanding.
I hope we come across some great questions today! See you in my next posts!