Has Machine Learning become Alchemy? This debate has many sides. It hinges on the arguments we use, and the assumptions that we bring with us. While AI has advanced a great deal, the ethical and legal considerations are not always clear. For example, alchemists knew that opening their trade secrets was dangerous.
However, when gold became plentiful, manufacturers worried about going bankrupt. Today, data from users is used to train algorithms and make money. Regulatory and legal responses to privacy concerns are also lacking.
The machine-learning community is beginning to realize this problem. With smart speakers and smartphones, the application space has expanded beyond the purely technical. Researchers at the beginning of their careers face special challenges. They are not yet familiar with the latest advances in deep learning and artificial intelligence (AI). But the broader community is increasingly saying that AI needs a second wave of innovation to move from a scientifically rigorous model to one that incorporates contextual intelligence.
While machine learning algorithms improve over time, the field itself is slipping. Some researchers compare the process to alchemy. One prominent researcher, Ali Rahimi of Google, described machine learning algorithms as “alchemy” because they were trained purely through trial and error. Moreover, they lack rigorous criteria for selecting AI architecture. In this book, Rahimi and his co-authors document cases of the “alchemy” problem and propose prescriptions to make AI rigorous.
Classic regression models are only intelligible when there are few parameters to study. Likewise, ANOVA, which is another form of alchemy, is often misused. It has little or no basis in reality, and most people who are trained to use it don’t fully understand it. The same goes for unsupervised learning. This method of modeling changes is never guaranteed to create reality-grounded models. Unlike supervised learning, unsupervised learning goes through different transformations before choosing the optimal algorithm.
The history of alchemy has been important for mankind. The goal of alchemy was to transmute base metals like lead into noble metals, such as gold. However, it was not as easy as it seemed. It took thousands of years for the technology to develop, and it still relies heavily on trial and error. In fact, most practitioners of machine learning are performing alchemy. That’s why they are so successful.
There are no algorithmic breakthroughs behind the recent dramatic improvements in machine learning, but massive computation power. These computational resources were not available when the current ML algorithms were invented. In fact, companies such as Google and Facebook are talking about using thousands of hours of specialised GPUs to solve problems. This approach is possible today due to Moore’s Law. However, the process is far from finished. In the meantime, the technology continues to advance.
The future of machine learning depends on the ability to combine abstract reasoning and perceiving. When these three processes are combined, the result is a higher level of intelligence than what we’ve seen before. Ultimately, it will lead to the Singularity. And we’re just beginning to understand how these three steps work together. It’s time to get to work on it! cunoaște More About Machine Learning
The first step in this process is to identify the target customers. Using machine learning and targeted marketing, Digital Alchemy clients can identify key variables and patterns in huge data sets. Through these systems, they can also understand which customer segments are prone to high churn. Once identified, they can then tailor offers, content, and messages to these groups and offer them deals that will appeal to them. It’s a win-win situation for both parties.
Deep learning is the next step. It’s an evolution of classical machine learning, which was developed by Fisher in 1936. But unlike classical machine learning, deep learning does not need bird feathers to be useful. However, it does have a great theoretical underpinning. And this theory is what allows this method to scale to a large scale. So, let’s take a look at some of the major factors influencing the development of machine learning.