Companies are losing money despite investing heavily in Artificial Intelligence. Here is why
I was browsing through a lot of articles on Medium regarding careers in Artificial Intelligence and Data Science. I happened to come across an article that explained certain statistics about companies that are losing the 21st-century tech world despite investing heavily in Artificial Intelligence and Data-driven strategies. Those articles made me feel amused. How can the field that was named as the ‘Sexiest field of the 21st-century’ be the reason for increase in fiscal loss in companies? I took the question to many HRs and CEOs of some successful AI and Data Science startups and discussed with them about this.
According to me and the other people in my team at PhospheneAI, hiring practices in Artificial Intelligence and Data Science in the past decade is one of the most misguided processes in most companies in the world. We understood that certain issues with hiring processes caused a huge misdirection for the upper management in companies making them take the wrong business decisions. This article focuses on what Modern Artificial Intelligence is, who should focus on a career in Artificial Intelligence and what is wrong with the hiring practices in companies that are invested in AI.
Most companies that claim to have used AI in their products are basically talking about a very small subset of algorithms in Artificial Intelligence known as Pattern Recognition. Pattern Recognition is a class of Artificial Intelligence algorithms that focus on replication of the human cognitive process of learning. This field is largely inspired by the connectionist theory, which is the foundation of most of the research in Neuroscience in the past few decades.
Pattern Recognition is also known by a fancy term called Machine Learning. Machine Learning algorithms are a class of algorithms in Artificial Intelligence that allow a computer to learn from data. I’ll not get into the technical details; I’ll reserve that for another article. But take it from me with a leap of faith that it is possible to understand and interpret how you think and process information by using certain algorithms. Let’s look at some common examples that we all see every day. Every time I open Youtube on my browser, I get some video recommendations displayed on the website. My Youtube recommendations list currently looks like this:
Now, you can try going to Youtube and see that your recommendation list looks entirely different from mine. The question is, how does Youtube decide what my recommendation list should look like? The motive of any content-based business is to ensure that its users spend more time on their platform. That is why Youtube recommended a different list for me and another list of videos for you. If I look randomly at those videos, it is very clear to me that those videos represent the kind of videos that I would generally watch.
Let me dissect it for you. I follow the major elections around the world and because it is election season in the USA, I recently watched a lot of videos related to the US presidential elections. To be specific, I am a huge fan of three presidential candidates, namely, Andrew Yang, Mayor Pete Buttigieg, and Senator Bernie Sanders. That’s probably why I have recommendations of townhall meetings and debate clips from those people. I also watch daily talk shows and that’s why I have Trevor Noah’s and Jimmy Kimmel’s video clips as recommendations. I previously watched lectures regarding Artificial Intelligence on Youtube and that explains why I have a lecture regarding Perceptrons. Finally, I also regularly watch music videos of boy bands and that explains why Youtube recommended a clip of Backstreet Boys’ hit song ‘Show me the meaning’.
You get the general picture, right? Youtube predicted the videos that I am most likely to click on and it was done using my past search history and the kind of videos that people with a similar search history have watched. This is known as Pattern recognition. Everything that we do is based on some pattern. Our eating habits, google search history, writing style, learning style, typing habits, etc all form a specific pattern. Using your data from your usage history, it is possible to predict what you are likely going to do next is the core of most AI-based applications. With the abundance of data in this digital age, it is much easier to understand what your patterns are.
This isn’t just limited to user personalization. This idea of pattern recognition forms the base of many business-related tasks that have greatly influenced our work culture in the past decade. I’ll mention some of the works that we’ve been working at Phosphene AI.
At Phopshene AI, we work on tasks that help creators improve their productivity by focusing on what they do best, which is content creation while automating the other curation related processes. By working with some cognitive psychologists, we’ve been able to make some huge strides in Image Memorability prediction, Image Gaze prediction, and Image aesthetics analysis. These tasks have historically been very subjective tasks that cannot be completed using classical algorithms. But by collecting loads of data from users, we’ve been able to do some great work.
I am a computer science graduate, but I have taken many certificate programs on economics and cognitive neuroscience. Without that understanding of how humans perceive beauty and how they react to certain emotions, we wouldn’t have been able to work on the tasks that we’ve been successful at. Me and my team’s understanding of the visual cortex helped us understand how humans perceive and collect the right data in the right way to help a computer understand how we humans perceive beauty. When working on those tasks, around 70% of the time involved collecting the right data and segregating them and only 30% of the time was spent on programming the right code to learn patterns in those data. This sets the premise of what we want to convey.
“Programming is not the core task in AI or Data Science”
The biggest mistake that companies have done is hiring computer science engineers to analyze and infer important decisions from data. How can a person who is proficient only in programming work on data-driven strategies that involve critical business decisions or Biology related data or Chemistry related data? Let’s say I’m hiring a computer science graduate to analyze economic data from my state of TamilNadu. How will she/he know what to infer or what to predict using that data or how to predict using that data? A common person would believe that analyzing GDP can help understand an economy, but an economist would know that GDP alone cannot be used to analyze an economy because there are other factors like life expectancy, wages, wealth distribution, etc.
Most companies fail to produce results using data because they hire ‘Computer Engineers’ to analyze data. Especially in India, candidates for the role of Data Scientist and Artificial Intelligence Engineer are expected to have more programming knowledge and computer science specific skills and are not expected to have domain knowledge in other fields. The main cause of this is the lack of proper understanding of Artificial Intelligence and Data Science by the CTOs and Product Leads.
We at PhospheneAI have done some good work only because of the diversity of our team. All of us are computer graduates, but each of us has some domain knowledge in cognitive neuroscience, psychology, economics, and micro-biology. Understanding cognitive psychology allowed our team to process social media information to understand why people make certain decisions in social media and take different decisions in real life. When we were analyzing HR data from various companies and predicting their churn rate, our knowledge in Microeconomics helped us understand why certain employees took certain decisions.
Artificial Intelligence and Data Science are interdisciplinary fields that don’t require programmers. At PhospheneAI we believe that this field of Artificial Intelligence and Data Science will grow better if it is run by people experienced in every other field other than computer science. Software engineering principles don’t apply here in Data Science and Artificial Intelligence. Anyone who cares about these fields knows this. AI and Data Science require people from the field of economics, biology, mathematics, physics, sociology, psychology, and the likes. Hiring software engineers to perform data science and Artificial Intelligence tasks will lead to the same disaster that many multinational companies are facing despite investing huge amounts of money in this field.
