7 Cases Where You’d Rather Avoid Machine Learning
Hey, machine learning can’t be the solution to everything, and even if it were possible, we need not apply machine learning to every problem! In this post today, let’s check out all those cases where we would rather not use machine learning.
Where a simple code or programming would do the task: This has to be obvious. Either as part of a big problem you’re trying to solve or be it an independent problem by itself, if you can solve it using a simple program, then why even consider machine learning? Machine learning is complicated, needs a lot of work, requires skilled people to do the job, and costs a lot. If there exists a simple fix, go for the simple fix.
When you can use a simple spreadsheet or dataframe to give the input values and the labels: Let us say that for a product, there are only two possible color choices: black and white, and you want to take the input from the user and run a process. It would surely be wiser to enter create a small table on your own for these two colours with all possible typos and variations you can think of and convert it all into lower case and then process the input, labelling all other input values as “other” or “invalid”. Here, you don’t need to upload a detailed database of colors and train the machine to identify all the input values and find the correct color match, right? That would be plain unnecessary.
When you don’t have sufficient data to make the machine understand the differences between the different classes: Let us say I have 4 photos of dogs and 3 photos of cats. Would you suggest that I use these 7 photos to create a “machine learning application” that would differentiate between cats and dogs? Would that even work? In this example, it’s quite evident that the data is insufficient. But in some cases, I don’t understand why people can’t see that there is not enough data, but they still want results. Without data, there is no machine learning.
As I had just mentioned, the example given above was simple enough, but usually, it’s never this straightforward. Machine learning problems are quite complicated. Let me try and explain. While the symptoms of different illnesses can itself make up a huge database with several overlaps when it comes to the symptoms of different conditions and ailments, how would a common, typical patient or caregiver express that? That’s where human agents and machines are different. It’s easier to train a human being to pick up these cues and that way, humans are more reliable and faster at understanding what the caller’s concern could be, but to train a machine to gain this level of knowledge, that’s going to require a LOT of data.
When an alternative would be much cheaper to maintain: Be it a simple program or having a human agent handle it, if all that you expect are two calls a week and you already have a team of employees taking care of most of the tasks, surely one of those could handle these two calls a week, too, no? Again, I intend to make my point clear, which is why I am using examples of this kind that would be easier to understand and relate. Now, would I want to use conversational AI services for which I have to pay some monthly fee to handle such rare cases? And that to, only for them to ask what the working hours of the shop are? Certainly, no.
When it is too dangerous to let a machine make serious judgments: No, I’m not talking about artificial intelligence taking over humanity! 😉 Imagine that you’re letting a machine diagnose whether people have a life-threatening, easily communicable disease like Covid-19. For as long as the machine does a good job, it would be great. But no machine learning algorithm is 100% accurate. Never. Even Google’s search algorithm can fail really bad every now and then. What if a machine labels someone as “not a threat” and the person boards an airplane? That would be putting the lives of all the passengers and crew members at risk, no?
When your idea itself is too unrealistic to begin with: Let’s say someone gives you this idea that you can use machine learning algorithms to test the personality and attitude of interview candidates and then recruit only those who would not quit easily so that you don’t have to deal with employees engaging in “quiet quitting” later. The idea, like so many ideas, sounds great on paper. But how will you train a machine learning model to detect those qualities in people? You don’t have the data to begin with. You can collect data, but to collect data, you need to know what you are looking for. You must then pay for it. And you won’t know how reliable that would be … Ask yourself every time you come across an “interesting AI application” idea if it is worth it in the first place!
You’re looking for a quick solution: While it is true that training a machine is probably the easiest part of machine learning, all it takes is a few lines of code or a few button clicks, there are more than a few demanding and exhausting processes involved. The data collection, the data pre-processing, picking the right parameters, understanding the roles of these parameters, avoiding conflicts and overlaps, keeping the error minimal, and all these sub-processes would have to go through several iterations and reviews until you would arrive at the most convincing results. This would easily take at least a few months of team work, especially if it’s a new or unique idea and you have to figure things out from scratch.
It’s understandable that we’d be drawn towards new trends that are exciting. There was a time when blogging became a trending hobby when popular platforms like WordPress would let anyone create their own blogs and share their thoughts, and just about anyone interested in writing with access to Internet wanted to give it a try. Now, there are smartwatches that are a happening trend in India, and so many youngsters purchase these and show keen interest in fitness. Similarly, if you’re running a business, anyone in your position would be expected to keep yourself updated with all the trends in the industry, be aware of customers’ and clients’ expectations, be aware of rival businesses and such. As part of your growth plan, you might consider using machine learning and automation in your projects. While it’s certainly a good thing to be curious and open-minded to new developments, it’s also equally important to be able to critically assess the pros and cons of major moves you’d make.
Hope this post has given you a clear idea of when you’d rather not need machine learning, either because there’s too much to pay or you’d have little to gain at the end of the day.
“Urgency and despair don’t get along well.”
― N.K. Jemisin
If you are confident that you’ve given it enough thought and you do want to take the path of machine learning models, then make sure you understand what you are signing up for. Wish you good luck! 😊
