Skip to main content

The limits of AI: why not every problem can be solved with machine learning

· 4 min read

There’s no denying that AI is a powerful tool. It has revolutionized the way we interact with technology, giving birth to intelligent virtual assistants, self-driving cars, and language processing tools. Some view it as a solution for all the world’s ailments, while others call it a curse that will bring down society as we know it.

In reality, however, it’s not that clear-cut.

The fact is, there are still many places where humans excel over AI. I believe that AI is incredibly useful in very specific circumstances. Our deepfake technology uses AI algorithms to face swap but doesn’t rely only on AI to complete the operation. And we only use AI when a traditional algorithm can’t solve the problem.

In the case of face swapping, we don’t train the AI to analyze an entire face at once. Instead, we give it a smaller, more detailed job: finding landmarks. AI extracts and analyzes only the relevant features necessary for the swap. From there, we’ve broken the steps down into the smallest chunks possible, such as finding the face, aligning the face, and putting the face in a specific orientation. After that, other algorithms take over.

The overarching theme here is: You should give your AI the least amount of work possible. The more specific, detailed, and rote the task is, the more likely AI will be able to successfully execute.

Let’s look at some reasons why we feel this way.

Things AI can’t do

AI lacks the innate creativity and innovative thinking that humans excel at. Sure, you can ask AI to provide suggestions or generate content based on existing data. But it will struggle if you ask it to come up with a truly original idea or suggest a unique solution that’s “outside of the box.”

And then there are ethical and moral judgments. AI algorithms make decisions based on patterns and data, without intrinsic values or a sense of right or wrong. If you want human emotions factored into the decision, well, you’ll need to ask a person — not your AI assistant.

Even in the case of self-driving cars, AI training data must be stripped down to the details. Developers don’t just give the AI video input and tell it to drive. No, it’s a meticulous, step-by-step process where the AI is trained to detect video inputs in small chunks. Then the data is fed into the model which learns to detect cars, then lanes, then signs, and so on.

This hands-free driving system works in part by using AI for specific tasks, but not every part of the drive. A driver can’t use it to drive through a crowded residential neighborhood or take a nap in the back seat during the trip. They can only be hands-free on specific parts of the drive.

That said, on a more positive note…

Things AI does extremely well

AI can handle tasks that put people in harm’s way, like working around machinery or in conditions with extreme temperatures or toxic fumes. AI also doesn’t get bored or need breaks. That makes it ideal for repetitive tasks or working long hours. It’s also able to do repetitive jobs without tiring or getting bored, so it does well with tasks like data entry.

It’s also incredibly valuable for solving challenging problems and recognizing patterns that may be too complex for human brains. The medical community is already putting this skill to use. AI can analyze thousands of medical images and identify patterns that may indicate a disease. And of course, AI excels at data analysis and visual recognition, both of which are necessary for deepfake and face swapping applications.

AI is a niche solution

When it comes to tackling small, specific problems, AI shines. For larger problems, it should be combined with other algorithms, tools, or other AIs to build a bigger structure. AI is a lot like the human body: Behind the scenes, specialized organs are focusing on their individual tasks. But put it all together, and our bodies can do almost anything as a whole.

So yes, for the moment, AI has some limitations. For now, AI tasks are best broken up so that you’re not wasting resources or efforts.

Don’t try to make AI be all things to all people. Let it do what it does best, and free people up to do the work they really enjoy.