The Great Promise and Huge Cost of Machine Learning

It seems like every company is incorporating machine learning and artificial intelligence (AI) in its products these days. I see it in tech press releases every day—from smartphone cameras to smart speakers and virtual assistants to DNA ancestry testing kits and business software. AI is a very exciting concept and has huge potential, but it starts to feel like snake oil when it’s incorporated into every startup’s business plan. Do AI and machine learning add value to the world or just boost the bottom line?

Artificial intelligence and machine learning are often used interchangeably, but they are not the same thing. AI is about recreating human intelligence in machines and software programs. AI applications include anything from spam filters to facial recognition technology. Machine learning, which was coined by IBM Labs in 1959, is when computers can analyze data and perform tasks without intervention from humans. Think of a computer that can beat a human at chess on its own, or IBM Watson winning Jeopardy! against two human competitors. It’s also what powers digital assistants such as Siri and the Google Assistant.

There’s a lot of potential for machine learning, especially in industries that already collect a lot of data, such as the government, tech companies like Google and Amazon, and healthcare organizations. To enable machine learning, you need data that the machine can absorb and analyze so it can understand a topic and its real-world context and then learn from that information. In 2016, a Guardian reporter tried to create an autocomplete tool using machine learning. It proved challenging—machine learning requires a lot more computer power than the average reporter has on hand (he ended up using Amazon cloud servers), and it also requires an enormous amount of a data, more than 119 MB of Guardian editorial columns it consumed.

The results, in which the author gave the computer the beginning of a sentence as a prompt, were a series of nonsensical phrases, in part because the computer didn’t have real-world context about what it means to be a member of the EU, for example. The experiment showed that machine learning isn’t something that’s perfected over a few days, and it’s not a magic bullet. Just because you use machine learning doesn’t mean that you’ll get extraordinary results; you may end up with a series of unintelligible sentences. While the goal of machine learning is to reduce the workload for humans, it still requires a lot of work up front.

Let’s go back to IBM Watson, which has extended into many industries, including cancer care. Watson for Oncology, for example, analyzes medical studies and literature to recommend patient treatments at a much faster rate than a human. However, proving whether Watson’s findings are right or wrong requires testing on human patients, which is subject to heavy regulation, not to mention liability. To say that Watson can cure cancer is a vast overstatement, but it can certainly give doctors and researchers complex analysis on the latest medical findings. Again, the humans still have quite a lot of work to do.

Machine learning and artificial intelligence can do amazing things, but currently, it comes with great effort and cost. In the future when it becomes more economical to gather massive amounts of data and less human intervention is required, when machine learning reaches the tipping point and machines can learn and make conclusions, there will be huge opportunities for business sectors, including medical, financial, and marketing.


Molly McLaughlin

Molly McLaughlin is a New York-based technology writer and editor with over a dozen years of experience covering consumer electronics and mobile. Her background includes editor positions at PC Magazine and, and a successful freelance business. Now she writes about Android, business software, and other techy subjects for a variety of publications, including PCMag and Lifewire.