If you’re in the computer science or computer engineering industry, news of this nature probably doesn’t affect you much. After all, the jobs being replaced are often entry-level, or those that require highly repetitive manual tasks. Computer science is a realm that would feasibly be in even higher demand with the advent of machine learning and AI. However, there are some important changes already developing thanks to the higher demand and higher excitement for these technologies.
How AI Startups and Companies Push for the “Sexy” Projects
For starters, AI and machine learning are hot topics in the realm of IT. Companies are clamoring for more machine learning solutions, even if they don’t fully understand them, pushing demand for new machine learning tools, scripts, and software to unprecedented new heights. For example, it’s estimated that by next year, roughly 20 percent of companies will have employees dedicated to monitoring and guiding neural networks, and more than 10 percent of new IT hires in customer service will be responsible for writing scripts for chatbots, one application of AI.
There are going to be several side effects of this push. First, novel projects that have the capacity to be even more impressive than conventional machine learning could be pushed aside, as our most talented engineers and computer scientists chase the positions that are offering the most money or the widest range of opportunities. This means the next truly original breakthroughs in computer science could be pushed back to make sure we explore this current trend to its fullest. Next, the demand for lower-level IT experts is going to shrink, making it harder to find entry-level positions, and virtually impossible to sustain a career indefinitely unless you have some kind of niche skillset.
AI’s Diversity Problems
We’re also facing increasing consequences from the lack of diversity in AI and machine learning fields. AI experts are overwhelmingly white and male, and the byproducts of an industry with an overwhelming majority are typically problematic, unaware of how other populations are affected by their work. For example, in the past, facial recognition technologies have had trouble recognizing people with darker skin tones, since these systems were developed by people who were mostly white.
As interest in AI continues to grow, it’s going to show up in more places. AI will be giving you search results, choosing which news articles to show you, gatekeeping access to your important files and personal information, and possibly keeping you safe. And because the field will likely grow faster than the diversity problem can be solved, it’s going to introduce many new headaches—both for consumers and for computer engineers and scientists trying to stay ahead of those problems.
Why It’s Hard to See Where an Algorithm Goes Wrong
Machine learning and AI also introduce a new layer of complexity that makes certain problems harder to solve—and it’s not just because they’re more complex than previous forms of coding. For the most part, machine learning algorithms aren’t necessarily a set of instructions; instead, they’re designed as a vague learning process for a machine to follow. The machine collects data, usually millions of examples of whatever it’s studying, and gradually learns about the concept, whether it’s recognizing faces in images or learning how to play Super Mario Bros.
The problem with this is that while we can see the evidence of the machine getting closer and closer to achieving its goal (or even surpassing human-level skills), the developers can’t “see inside” the algorithm to determine which pieces of information led to a specific conclusion. In other words, it’s incredibly hard to diagnose, specifically, where an algorithm goes right or wrong.
As AI and machine learning become more common, this is going to become an increasingly complex problem for computer scientists and engineers to address, requiring more insight, higher level skills, and possibly, entire new approaches to developing AI in the first place.
The Automation of Automation
We also need to take seriously the possibility of one day automating the process of creating new machine learning or, in an Inception-like twist, new automation software. Such a multi-layered approach to computer science would open a new branch of study, and require the creation of entirely new ways to look at problems in the computer industry.
AI and machine learning aren’t merely passing fads, though there may be even more novel computing breakthroughs on the near horizon. As a computer scientist or engineer, it’s your responsibility to predict and adapt to the massive changes in store for the industry in the coming years, as machine learning and AI become even more in demand.