Читать книгу Machine Learning For Dummies - John Paul Mueller, John Mueller Paul, Luca Massaron - Страница 63

Any new technology comes with potential pitfalls. The higher the expectations for that technology, the more severe the pitfalls become. Unrealistic expectations cause all sorts of problems with machine learning because people think that what they see in movies is what they’ll get in the real world. It’s essential to remember the basic concepts presented in Chapter 1 — that machine learning algorithms currently can’t feel, think independently, or create anything. Unlike those movie AIs, a machine learning algorithm does precisely what you expect it to do, nothing more. Of course, some of the results are amazing, but keeping expectations in line with what the actual technology can do is important. Otherwise, you’ll promise something that the technology can never deliver, and those adherents whom you were expecting will go looking for the next big thing.

In fact, the uses for machine learning today are quite narrow. As described in the article at https://www.linkedin.com/pulse/machine-learning-its-hard-problems-valuable-toby-coppel, narrow AI, such as the business use of AI to gain insights into huge datasets, relies on well-understood techniques that companies have started to employ within the past decade. The machine can’t infer anything, which limits the use of the machine to the task for which the developer or data scientist designed it. In fact, a good analogy for today’s algorithms is that they’re like a tailored shirt (see the article at https://www.computerworld.com/article/3006525/cloud-computing/why-microsofts-data-chief-thinks-machine-learning-tools-are-like-tailored-shirts.html for more details). You need specialized skills to create an algorithm that is tailored to meet specific needs today, but the future could see algorithms that can tackle nearly any task. Companies that rely on narrow AI need to exercise care in how they develop products or services. A change in product or service offerings might place the data used for the machine learning environment outside the learner algorithm’s domain, reducing the output of the machine learning algorithm to gibberish (or at least making it unreliable).

Using machine learning in an organization also requires that you hire people with the right set of skills and create a team. Machine learning in the corporate environment, where results mean an improvement in the bottom line, is relatively new. Companies face challenges in getting the right team together, developing a reasonable set of goals, and then actually accomplishing those goals. To attract a world-class team, your company has to offer a problem that’s exciting enough to entice the people needed from other organizations. It isn’t an easy task, and you need to think about it as part of defining the goals for creating a machine learning environment.