One of the hottest topics in the technology industry right now, at least judging by the frequency with which it’s covered in news articles, podcasts, and video shows, is artificial intelligence or AI development. Some people say AI will revolutionize the world while others think it could be the end of mankind as we know it. If you listen to enough podcasts about AI development, however, you’ll notice that at least one central question seems to come up over and over again: Is the pace of AI development slowing down? To answer this question, let’s take a look at some interesting data that came out recently…
| Data Preparation
Most, if not all, machine learning and AI algorithms require data. It’s a key step in making sure your algorithm is interpreting your training data correctly, and creating any sort of generalized intelligence is impossible without access to significant amounts of diverse data. For most organizations, accessing suitable training datasets can be difficult or even impossible (especially for consumer applications), which is why many companies are turning to APIs that collect and aggregate public data. API aggregators like Google Maps API and Facebook API provide an easy way to build on top of large-scale collections of user-generated content in a way that doesn’t sacrifice privacy but also ensures that your AI product benefits from access to more relevant and unique information than would otherwise be possible.
| Feature Engineering
A common pitfall in machine learning is to jump straight into building a model when, in fact, feature engineering is often more important. With so many data scientists and tools available today it’s tempting to think we can let someone else figure out how to use features and get on with modeling. However, investing time and effort into understanding your data and what features will maximize performance is often more important than simply choosing the right algorithm. If you can do that, your models will generalize better, which leads to higher prediction accuracy. This also makes it easier for others to interpret your models by helping them understand why they perform as they do—particularly useful when communicating results with non-technical users.
| Machine Learning
The pace of AI development has slowed down because many of us are taking a step back to reevaluate what exactly makes something intelligent. Artificial Intelligence (AI) is an umbrella term that encompasses any method in which human knowledge and activities can be automated. Machine learning is one branch of AI, in which computers use training sets to learn how to make decisions on their own without specific directions. Until recently, developers have been making leaps and bounds when it comes to machine learning—the progress being so rapid that some have speculated we’re approaching a singularity period where machines will surpass humans’ abilities. This view has come under fire from other scientists, who see a slowdown in machine learning development.
| FE Automation
A recent report from Forrester Research indicates that automated business processes are becoming more popular, with 43% of enterprises planning to implement at least one next year. That’s good news for enterprise efficiency, but it could mean some changes for employees, who may have to adapt to new roles or lose their jobs entirely. Because of AI development in business process automation, more companies are able to reduce costs and increase efficiency by implementing streamlined processes. However, that also means that sometimes employees have to adapt or even lose their jobs because of AI development in business process automation. A recent report from Forrester Research indicates that automated business processes are becoming more popular, with 43% of enterprises planning to implement at least one next year.
| Shortage of AI training data
The pace of AI development has slowed down because engineers and data scientists haven’t been able to get their hands on enough high-quality training data. Google Brain research scientist Ruslan Salakhutdinov puts it bluntly: I would argue that almost all fundamental breakthroughs in machine learning over the past several decades could be traced back to better and more comprehensive training sets. Salakhutdinov says we need to find new ways to train our algorithms, and while deep learning may lead us there, we can’t become complacent. We still need other techniques if we want our AI systems to advance.
Why don’t we have enough AI training data?
To train computers to understand natural language, we need a large supply of text. To train robots to carry out tasks, we need videos showing them how. In other words, we need loads of data—lots and lots of data—in order to make artificial intelligence systems smarter. That’s why experts have been concerned about a recent report from China that suggests that China is slowing down its development of artificial intelligence—because it doesn’t have enough high-quality training data on which to feed its algorithms.
Companies underestimate the amount of data they need and the time it takes to collect it
If a product is going to use AI to learn from its customers, then it must have access to that data. Companies often underestimate how much time and effort it takes to collect that data, especially if they need customers to manually provide it. In many cases, companies end up delaying their rollouts because they don’t have enough data. That’s why Amazon recently launched Amazon Lex for Business—which simplifies common tasks by giving developers access to Amazon Web Services features through an easy-to-use conversational interface so developers can build a chatbot without having any deep knowledge of artificial intelligence or machine learning.
| Conclusion
While we believe that machine learning and artificial intelligence will continue to revolutionize nearly every industry, it is becoming clear that progress in these areas is not happening as quickly as it did two years ago. Rather than advancing exponentially, we believe that steady improvement of existing technologies is more likely. We’ll begin to see new developments emerge in fields such as medical diagnostics and service robotics while breakthroughs like truly autonomous cars and deep neural networks become more difficult to achieve given the current limitations of computer hardware and data sets available for training. If you are looking for an AI developer there are many AI development companies that can help in developing AI-based applications.
