What’s Slowing Down the Pace of AI Innovation?

Article by Rei Morikawa | July 11, 2019

With constant media attention on how AI innovation is accelerating at breakneck speed, it’s no wonder that people are so curious as to how everything will be automated 5, 10, or 20 years in the future. But the current speed of AI innovation is slower than what scientists originally expected. This is due to several problems and blockers in AI innovation, such as the shortage of AI engineers and training data.

 

The Current Shortage of AI Engineers

There are only 300,000 AI engineers worldwide, which simply isn’t enough people who truly understand the complex technologies behind AI. Right now, most people who are interested in learning AI are motivated on the user-end about the potential to cash the market. But AI is becoming an increasingly hot topic in all industries, from law to finance and banking.  Hopefully, more aspiring young scientists will join the field soon and make an impact.

 

The Current Shortage of AI Training Data

High-quality algorithms can only be created from high-quality data. But it’s difficult for companies to collect large amounts of clean, unbiased data that is representative of all possible scenarios.

 

Let’s Explore the Current Shortage of AI Training Data in More Depth

AI training data is used to build algorithms and teach them to perform tasks. Researchers use the training data over and over again to fine-tune the algorithm’s predictions and improve its success rate. To train the algorithms effectively, researchers need a large amount of data – but that’s only half of the definition. We also need high-quality data. AI researchers must confirm that the data is clean and organized before using it to train an algorithm. Duplicate, incorrect, or irrelevant data can mess up an algorithm’s ability to recognize patterns, or create biased results. Even small errors, such as incorrectly tagging a word as a noun instead of verb, can create a grave impact. Therefore, AI researchers must be careful and triple-check the AI training data quality before using it.

AI training data usually contains pairs of input information and corresponding labeled answers. In some fields, the input information will also have relevant tags to help the algorithm make accurate predictions. For example, in sentiment analysis, the AI training dataset usually includes input text with output labels of positive, negative, or neutral. In image recognition, the input would be the image, and the label would suggest what is depicted in that image (ex. table, chair, etc.).

 

Why don’t we have enough AI training data?

Companies don’t know how to get started with their machine learning projects.

 The AI hype is everywhere in 2019, and it seems like the world’s biggest tech companies are all embracing it. If you’re managing a company hasn’t implemented AI technology yet, you might feel pressured not to get left behind in the technological revolution, especially if you’ve heard that your competitors have already jumped on the AI bandwagon. But where do you start, and what can AI do for your company? 

It’s impressive to see that so many companies nowadays are interested in adopting new technology to improve their business, but AI isn’t a one size fits all solution. You need to have a plan of how and why your company should implement AI technology. The first step is to define a specific need or problem that you’d like to solve using AI technology. Then, you can think about whether AI is the right solution. If yes, you can move on to researching about what kind of machine learning algorithm to use, such as classic or neural networks.

 

Companies underestimate the amount of data they need, and the time to collect that data.

Companies often decide at the last minute to implement machine learning, right after seeing that their competitor released a machine learning product. This leads to a stressful scramble to collect data, and sometimes it’s a lost cause. For example, you need to collect data for months to train a high-quality fraud detection algorithm. If you rush the process or build the algorithm with only a few weeks of data, then you’ll end up with a poor model that might fail in the real world.

 

Collecting and labeling datasets is a time-consuming task.

Some machine learning algorithms, such as spiking neural networks, require specialized datasets which are often difficult and time-consuming to build. In addition, some tasks such as image labeling are also tedious and require a lot of manual labor. Small and mid-size companies would hesitate to invest in machine learning projects unless they are 100% confident that the implementation will pay off.

 

The few companies invested in data collection often refuse to share their datasets with others.

This data hoarding usually stems from privacy concerns or fear of handing an advantage to their competitors. While it’s true that high-quality data is often what separates good algorithms from great ones, it’s important to also consider the need for an appropriate quantity of data. If a spirit of cooperation gives you access to clean, structured data that is relevant to your existing dataset, then it may be worth sharing what you have. After all, the more high-quality data you have, the better your algorithm will perform on edge cases – and the greater your market advantage will be.

AI innovation requires big data, which requires humans to contribute and collect data. So how can different kinds of companies bring together the human power needed to get enough data? Large companies like Google and Uber have the resources to hire employees whose jobs focus solely on AI training.

But how about small to midsize companies that don’t have an AI research department? For those companies, it’s it’s especially important to store their data in a way that is simple and ready to use. If the companies store data in messy logs, their employees will struggle to work with it and need to perform extra pre-processing steps.

That’s where Lionbridge AI comes in. We offer crowdsourcing tech services to accurately clean and tag your data so that it’s ready to use for your next machine learning investment. We have a decade of experience in managing a crowd of over 500,000 qualified contributors. Our crowdsourcing services include sentiment analysis, product categorization, image annotation, data entry, and more.

The Author
Rei Morikawa

Rei writes content for Lionbridge’s website, blog articles, and social media. Born and raised in Tokyo, but also studied abroad in the US. A huge people person, and passionate about long-distance running, traveling, and discovering new music on Spotify.

Welcome!

Sign up to our newsletter for fresh developments from the world of training data. Lionbridge brings you interviews with industry experts, dataset collections and more.