Forbes
Hashtags Are Data Treasure
By: Enrique Dans
Facebook has explained at its F8 how it’s using the hashtags from the billions of photographs users upload to Instagram to train its image recognition algorithms and how this has allowed it to improve industry standards: according to the company, its algorithms are now 85.4% accurate.
Google has long been using tagged videos uploaded to YouTube for exactly the same purpose: when explaining to an algorithm what verbs like hugging, fighting or cooking mean, what better way than through media tagged with those words. The system is not perfect, but the tags people use for photographs or videos are rarely wrong, given that they want the content to be easily found, kept within a conversation or tied to an event.
Accurately tagged data is very valuable, because it can be used to train an algorithm and get a machine to be able to understand the meaning of that data. I recently knew of Qure.ai, a company I came into contact with through the French innovation fair Netexplo, which collected radiologists’ and other medical professionals’ images, which were typically filed on the basis of a diagnosis. By feeding them to an algorithm, the company developed a system able to diagnose tumors from medical images, which is still carried out by hand, and thus prone to error. The company says that as the algorithm’s reliability increases, doctors probably stop diagnosing themselves for fear of overlooking certain patterns that AI can recognize. As a result, this skill may well disappear.
Businesses and organizations need to understand the importance of AI. Having the best professionals is no longer a guarantee of success, unless they generate structured data that can be used to train algorithms. Whatever sector you are in, ask yourself how you can make your company’s day-to-day operations generate correctly labeled data that can be supplied to an algorithm for learning? Then ask yourself what your organization could do with highly reliable algorithms.
Obviously, the transition to using AI will require a change in mindset. Understanding the problem makes it easier not only to define reasonable objectives, but to identify potential issues. When it comes to imagining what tasks an algorithm can do, most managers still have few ideas: they see a large number of their processes as intrinsically human and are unable to imagine them being carried out more efficiently by a machine. Obviously, algorithms are not going to take over all tasks, or at least not overnight: machine learning projects require a highly complex and difficult first phase requiring the identification of specific objectives, the collection of data that must then be transformed and supplemented with additional data before building models and making predictions or diagnoses… but despite the difficulties, one thing is clear: data is the new entry barrier, the thing that will differentiate companies.
Steering a company toward generating analyzable data isn’t a Herculean task that involves root and branch change: it simply requires understanding the objectives and attacking the problem in the right way. Getting going on this as soon as possible means that your organization will be ready for the future. Data is the new competitive advantage, as the technology companies have known for a while. When is your organization going to get with the program?
