Facebook Just Keeps on Trying – Maybe It Should Stay Away from AI
Poor Facebook. Seems it manages to hog the newswires with stories that just make you want to shake your head and ask how it could have made so much money.
One of its fiascos, albeit from a few years back, was to improve its trending topics section by taking away from human editors the responsibility of writing news story descriptions, instead using artificial intelligence (AI) software. This was done as Facebook was afraid that conservative users would abandon the site over claims of bias. Although insiders claim that around 18 contractors reportedly lost their jobs when Facebook decided it would rely solely on its news algorithms, and amongst those who left were engineers who could have corrected any flaws in those algorithms.
The purpose of the AI algorithm was to prevent potential misuse and “to minimize risks where human judgment is involved.” Also according to Facebook, “Topics that are eligible to appear in the product are surfaced by our algorithms, not people.” Does Mr. Zuckerberg know that AI systems and algorithms are trained and written by humans? It seems that the software algorithm began to select posts that were getting the most attention, even if the information in them wasn’t true. Back to the drawing board.
In 2017, Facebook tried again to fix the trending section by including topics covered only by Yahoo! News. Then it changed its mind and included topics by other news publishers. Its reasoning was that covering just one outlet could be a sign that the news was fake. Oh, so once again instead of fixing the algorithm that was spewing out false information, it changed the sources for the news.
Poor Facebook. The problem, which Facebook seems to blame on humans, highlights the potential of embarrassing and sometimes disastrous errors that AI can cause. This incident caused a big brouhaha and, at least from a Facebook perspective, took time to recover loss of brand. Can AI be a benefit? Yes, of course. In this case, it wasn’t ready for prime time and the poor humans kept getting blamed.
Part of the problem with Facebook is that the AI algorithm needed to understand text – meaning unstructured content. Data is machine driven, whereas unstructured content is driven by people, which makes nuances, insights, relationships between disparate content, sentiment, and knowledge capital difficult to extract and difficult for a machine to understand.
Whichever software you use, it needs to be able to differentiate between a news article and a spreadsheet. Interestingly, as I was writing this it occurred to me that we have a client that culls news from literally hundreds of sources. The financial firm invests in over 27 countries and its personnel retrieve highly relevant information from millions of pages of content on a daily basis. Information can come from the public sector, regulatory bodies, and international websites, such as the World Bank.
Using multi-word terms to index material, our insight engine conceptSearch creates an intelligent understanding of text within unstructured or semi-structured information. This enables professional users to easily sort through several hundred documents, some quite lengthy, by providing a synopsis of relevant concepts highlighted within the content.
What does it do? conceptSearch is a statistical, language-independent, scalable technology that can accept queries in natural language, with users typing words, phrases or whole sentences. The system analyzes the query to extract keywords and phrases, to identify the main concepts and retrieve content that is highly relevant. It will even return content that is related but may not contain the search string. For example, a search for ‘triple heart bypass’ will also return content that may contain ‘pulmonary disease.’
Maybe not as intriguing as AI, but hey it works. That’s saying a lot more than Facebook can say.
Join us for our Getting Lost in Semantics – Selecting the Right Search Engine webinar, on Wednesday, September 12. The definition of semantics can vary, depending on who is using it. This webinar takes the mystery and confusion out of search. It explains which technologies work, which don’t, and why. And it demonstrates how to select the right technology, to get the right results, all the time.