AI developing an understanding of semantics is the next step in its evolution, suggests machine learning expert, but when are we likely to reach this milestone?
AI is close to reaching an understanding of semantics as the next step in its evolution
AI is a misnomer, or so it is often suggested. The first letter — artificial — is about right. As for the second word — well, there is nothing intelligent about it. Take semantics as an example, there is nothing remotely intelligent, or otherwise, about artificial technology understanding the meaning in sentences, paragraphs and books for the simple reason, it is unremittingly bad at it.
But could this be about to change? Recently, we sat down with Hadayat Seddiqi, director of machine learning at legal tech company InCloudCounsel. He reckons that AI is close to reaching an understanding of semantics as the next step in its evolution. We asked: “when are we likely to reach this milestone?”
SEO: if you are a writer, who takes pride in your writing, in applying your lexicon to express an idea without repetition, then you probably hate SEO. If you like to make a complex argument, that can only really be expressed over a sentence, a chapter, an article or even a book, then SEO, because of the way it makes you dumb arguments down into two or three or four words, such as AI and understanding semantics, is the enemy. (See what we did, there?)
"NLP to break down human communication: How AI platforms are using natural language processing"
Whenever we go online it’s a safe bet that we’re only a click away from being on a site using AI. Sites and applications from industry giants such as Google and Facebook are constantly using AI in order to improve their user experience and provide more efficient services. Many programs such as chatbots are made possible through the use of natural language processing or NLP. RPA applies it too
Wouldn’t it be great if search, using AI to understand semantics (there we go again) could create search results based on a much more sophisticated — intelligent, if you will — understanding, removing the need for tedious repetition!
The natural language processing market is in fact expected to reach $22.3billion by 2025– which illustrates how far the technology has come, particularly in how we communicate and do business.
Currently, these technologies are being used for a variety of purposes within organisations, including reputation monitoring of brands to determine sentiment analysis, providing insights on ad placements via keyword matching or sense disambiguation, and can even be used in regulatory compliance to make sure products don’t become a liability.
"Six surprising ways businesses are impacted by RPA, OCR and NLP"
Robotic process automation, optical character recognition, and natural language processing, or RPA, OCR and NLP, are some examples of newer technologies that positively affect businesses.
In that case, AI’s true power to revolutionise industries and determine key business insights, lies in its ability to read text and understand the semantics (or relationship between words) to help organisations further mitigate risk and uncover liabilities. In turn this creates massive value in natural language processing.
So, in the story of AI enabling understanding of semantics, what is the next step in its evolution and when are we likely to reach this milestone?
Seddiqi says: “Reaching semantic understanding in AI will require several key milestones. It’s helpful to think in terms of the iterative progress leading us to semantic understanding in AI and what each milestone means.”
“A previous milestone,” says Seddiqi is Word Vectors: “Let’s frame this milestone in terms of a common use case that most people are familiar with: search functionality on a computer. Everyone has used Ctrl+F/Command+F to find something on their system, which finds what you’re looking for by matching keywords exactly. Moreover, using a search engine such as Google to find information will include a ‘spellcheck’ component to address potential misspelling.
“But what about words that look very different but mean similar things? Around 2013, the AI community found an efficient way to model this called ‘word vectors’. You can do fun word algebra stuff, like King + Woman will give you Queen. More practically, you can now expand your search to include semantically related words.”
"Five AI advancements that are making intelligent automation more intelligent, by Sarah Burnett"
Sarah Burnett, from Everest Group, one of the top analysts in RPA, explains what intelligent automation is and why it can be a massive benefit to enterprises.
Upcoming milestone in AI understanding semantics
“Word vectors were a game-changer, but they still required you to express your idea in terms of one or a few keywords. But what if your idea requires an entire sentence to express? That is the next milestone, which we’re arriving at thanks to a big push in research last year.
“The idea is that you can take a sentence, encode it into a sentence (or thought) vector and then find similar sentence vectors. If encoded well, your search function can find very different looking sentences that express the same idea.
“It’s not unreasonable to say this technology will mature within the next few years, based on current research progress.
Future milestones: AI understanding beyond sentences
“There is a clear pattern of hierarchy emerging in the progression of this technology. We’re getting close to AI understanding ideas at a sentence level using similar techniques from the word level and scaling them up. This opens up exciting applications for AI understanding ideas requiring paragraphs, entire documents, or even entire books.
“AI’s recent leap to understanding sentences from words has not been trivial as the ability to do so has largely been constrained by dataset size and computational power. Our ability to create models to handle these bigger problems has so far been shown to hinge on these two resource constraints.
“As these costs decline from advancements in AI hardware, we will see ourselves getting closer to models that understand larger collections of text. This is somewhat proven by Open AI’s GPT-2 model, which shows that using the same sentence encoding model designs with a large amount of data, produces models that already understand high-level concepts across many sentences. For example, GPT-2 understands enough to write entire news articles with astonishing coherence.