Industry executives and experts share their predictions for 2023. Read them in this 15th annual VMblog.com series exclusive.
AI language models will disrupt how humans interact with technology
By Pieter Buteneers, Director of Engineering in AI & ML at Sinch
Science fiction movies have given us some pretty high expectations of artificial intelligence (AI). In these movies, AI has highly advanced language abilities – capable not only of complex thought but also of understanding and manipulating human language just as well as a human. While most contemporary AI doesn’t have quite the level of language proficiency that the movie machines do, it’s getting closer by the day.
In years past, AI has relied heavily on keywords to search a database or maintain a conversation. Being this literal is, of course, limiting: an unfamiliar idiom or slang term could easily confuse a chatbot enough for it to fumble a conversation and, if the AI responsible for searching a database doesn’t receive the right keyword, it could come up with the wrong search results, or have no results at all.
Greater Understanding, Fewer Words
In 2023, we’ll see AI continue to move further away from keywords and toward actual comprehension and understanding. Modern advancements in natural language processing (NLP) are reshaping the ways that humans and AI interact, further enhancing AI to be able to comprehend, write, and even speak languages as well as humans – and in some cases, better. NLP-powered conversational AI can draw on contextual clues to intuitively derive meaning from speech and text even with misspellings and misused words, just as the average human can. Not only are such models unfazed by slang terms and regional idioms, they can even be adjusted so their tone and dialect mimics that of their human counterparts.
Language-agnostic AI, already existent within certain AI and chatbot platforms (like OpenAI’s recent beta release of its ChatGPT model), will understand hundreds of languages – and even interchange them within a single search or conversation – because it’s not learning language like you or I would. Instead, this advanced conversational AI focuses on meaning, and attaches code to words accordingly, so language is more of a finishing touch than the crux of a conversation or search query. Language-agnostic AI equipped with NLP can scan bodies of text for answers to questions or irregularities, and some can even identify when messages are fraudulent, which can help businesses and consumers alike to filter out spam messages. It will power stronger search results – both from external (the Internet) and internal (company database) sources – and less robotic chatbot conversations, enabling companies to lean on automation to reduce resources and strain on staff.
…Or No Words At All
This same concept extends to images as well. With AI that actually understands language, rather than piecing it together one word at a time, we’ll also be able to describe an image that we’re looking for, and have the AI spit out the correct image. Similarly, this AI will also handle image-to-image search: we provide it with an image, and the AI finds another image resembling the first. Beyond its consumer search engine applications, this technology can be applied commercially. For example, retail brands can use it to show customers exactly what they are looking for (even if they can’t remember what it’s called), or suggest alternative products when items are out of stock.
Change On the Horizon
Though far from flawless, cutting-edge AI models are already providing evidence of the disruption still to come in this field. OpenAI’s ChatGPT serves as proof positive of the immense power that a refined language model can wield – particularly when used in place of a search engine. Even in its current beta release form, the range of ChatGPT’s potential applications is staggering. Its ability to draw from the internet to inform and enhance responses, allowing users to pinpoint precise topics and criteria for the information they receive back, already blows conventional search engines out of the water; if its development continues at this trajectory, major keyword-based search engines like Google will soon be rendered obsolete.
Augmenting, Not Replacing Jobs
Still, AI is a machine at its core. Lacking human emotion and common sense, it will still make certain errors (with confidence) that humans never would. While impressive, existing models like ChatGPT lack the ability to fully determine if the answers they give are correct, and though they will deliver incorrect responses with strikingly convincing confidence and fluency, they are not yet reliable enough to be fully depended on. As such, we still need humans working alongside AI to keep it in check, keep silly mistakes at bay and maintain a human touch to business – not to mention the element of true creativity, as AI models can only develop responses by leveraging existing knowledge bases rather than formulating novel ideas.
The popular debate around whether models like ChatGPT will make human professions like copywriting obsolete is misleading: instead of being replaced by AI models, such professionals will increasingly use them as tools to make their work vastly more efficient, and by extension, cheaper. As they are refined in the coming months and years, we will see many more professionals adopting AI as a utility to streamline their work in a variety of ways, from proofreading and suggested copy to even helping them write code.
The Hollywood version of AI – one with a true understanding of human language and inputs – is steadily approaching reality with every passing day. Even in its current state, conversational AI is a vastly powerful, endlessly dynamic tool for modern businesses, and with the speed at which it is evolving, whatever it cannot do at present should always be qualified with “not yet.” Those who manage to keep pace with AI’s progression through 2023 and harness its potential will hold a clear advantage over those who underestimate it.
##
ABOUT THE AUTHOR
Pieter Buteneers is an industrial and ICT-electronics engineer. He started his career in academia, first as a PhD student and later as a postdoc, where he did research on Machine Learning, Deep Learning, Brain Computer Interfaces and Epilepsy. He won the first prize in the biggest Deep Learning competition of 2015 together with a team of machine learners from Ghent University: the National Data Science Bowl hosted on kaggle.com. In the same year he gave a TEDx talk on Brain Computer Interfaces. In 2019 he became the CTO of Chatlayer.ai, a platform to build multilingual chatbots ‘who’ communicate on a human level. In 2020 Chatlayer.ai was acquired by Sinch and now Pieter leads all Machine Learning efforts at Sinch as Director of Engineering in ML & AI.
