AI is the buzzword taking the world by storm, promising to revolutionize every facet of our lives. Indeed, we’ve already discussed the mind-blowing challenges AI is currently tackling, from diagnosing malaria to finding missing children. But what about the future of AI?
Whilst companies may be promising that generalized AI is just around the corner, we still have a way to go before sentient AI becomes a reality. However, we are going to see a boom in AI adoption in 2019- here are 5 trends we can expect to see in the next 11 months.
AI & Big Data
Those of you who keep a close eye on tech buzzwords may think big data has died- but AI is almost certainly going to help resurrect it. One of the key challenges facing many data science projects is the lack of quality, organized, structured data. This is where AI could have a massive impact. If 80% of the world’s data truly is unstructured, AI capabilities that allow companies to structure and model data, and engineer pipelines, could have a profound impact on business- and make data scientists' lives a whole lot easier!
Chatbots & Virtual Assistants
Chatbot hype is everywhere, and it’s true that as machine learning algorithms including NLP become more widely adopted, we’re likely to interact with chatbots more and more. What we’re most likely to see in 2019 is not only more chatbots, but a shift towards “virtual assistants”- essentially, bots with greater language fluency and distinguishable personalities. Autodesk have been publicly singing the praises of Ava, their virtual assistant, and how she’s been much more successful than their depersonalised chatbot Otto. Expect to get to know virtual agents on a personal level in 2019.
Our lives are becoming increasingly digitized- and with that, we become more and more vulnerable to cyber attacks. As the number of cyber attack threats increases, cybersecurity solutions have to process more and more data, and find potential threats at increased rates. You guessed it: this is where AI comes in. AI algorithms are able to sift through potential threats at speeds that human cybersecurity experts simple can’t achieve. Whilst human expertise will remain vital in this field, being able to churn through data and prioritise the biggest threats is a job for the robots.
Interoperability and ONXX
One of the main obstacles facing widespread neural network applications is the lack of interoperability. There’s a wide range of machine learning frameworks out there- from PyTorch to Tensorflow, from Keras to Caffe2- but once a practitioner is locked into their ecosystem, transferring neural net models to other toolchains can be incredibly tricky.
Luckily, 2019 is the year that this is set to change. The Open Neural Network Exchange (ONXX) developed by Microsoft and Facebook is designed to make porting neural nets between different frameworks much easier and faster. ONXX supports a huge range of frameworks, converters and runtimes, and we should expect to see widespread adoption throughout the year.
As well as an interoperability problem, current AI solutions suffer from an interpretability problem. Which is to say: most AI applications don’t offer customers and businesses the opportunity to look “under the hood” and understand why the algorithms have come to certain outcomes, or made recommendations. From the business side of things, explainable AI means that companies can get a critical view into machine learning models; on the customer side, explainable AI increases trust, and assuages fears about algorithmic bias. Perhaps surprisingly, explainable AI is one of the most contested trends on this list– but in a time where AI has such a tangible impact on our lives, we will hopefully see greater transparency in the coming months- and indeed, years.