Kathy Gibson at Gartner Symposium, Cape Town — There are many misconceptions around artificial intelligence (AI) — it may not be happening quite the way science-fiction would have it, but it is definitely on a growth trajectory.
“People worry that it will take all our jobs — and we can all go home; that it will soon win Nobel prizes and Pulitzer prizes,” says Alexander Linden, research vice-president at Gartner.
The fact that organisations like the World Economic Forum make predictions about AI taking people’s jobs fuels the narrative, Linden says. But he believes the days when computers will think like humans is not close enough for us to worry about yet.
“We don’t want technology to think like a human: what we want it to do is not get tired, to think fast, to be consistent. But there is no way we will ever be able to figure out exactly what is happening in the brain.”
Ai shouldn’t be confused with analytics, Linden adds.
AI refers to machine capabilities which solve complex tasks that were recently only possible by humans — equally well or better. The analytics applications are technical disciplines that solve business problems through the extraction of knowledge from data.
There are a number of applications where AI is already proving useful , and more will come to fruition soon. Linden says AI is currently very good at making operational decisions that are routine, take less than two seconds and have a simple I/O. Use cases include cross-selling, image recognition, marketing, retention, management, task assignment and failure prediction.
Gartner predicts that, by 2019, deep learning will be a critical driver for best-in-class performance for demand, fraud and failure predictions.
“Deep learning is here to stay and expands machine learning (ML) by allowing intermediate representations of the data,” says Linden. “It ultimately solves complex, data-rich business problems. Deep learning can, for example, give promising results when interpreting medical images in order to diagnose cancer early. It can also help improve the sight of visually impaired people, control self-driving vehicles, or recognise and understand a specific person’s speech.”
Deep learning also inherits all the benefits of ML. Several breakthroughs in cognitive domains demonstrate this. Baidu’s speech-to-text services are outperforming humans in similar tasks; PayPal is using deep learning as a best-in-class approach to block fraudulent payments and has cut its false-alarm rate in half; and Amazon is also applying deep learning for best-in-class product recommendations.
Today, most common use cases of ML through deep learning are in image, text and audio processing — but increasingly also in predicting demand, determining deficiencies around service and product quality, detecting new types of fraud, streaming analytics on data in motion, and providing predictive or even prescriptive maintenance. However, ML and AI initiatives require more than just data and algorithms to be successful. They need a blend of skills, infrastructure and business buy-in.
Most organisations lack the necessary data science skills for simple ML solutions, let alone deep learning. If ML projects cannot be addressed with easy-to-use applications, IT leaders will require ML expertise.
“In this situation, IT leaders will be seeking data scientists,” Linden says. “Data scientists can extract a wide range of knowledge from data, can see an overview of the end-to-end process, and can solve data science problems.”
Gartner predicts that 80% of data scientists will have deep learning in their toolkits by 2018. “If one of your teams possesses a good understanding of data, has business domain expertise and can interpret outputs, it is ready to start ML experiments,” says Linden. “Even if your team lacks experience with algorithms, it can start with packaged applications or APIs.”
Using ML and AI to add value to a business is complicated. “Don’t deliberately meet all ML prerequisites exactly — instead find the right problem to solve,” Linden advises. “It is a good idea to start ML by using the same data you use in your popular reports, such as orders by a region. Then you can apply ML to make forward-looking predictions, for example a forecast for the same orders by a region for the next month. This way it extends on the after-the-fact reports to show business stakeholders the art of the possible with ML.”
Nevertheless, ML has limitations. “An ML system can make the best possible decision if it has enough data to learn from — such as millions of priced items and their availability — but it cannot judge whether any of the resulting decisions are OK ethically,” Linden cautions. A combination of data scientists’ current experience and skills with new ML capabilities will be required for successful ML and AI adoption.
“What’s hard for people is easy for ML, and what’s hard for ML is easy for people.”