AI, machine learning, deep learning, algorithms, neural networks – what does it all mean? At this year’s Accountex Caroline Plumb, the founder of Fluidly, wants to cut through the hype.
Fluidly uses the data that resides within accounting packages and applies algorithms to model and predict a business’s cash flow and bank balance. It relies on AI, explained Plumb. So obviously she has high hopes for artificial intelligence.
But there are growing pains, she told AccountingWEB. “We’re basically at peak hype, right at the top of the hype cycle.” But AI isn’t necessarily overhyped, either, Plumb explained, citing Amara’s Law.
The law, coined by the American futurist Roy Amara, states: Humans overestimate the effect of technology in the short term and and underestimate the effect in the long run. “AI will have a massively transformative effect, but it’ll be steadier progress than people think.”
What’s driving that progress are ‘deep learning’ and the explosion of data in modern life. Plumb points at deep learning as a particularly important development. In a very simple sense, deep learning refers to a process that mimics how the human brain learns.
“Deep learning extracts features and interprets more and more with each layer of the neural network,” said Plumb. Again, the neural network here refers to computing that’s patterned on how our brains receive and handle information.
Deep learning and neural networks help create algorithms which can learn and make predictions on data. Instead of following programmed instructions, these algorithms can ‘learn’ in a self-directed way.
Content seriesView full content series
“It’s the emergence of deep learning that’s seen the rapid improvement in algorithm,” Plumb explained. “The computer’s ability to recognise human speech. It’s a pretty steady error rate of about 20% until the emergence of deep learning emerged.”
Data is the last part of this equation. These algorithms run on data. Fluidly, for example, simply couldn’t exist without data. Accounting packages (and newer developments like Open Banking) churn out heaps of data and Fluidly builds an algorithm on top of this vast set of information.
The central challenge of deep learning will be a familiar one to both accountants in business and practice: “How do you take relevant data sets and get insight?” said Plumb. “It’s a matter of taking data sets, standardising them and piping them into one huge data lake where they can be cleaned and used.”
But Plumb’s Accountex session won’t just be about the future. It’s about the past, too. “The talk will be a bit of a history of AI,” she said. “It’s been around since the 1950s. So what are the key themes from AI, from the initial rules based learning, through to machine learning in the 1970s, to the emergence of this new category of deep learning.
"I’ve got an explanation for each, using random Twitter memes and videos. So hopefully it’ll be entertaining, as well as informative.”
Caroline will be speaking about all things AI and deep learning at Accountex 2018. You can see the details here.
About Francois Badenhorst
I'm AccountingWEB's business editor. Feel free to get in touch with comments, tips, scoops or irreverent banter.