The buzz around artificial intelligence (AI) and machine learning may have subsided, but the arguments are still taking place within accountancy. John Stokdyk and Francois Badenhorst assess how these technologies are currently affecting the profession.
AI pushed itself to the top of accountancy’s agenda during the bot-fest of 2016-17, when every specialist software developer was touting its accounts assistant bot or the machine learning capabilities of its automatic coding algorithms.
Richard and Daniel Susskind crystallised the key debates around AI in their 2016 book, The Future of the Professions, which predicted that within just a few years, AI would sweep aside many of the attitudes and jobs of status quo accountants. The percentage estimates vary, but it is not uncommon to hear how accountancy is one of the most vulnerable professions, with 80-90% of its workload open to automation.
The promise of AI
Before giving into the “bots are coming” apocalypse, it’s worth looking more closely at the portfolio of concepts and processes surrounding AI to develop a better understanding of the changes that are taking place.
At one end of the spectrum are the natural language processing (NLP) tools that drive communications bots – interpreting the varied syntax of human speech and text and formulating an appropriate response. This was a fundamental element of mathematician Alan Turing’s famous test of machine intelligence: in his view a computer could be said to possess artificial intelligence if could mimic human responses under specific conditions.
Unit4’s Wanda, Sage's Pegg, QB Assistant and thousands of bots on public websites are proving that some chat interfaces are now able to pass the Turing test, but in the words of one banking executive at last year’s FinTech Connect event, “Chatbots are really bad at service.”
The other end of the AI spectrum lies deep within the world’s burgeoning data mines, where self-learning algorithms are used to determine usual patterns within the data and then spot uncharacteristic outliers. As Mindbridge’s John Colthart explained recently, these tools can help auditors discover material misstatements within accounting ledger data.
The ultimate example of “unsupervised learning” AI is Google’s AlphaGo system, which was able to learn and build on the rules of the ancient Chinese board game Go by playing itself around 5m times in three days. Not long after that it was able to defeat the world’s leading human player.
One of the building blocks of AI analysis is machine learning, which starts by matching common patterns and then learning what to do with similar items by copying what its human guides do. These systems are being applied to transaction autocoding in accounting systems like Xero and QuickBooks and numerous optical character recognition (OCR) and transaction capture systems.
Gerd Leonhard from the Futures Agency addressed the promise and pitfalls of AI at last November’s Xerocon. For him, the key question that Turning set had moved on from “Can machines behave like us?” to “Can machines be intelligent like we are?”
Machines might be able to look at data and process it faster, but that “doesn’t make them intelligent. It makes them faster at processing data,” he said.
“AI is a computer system that turns data and information into knowledge,” he continued, citing the cybersecurity intrusion detection, technical support, financial trading and even TripAdvisor as examples of intelligent assistance. But in the case of the latter, “What knowledge does it have about food? Has it ever eaten?” he asked.
Intelligent assistance is a powerful tool, but it can’t decide why you don’t want to have a particular expense on your books. Turning to the need for human intelligence, he explained that a modern airplane can fly itself – probably better than a human – but human pilots are needed to handle the tasks that a computer can’t – like negotiating with air traffic controllers or pacifying recalcitrant passengers. Leonhart’s summarised his argument: “You cannot automate ingenuity.”
Wherever innovation leads, marketing inevitably follows. What this means in practice is that anytime a new buzzword appears on the scene, people in that marketplace race to apply the label to their products. Ask a supplier of accountancy software how they define AI, and their answer will be wrapped like cling film around the product they are trying to sell you.
Because AI is so wrapped up in abstract concepts and cognitive semantics, the opportunities are endless: when does an expense algorithm become a machine-learning aid, and how can you prove it, short of examining the developer’s code? That's assuming they would let you and that you had the skills to assess the sophistication of the logic – both unlikely longshots.
By this point, every software developer involved in this market will talk up their plans for embedding machine learning and AI into their products, and like Shangri-La, the fruits of these efforts are tantalisingly just over the horizon. These ever broadening interpretations fuel cynicism and disillusionment among prospective purchasers.
A botwash backlash recently broke out in the US around the claims of outsourcer Botkeeper when tech commentators Blake Oliver and Patti Scharf argued that the company’s bookkeeping bots were human bookkeepers in the Philippines rather than the automated accounting robots the company energetically implied in all the communications.
Just as Bill Clinton once argued, “It depends upon what the meaning of the word ‘is’ is,” Botkeeper CEO Enrico Palmerino tied himself into all sorts of mental knots to justify his company’s ambiguous claims.
Listen to the Fox interview at minute 2:21 on, you will hear me clearly state botkeeper "enhances the workload performed by an accountant, allowing them to do more critical thinking and less manual mundane data entry: bookkeeping." AI = Do More with Less Resources (human trained)
— Enrico Palmerino (@EnricoPalmerino) February 20, 2019
On the strength of its disruptive, technology-driven approach to outsourced bookkeeping, Botkeeper raised $18m in Series A funding investors in November 2018. But while VCs have eagerly gobbled up anything labelled as AI, the marketing veil around AI has started to slip.
Even Google, feted for cutting edge developments like AlphaGo, was recently exposed in a manner similar to Botkeeper. Its Google Assistant relies heavily on massive data sets built by a squadron of overworked, subcontracted linguists.
“Artificial intelligence is not that artificial; it’s human beings that are doing the work,” a Google employee told The Guardian. Another worker called it a “White collar sweatshop”.
The academics Mary Gray and Siddharth Suri detail this reality in their new book ‘Ghost Work’. Beyond a fairly limited repertoire of decisions, Gray and Suri write, AI can’t function without humans in the loop. Human workers, like Botkeeper’s Filipino workforce, sweep up the loose ends.
The conception of Artificial intelligence, widely understood as processes making human labour obsolete, couldn’t be further from the truth – at least for now. For accountants, sold an AI bill of goods, along with the concomitant price tag, it’s certainly reason to maintain some degree of cynicism.
Tangible productivity gains
A degree of scepticism is clearly a healthy trait in the current market, but should not blind accountants to the real gains being achieved by the array of AI systems now being deployed within the profession.
MindBridge, quoted earlier, has been leading the way with audit aids and has broadened its reach in partnership with IRIS Software.
The millions invested by Big Four firms in these technologies are also beginning to emerge in public and interconnect, as an ICAEW IT Faculty post on a collaborative project called “Engine B” demonstrated. Engine B is the codename for a set of common data models (CDMs) for accessing client information. Now that leading auditors are on board, the big challenge is getting audit and accounting software developers to open up their databases and programming interfaces enough for the standardised models to operate effectively on different systems.
According to John Colthart, AI enables auditors, finance managers and regulators to test and report on complete transaction datasets rather than samples. “AI is not displacing the role of the auditor rather AI works alongside people to automate large data analysis tasks and provide new insights,” he said.
“By enabling auditors to dive deep into 100% of the data, clients can be advised on their financial health and compliance with more comprehensive evidence and greater confidence in risk assurance.” The potential savings are enormous, and could reveal new insights to help the profession deliver on its aspiration to deliver relevant, timely business advice.
You might also be interested in
AccountingWEB’s Editor at large has been with the site since 1999, rising from news editor to editor in chief, global editor and head of insight. As a roving editor, he continues to investigate the profession's use of technology around the world. He devotes his spare time to technology history and an oddball collection of stringed instruments...