AI’s Future: Combining RPA With AI to Augment Knowledge Workers

Counter-terrorism involves analysts digging through millions of Twitter and Facebook messages, YouTube videos, and websites in multiple languages, which is way too much work for humans. A combination of AI (Artificial Intelligence) and RPA (Robotic Process Automation) can help people with this work, leaving people in charge of the most complex and important decisions. AI systems can search documents in any language, automatically translate them, extract names of people and organizations, and conduct sentiment analysis for conversations to identify key texts that should later be considered by humans. This text can be automatically organized into appropriate bins using RPA. This enables a data processing and analysis pipeline that can process large amounts of content at a rate never possible in the past.

This is a very different result from the predictions made by many AI experts that AI will replace accountants, lawyers, financial and security analysts, doctors, and other knowledge workers. From Erik Brynjolfsson and Andrew McAfee’s Race Against the Machine to Martin Ford’s The Rise of the Robots and Daniel Susskinds (2020), many books argue that AI will result in a combination of great wealth and massive unemployment, and therefore requires huge payments to the unemployed .

We see another world where computer systems primarily promote knowledge workers. While there are many ways this scenario could develop, this article describes how a combination of AI and RPA can automate simple tasks that are more difficult for humans, usually judgmental tasks where computers have performed poorly in the past to have. As can be seen in the high-end counter-espionage example, we expect the number of these automated tasks to increase in the coming decades as systems improve, also as the underlying AI and RPA get better.

What are the easiest tasks for RPA and AI? This depends on the degree of standardization of documents, other data sources and processes. For documents with fixed structures such as invoices, receipts, passports and licenses, the work processes and decision automation paths are usually clear and standardized. In this way, RPA can extract most of the key information and store it in other documents, be it in Excel files or in company information systems such as those provided by SAP or Oracle. It does this by mimicking the actions of human workers, recording their clicks, determining the places where people must make judgments, and the rules that are followed. However, when the document structure and other data inputs are less standardized, the only way to get RPA working is to use AI to transform the input data and extract meaningful information from it.

The fastest improving subset of AI is natural language processing. NLP is getting better at transforming, translating, interpreting, classifying and categorizing documents or parts of them, and determining the location of data in documents, even if the layout changes. The latter allows RPA to extract more data and make more decisions based on that data. The former can allow AI to assign different parts of the text to different people, reducing the need for everyone to read the entire text.

However, in assessing the various tasks that AI or RPA can perform, it does not matter whether AI can understand the documents or answer simple questions about them and the person. This topic is discussed endlessly by people evaluating OpenAIs GPT-3. This is a completely different topic that we believe is much less relevant to improving productivity in the 2020s. Instead, it’s about whether AI can roughly understand text and classify information to reduce the work of knowledge workers. Terms like “intelligent automation” and “hyper automation” have been so misunderstood and overwritten that expectations of what RPA and AI can achieve are unrealistic.

This path will lead to very different challenges than those of Brynjolfsson, McAfee, Ford and Susskind. Rather than unemployment and guaranteed income, the challenge is to understand what part of white collar work, including that of professionals, can be done by AI and RPA in the short and medium term. This requires an understanding of the processes carried out by professionals such as financial analysts, accountants, lawyers, engineers, doctors, nurses, and architects.

Look at financial analysts. You need to make deposits over billions of dollars based on large amounts of structured and unstructured data. How can you evaluate the data quickly and effectively? Trying to automate all of a financial analyst’s work is a large and risky endeavor that is likely to fail. Instead, AI can search the Internet for articles about various companies, automatically translate, summarize, and classify them, and extract and summarize financial data for analysts to take a closer look at. RPA automates the tasks of summarizing, classifying, and extracting information by mimicking what people once did while slowly searching the web. The better NLP gets, the better RPA and NLP can organize the information for financial analysts together so that they can take a closer look at the relevant information.

Consider contract lawyers whose job it is to review, understand, and act on contracts. The complexity of these documents should be underestimated. Nobody can understand an entire contract, especially not a newbie who often clicks on “I agree” without reading his mobile phone contract. While NLP may not be able to write legal briefs to replace attorneys, it can assign sections to attorneys who specialize in different types of contract law, thereby reducing the overall working time for them. The reason AI can do this is because many contracts are organized in a similar way and therefore NLP can be trained in similar contracts to learn which parts should be assigned to which lawyers.

Look at the bookkeeping. Downloading reports and simply calculating data in these reports is relatively straightforward. Data can easily be extracted from standard invoices, spreadsheets and other standardized documents and entered into corporate information systems such as those provided by SAP or Oracle. However, the number of calculations RPA can perform often depends on the number of rules associated with those calculations. As the number of rules and thus contingent liabilities increases, RPA requires more complex code and more NLP. Additionally, the need for AI increases as we move from simple accounting tasks like accounts payable and reporting to more complex tasks like auditing.

To sum up, the future of AI can be seen in decades of incremental advances that will primarily advance knowledge workers. This conclusion is consistent with the past 40 years of employee progress. Computers provided employees with document, spreadsheet and powerpoint functions in the 1980s and business software in the 1990s. Although some professions such as secretary, data entry clerk, and accountant have been eliminated, advances in automation have made the work of accountants, journalists, engineers, lawyers, and architects much more work. This process of incremental change and expansion is likely to continue with RPA and AI.

You might also like to read: We Will Never Return To Work Before COVID. The virus forced us to realize: it has never been easier to stay together. (Jonathan Bartlett)

May 21, 2021