Entering the age of AI-powered digital employees
The future of work is increasingly digital, and with the infusion of AI into automation, we’ll see an accelerated adoption of intelligent digital employees that will support knowledge workers.
The future of work is increasingly digital, and with the infusion of AI into automation, we’ll see an accelerated adoption of intelligent digital employees that will support knowledge workers.
According to a recent McKinsey survey of executives, companies have pushed the time frame for digitizing many aspects of their business, from internal operations to supply chain and customer interactions, by three to four years. Digital products in those companies’ portfolios have also shot some seven years ahead of where they had expected to be prior to the pandemic.
The Great Resignation, skill shortages, supply chain disruptions, working from home, touchless customer experience, and agile process redesigns are paradigm shifts that businesses have rapidly needed to adapt to. But how do companies stay dynamic, resilient, and efficient in this new era? We believe that a big part of the solution may be found in digital employees, powered by automation and AI.
Digital employees rely on the latest in AI and automation technologies to collaborate with knowledge workers to automate mundane tasks and enhance their decision-making capabilities. Work that used to take hours can now be completed in minutes. Digital employees can send email, schedule meetings, filter resumes, or approve loans. They could serve as an interface for working seamlessly across the apps and tools that knowledge workers use daily. And the more they work, the smarter they get.
At IBM Research, we’ve been working to create cutting-edge automation tools that go beyond supporting data scientists and engineers. Instead, through digital employees, we can democratize automated support across entire business operations to any subject matter or domain expert.
For certain domains, a digital employee could take on tasks from a range of traditional job roles. An accounts payable digital employee, for example, could autonomously perform parts of three traditional job roles — a customer service representative, billing agent, and cash applicator or dispute resolver — to complete an order to cash (OTC) process. Since digital employees increase the bandwidth of a hybrid workforce, they can help companies reallocate their workforce to more strategic tasks.
We at IBM Research are working to build a successful digital workforce. To achieve this, we have to create what are known as augmented business process management systems (ABPMSs). These are AI-empowered, trustworthy, and process-aware information systems that can reason and act upon data within a set of constraints and assumptions, with the goal of continuously adapting and improving a set of business processes based on one or more performance indicators.
IBM Research, along with several leading academic institutions in the business process community recently published an augmented business process manifesto that outlines what a system needs to achieve to actually be considered a digital employee. They have to be:
- autonomous to act independently and proactively
- conversationally actionable to seamlessly interact with humans whenever necessary
- adaptive and dynamic to react to changes in its environment
- self-improving to ensure the optimal achievement of its goals
- explainable to ensure the trust and hence the cooperation of the human agents
IBM Research’s automation team has been working on building new automation and AI-based technologies that can mimic workers’ capabilities and inform a future digital workforce that allows machine and human colleagues to collaborate more intelligently since 2019. Our vision and technical advancements can be found in IBM’s newest product: Watson Orchestrate.
To fulfill our goal of building useful digital employees, we had a twofold strategy:
Intelligent orchestration with AI planning
In automation, enterprise users can identify a goal (such as approving a loan application or processing an insurance claim), describe it in natural language, and leave it to our algorithms to piece together the logical steps needed to achieve that goal.1 One of the methods IBM uses for this composition is called AI planning, a sub-area of AI that can create and apply complex action sequences to achieve a goal in an explainable way.2 Intelligent Orchestration enables composability, reusability, and dynamicity through modular components.
AI planning is notably non-deterministic in nature, meaning the AI can help an employee determine a step-by-step path forward, even if unexpected circumstances come up. By creating systems that can factor in the unexpected, we can help keep business operations flexible in ways that are not possible with traditional deterministic workflows or sequencing systems.
This intelligent orchestration drives Watson Orchestrate’s runtime experience, allowing digital employees to collaborate with human workers on various tasks. (Watch our tutorial on AI Planning.)
Bootstrapping new capabilities by easily transforming existing APIs
Watson Orchestrate requires modular skills to orchestrate. However, skill creation can be tedious, time consuming, and may require programming capabilities beyond those typically found in line of business users.
IBM Research has created multiple technical solutions to address these issues by leveraging accessible APIs that automate workflows. These include approaches that leverage deep learning models3 and business knowledge to generate training data for skills.
More work needs to be done to better support businesses beyond current solutions that combine automation technologies with AI. One area in particular is the limited availability of developer talent. Many IT projects get relegated to the “pending” file due to a shortage of resources with specialized technical skills. As a result, operational inefficiencies continue to exist and time-to-market — a crucial factor for businesses to remain competitive — is compromised.
A viable solution to this challenge is what has become known as low-code/no-code (LCNC) development. The user interface of a LCNC platform can automate aspects of the development process, eliminating dependencies on more traditional computer programming skills. LCNC lowers skills barriers, particularly for business users and domain experts with little formal coding experience, such as analysts or project managers.
Both professional developers and non-technical users, though, can benefit from the core capabilities, such as a visual integrated development environment (IDE), a graphical user interface, built-in data connectors, APIs, and pre-built templates. Multi-modal interactions through NLP, voice and visual recognitions, plus semantic knowledge through ontologies and foundation models, are critical enablers for next-generation low code/no code development and user experience.
Another ongoing challenge is the need to ensure the trustworthiness of digital employees with AI explainability. Specific to automation, we must ensure transparency and fairness in how a system learns, remembers preferences, adequately reasons, and acts towards the attainment of users' goals. One way we can do this is by offering situation-aware explanations (or SAX) that consider the context in which subsequent choices ae made, such as user goals and assumptions.
Recent IBM research addressing this need includes IBM’s AI Model Explorer and Editor (or AIMEE), a tool that allows for the creation of an interpretable rule-based model where the rules become the model, or can be used to give the end user a better understanding of the decision-making criteria their machine-learning model has learned.
Digital employees have enormous potential to shift how we work and improve overall productivity of a hybrid workforce. But for them to succeed, businesses must embrace AI and automation techniques that save time and cost in favor of more mission-critical tasks. To learn more about how IBM Research is advancing automation and AI, visit our team page.