by Kevin Mooney
Director, Customer Assurance & Trust, UiPath
The impact of AI at work has been inspirational. We’ve been inspired by stories of artificial intelligence (AI) working in tandem with people and automation to achieve greater business potential—accelerating the creative process, augmenting human abilities, and enhancing productivity. But we can’t ignore the capacity of AI to be used for harm as well.
The growing popularity of Generative AI models has raised many questions over their safety, security, and, in particular, data privacy. According to findings from MIT Sloan Management Review and Boston Consulting Group, 84% of AI experts and implementers believe responsible AI should be a top management priority. However, just over half (56%) believe it’s being taken seriously enough by business leaders.
Consider the damage that could be done by biased training data that warps AI decision making, or the capacity of these systems to spread misinformation on a massive scale. There’s a lack of transparency in how these systems work, and the risk that the data you share with them could be used for training and appear in front of another user.
To ensure the ethical use of AI, there’s an urgent and pressing need for proper governance. AI progresses too fast for policymakers to keep up, so the responsibility falls on creators and users to ensure their use of AI is beneficial and ethical. As an AI innovator that allows for the implementation of AI across our platform, UiPath is no exception to this rule.
This article will help to explain the various approaches to our implementation of AI at work, along with the controls put in place to protect customers who use them. As always, our approach is founded on our core AI and automation principles: open, flexible, and responsible, as highlighted here.
1. Creating an open ecosystem for AI excellence
We cannot believe a company will ever have a monopoly over the best AI capabilities. After all, ‘AI’ refers to a diverse range of different tools and technologies. There’s Generative AI, of course, but also Specialized AI which includes models that are trained for a specific business task or process, like document processing or sentiment analysis. Each has value for the enterprise and no company can ever be the best for all of them. We aim to maintain this diversity for the benefit of our customers and the AI industry.
That’s why UiPath follows an open approach to AI. We want our users to be able to combine the best of Generative and Specialized AI models. We enable them to choose from a list of top-tier AI models created by UiPath, or they can take complete control over their AI governance and utilize user-contributed ones, or even models from external providers like OpenAI.
Specialized AI models have long been developed by UiPath. Our platform provides numerous offerings such as UiPath Task Mining, Process Mining, and Document Understanding. Then there’s UiPath Communications Mining, which provides businesses access to a state-of-the-art large language model (LLM)—a foundational technology for Generative AI—to understand and automate their business communications.
Yet, we recognize the immense value of introducing features with Generative AI capabilities. We recently announced that UiPath Document Understanding and UiPath Communications Mining will leverage Generative AI to accelerate model training for users. Our upcoming Project «Wingman”, now available in private preview, enables customers to create powerful automations simply by using natural language prompts.
In all cases, AI capabilities are clearly marked to ensure UiPath customers know what their options are. This enables users to make informed and ethical decisions about their use of the technology.
2. Flexible AI that adapts to the user—not the other way around
It’s vital, as creators of AI solutions, that our models can adapt to users’ needs. When a customer has no choice but to choose an off-the-shelf model that doesn’t quite meet their use case, that’s when accidents happen. Accuracy means the world when you’re using AI to make decisions that directly impact your customers, or which need to be compliant.
That’s why we take model flexibility seriously. The UiPath Business Automation Platform is a flexible platform designed to meet the exact needs of our customers. The UiPath Platform enables users to build custom workflows by integrating multiple AI models, user interfaces, and application programming interfaces through UiPath Integration Service and Activities. These offerings make it easy for customers to integrate the model of their choice directly into the UiPath Platform.
What’s more, users can customize our models to align with their specific tasks and domains. Consider how different and unique every enterprise is. There’s no room for a ‘one-size-fits-all’ solution. Thankfully, tailored and precise automations can be created with the fine-tuning features of Communications Mining and Document Understanding. This results in improved accuracy, performance, and safety for everyone.
3. Guardrails for responsible AI
AI systems need data to improve. But users deserve the right to know their data is protected. UiPath has a responsibility to ensure data used to build our tools is of good quality, sourced lawfully, and securely managed. We do this in a variety of ways:
-
Legal and compliant data collection
-
Appropriate measures so data processed in the UiPath Platform is protected
-
Accurate and balanced data in UiPath models and algorithms to address bias in the training data
-
GDPR compliance, including for data deletion
The platform is actively reviewed for privacy, bias, and data security concerns, aiding the development of trustworthy AI. We’ve even achieved some innovation here. Usually, inaccurate or biased models are the result of poor training rather than bad intentions. It can be difficult, especially for ordinary business users, to know when their AI model is sufficiently trained and balanced. UiPath Communications Mining solves this problem through its Model Rating feature. The capability largely automates the process of model validation, assigning it a score for performance, accuracy, and balance. The UiPath Platform then gives users a fully guided improvement experience, showing them the steps they should take to improve the model.
UiPath promotes human-centricity and transparency in AI. We’ve given customers the tools to govern their automations embedded directly into UiPath Studio and Robot via Workflow Analyzer and Automation Ops. We’ve also developed strong standards to evaluate third-party AI and LLM technologies. Our aim is to ensure steps are taken to mitigate bias, bad intentions, and other threats.
Read the full list of responsible AI principles.
To make ethical AI, first enable ethical humans
Providing The Foundation of Innovation™ means embracing new technologies that push the world forward. In our approach to AI at work, UiPath enables customers to implement these powerful capabilities in multiple ways that align to their automation needs but also their security, governance, and risk posture. It’s important that these powerful technologies are deployed in ways that are as responsible and ethical as the people they serve.
Furthermore, we believe emerging technologies need to be democratized and tools placed into the hands of users to understand and use them accordingly. It’s why we’re developing AI-dedicated courses as part of the curricula in the UiPath Academy.
For all the stunning progress of the last few years, the AI story has only begun. It’s a story that will be largely told by its developers and users, who have a responsibility to distribute accountability and push for the ethical advancement of the technology. UiPath has their back. We’re committed to building trustworthy AI features, giving our customers the tools and information they need to use AI responsibly.
If you’d like to know more about the principles that guide our AI development, and see what’s coming next, make sure to reserve your place at FORWARD VI.