Artificial intelligence (AI) discussions rarely veer into topics of regulatory or governance issues, instead focusing on the promised wonders and potential it holds.
But with its latest release, Cambridge, Mass.-based Pegasystems opened up that conversation. The latest version of its Pega Customer Decision Hub (CDH) includes a new technology which gives organizations direct control over the level of transparency within their AI customer engagement models.
Building Transparency Into AI
With the latest version of CDH, users can safely deploy AI algorithms based on transparency thresholds set by their business. Pegasystems also claims it is easy to use.
The transparency is offered through what it calls a "T-Switch."
“The T-Switch feature of our AI-powered Pega Customer Decision Hub enables organizations to set the appropriate thresholds for AI model transparency or opaqueness," said Vince Jeffs, director of Strategy and Product Marketing at Pegasystems.
“Businesses can predefine these levels for each AI model using a simple sliding scale from one (most opaque) to five (most transparent). The transparency scores guide business users to build AI systems using the right models to get the job done while also meeting their organization’s transparency requirements.”
For organizations working in highly regulated industries, this is particularly important, especially with the General Data Protection Regulation (GDPR) regulations governing the use of personal data of EU citizens coming into force next May.
Meeting GDPR Requirements
The GDPR requires businesses to readily show the data they have on European customers and how they use it to make decisions. This includes any new data and insights derived from AI models, putting businesses in the position of having to explain how their AI models work.Not all AI models are built with the levels of transparency businesses need to fully explain how the systems make their decisions. The Transparent AI Pegasystems is offering enables organizations to more clearly see why the algorithm made the decision it did, giving users control to fine-tune performance and to ensure full compliance with the transparency requirement in the regulation.
It may be possible, in situations that are not heavily regulated, to use opaque AI. Jeffs explained that while some opaque AI algorithms may drive powerful performance, the complex logic behind these ‘black boxes’ can’t be fully explained — a tradeoff that becomes more problematic when the model causes unintended actions.
The result is global companies are now scrambling to ensure they both have enough transparency built into their data and that their AI systems will be compliant before the May deadline.
A Growing Debate Around AI Ethics
“There’s a growing debate emerging about the ethical responsibility that businesses have to safely and responsibly deploy AI. Just look at the ongoing public sparring between Elon Musk and Mark Zuckerberg that started over the summer, and you can see this issue isn’t going away,” Jeffs said.
“In fact, it will only get more complex as more organizations tap into the new ways AI can improve customer experiences. Many of the red flags people like Musk raise are quite valid — we should not allow AI to operate unchecked, running amok and causing unintended consequences.”
Without digging too deeply into the dispute, Fortune magazine reported in July that Tesla CEO Elon Musk claimed AI is an “existential risk for human civilization.” Facebook’s Mark Zuckerberg retorted that comments like this are “irresponsible,” to which Musk tweeted a retort that Zuckerberg’s “understanding of the subject is limited.”
Pegasystems claims the launch of the T-Switch capability is a step towards helping businesses gain that understanding, control and governance.
Building Transparency Guardrails
It essentially puts transparency guardrails in place to guide users when deploying AI models.
Organizations will still need experts who understand how transparent or opaque its models are and to set the appropriate transparency thresholds, while business users can then select the right models to meet their objective that fall within those thresholds.
There is a concern, though, that transparent AI will add to the data load of workers who use that data to do business. To those concerns, Jeffs argues the new additions will make it easy to manage.
“If, for example, a marketer turns to the CDH to drive a new campaign, he or she will know how transparent or opaque the models are based on simple scale of one to five. This makes it easy for the marketer to know which models will keep the company within accepted compliance thresholds while still delivering the performance needed to do the job,” he said.
Pega has also taken a no-code approach to AI and its applications, meaning business users with no programming skills can take control for themselves if they choose.
Moving Into General AI
The upgrades to CDH reflect wider developments in AI and its ability to master and automate certain tasks.
AI systems are typically very one-dimensional in what task they can accomplish. An emerging trend is AI systems initially developed for one task taking that expertise and applying it to other tasks. This moves us closer to what is known as General AI.
Another emerging trend is using AI to more efficiently gather and refine the data needed for other AI systems to learn. This is one of the biggest bottlenecks to AI adoption today.
Pegasystems is already exploring other ways to use the T-Switch functionality.
“We will be making further advancements to the idea of the T-Switch, among many other AI advancements. For the T-Switch, we will be using AI to set the score, essentially using AI to help explain other AI systems,” said Jeffs.
The T-Switch capabilities within the Customer Decision Hub will be available by the end of October.