“We’ve had a year of hype with regard to Artificial Intelligence, a year when you could say anything without backing it up. As 2024 approaches, we’re getting to the point where you can’t just make a wild prediction and get attention. You’re going to have to say, ‘I’ve got this technology. I’ve shown it works. It benefitted someone.’ It’s becoming a ‘show me’ conversation.”
These are the observations of Matt Calkins, CEO and Founder of Appian. Dynamic Business sat down with Calkins to discuss how generative AI is evolving and the challenges and opportunities businesses need to consider.
Data security concerns loom large
There has been growing concern since large language generative AI interface models such as Chat GPT entered the zeitgeist a year ago that the technology is powerful but not private. According to Calkins, it’s time to have an honest conversation about the inherent risks of sharing your data in a public AI model.
“We have been talking for a decade now about how the most important asset an organisation owns is its data,” he says. “We haven’t internalised the fact that because AI is a way of accessing and synthesizing data, it has made your data even more valuable than it was before.
“I can’t find anyone who wants to upload their data to train an AI algorithm that exists on the internet, but they don’t own, even though they are all very excited about generative AI.”
How a private AI approach keeps data secure
In response to emerging data privacy concerns, Appian has developed a private AI approach that allows people who are unwilling to divulge their customer information and risk exposing it to a competitor or the public to still enjoy the benefits of generative AI.
Appian provides a set of algorithms that exist exclusively for the customer’s use. The customer controls access to these algorithms and the data used to train them, which protects their customer information during model creation, training and use.
Calkins sees the development of private AI algorithms as a way for data to be controlled by the organisation that is trying to create, teach, expand and make the algorithm more efficient.
He notes that Appian’s use of AI richly informs the question but does not inform the algorithm. This approach is known to AI experts as request augmented generation.
“There’s no training involved in a private AI model,” Calkins explains. “Instead, when you have a question to ask AI, you first query your enterprise before you send it. You ask all the databases and the data sources in your enterprise, ‘Do you have any data that’s relevant to this question?’ You gather that data in a little bundle and send it to an untrained AI along with the question.
“If you provide an untrained AI with a dataset of pertinent information, it does a pretty good job despite the fact that you never uploaded your database. In fact, in many cases, it does a superior job. A trained AI has difficulty learning certain things, so if your business changes and you want to delete some rows of data from the memory of AI, there is no good way to do that. AI doesn’t look like a relational database. You can’t just delete stuff. It’s a black box. Once you’ve trained it, you can’t take something away. That’s one reason why our approach is better than the public model.
“Another way it’s better is that you can apply low-level security. You can ask every question with only the information that the questioner is allowed to access. In typical AI, you train it on all the data, but there’s no good way of saying, ‘Please answer this question as if you only knew what a low-level employee would know. And please answer this next question as if you knew everything that the director would know.’ You can’t do that with [public] AI, but you can with private AI.”
Calkins says that Appian’s solution is more auditable than a public model. “When you get a result back from AI, you may not know why it would say something like that. Sometimes it is a mistake. Sometimes, you want to correct it, but certainly, you want to get to the root of why it said this strange thing. With private AI, it’s easy. You can audit it very well. You can look at the data you sent and see why the wrong answer is coming back. You can’t do that in trained, public AI because you have nowhere to look for what data informed the answer because all the data was processed.”
Why a private AI approach isn’t being universally embraced
Given the benefits of a private AI approach, why are the major industry players resisting moves towards this? Calkins says the reason is self-interest.
“These companies are gigantic, and they’re generally much larger than their customers. And so, they are all hoping, I believe, to use their leverage to force their customers to give them their data. They want to control their customer’s data, and they’re big enough to harbour hopes that someday they will; therefore, they’re not interested in an open data strategy. They want public AI where they can learn and train on the information they’re given. They are not acting in the interests of their customers.”
“Your data is yours alone. I wouldn’t play any games with that, and I wouldn’t depend on thin contractual protections for that,” he warns.
Integrating AI and process automation is a priority
When the AI hype began people speculated that it would take over our jobs and run our companies. Calkins notes that this reliance has yet to eventuate because there must be a synergy between people, AI, and existing business processes.
“AI is a partner to people, and we need to make decisions together,” he explains. “I call it mixed autonomy. Mixed autonomy where partly it’s the AI and partly it’s the person who’s watching and being careful. We’re going to have a long time in this mixed autonomy situation where AI proposes things, but we need a human with good judgment to sign off on everything. AI alone will make too many mistakes.
“Because AI cannot act alone, because it must be a member of the team, we need to find a way to get work to it and from it. We need to deal it into the complex interchange of responsibilities by which modern work gets done.
“There are business rules, there’s robotic process automation, there’s other things that do work other than people. Now, AI is going to be one of those. It’s a digital worker, but it’s part of a process. You see, the work is going to go from one entity to another and another, and AI is a station on that road.
“But we need processes to coordinate the way we’re going to work together with AI. We can’t just throw it the whole job. There are enormous and obvious synergies between process and AI due to the fact that AI can do some things exceptionally well but can certainly not be trusted to do everything.”
How private AI safely improves business efficiency
A low-code platform such as Appian orchestrates people, bots, systems and AI to improve efficiency while keeping your data secure. The benefits to Australian businesses include the facilitation of end-to-end automation and increased productivity by improving throughput. This is achieved by automatically routing work and processing content accordingly.
AI can be used within a low-code environment to build and refine apps faster and instantly create digitised forms and applications from existing PDFs. Businesses can also save significant time by extracting data via AI-driven document processing.
Low-code platforms which harnesses AI can quickly complete routine, high-volume tasks, which reduces errors and allows workers to accomplish more. AI also improves decision-making within an organisation by providing predictive recommendations based on patterns within a given data set.
AI has broad applications for customer service departments. The customer journey can be enhanced from the beginning of the sales process through chatbots that can field basic inquiries to the end of the sales process by analysing customer feedback for areas of improvement.
According to Calkins, a private AI approach allows you to achieve these business improvements without compromising the security of your customer data.
How should business owners embrace AI?
It is easy for business owners to become overwhelmed by the speed, divergence and complexity of the AI debate. Calkins recommends a cautious approach to using AI in your business.
“Every professional who cares about AI is engaged in a bit of a race to be the first in their industry or in their office to create real value using AI. We’ve had a lot of interesting discussions about it, but now I think it’s time to do something practical.
“I advise people to think in small steps, show value, don’t do a long project that may someday be impressive. Do a short project that demonstrates that you can create value without taking any risks at all. Take small, predictable steps with a partner you can trust. That would be my advice.”