By MARTY LEVINE
AI’s usefulness and capabilities are expanding every day, said Victoria Lancaster, but soon the price will be going up — and its security risk “keeps me up at night,” she added.
Lancaster — recently named executive director of AI Service Delivery at Pitt Digital — spoke on Sept. 18 at the latest Staff Council Spotlight event.
Lancaster noted that she is the only non-technologist at Pitt Digital – which is perhaps why no one at the event asked about AI as an existential threat, despite current news stories. “I honestly believe that AI has the ability to solve a lot of the world's problems,” she said. “I believe the technology, when used appropriately – I hate to say for good, not evil – can really accelerate key research that's being done.
However, she added: “You've heard me say this for a long time: you need the human in the loop. Number one, we need AI governance, which means that we need senior leadership from the top down to say, yes, this is what we're going to do with AI; this is how we're going to protect you. I lose sleep because I am worried that people are going out and buying whatever AI model they want” in addition to, or instead of, using the free offerings available at Pitt. Sometimes these acquisitions are created by specialized research needs, she allowed, but at other times they may occur simply because users are unaware of what Pitt already provides.
“I will tell you OpenAI has 15,000 Pitt emails. There are 15,000 people that have purchased OpenAI, and they have a Pitt e-mail. That's not covered” by Pitt’s contracts and agreements that keep Pitt data secure when using AI directly through Pitt Digital. “I don't have any security over them. I can't stop them from putting data into that and if they lose the data, if it gets stolen, there's no recourse for it. I lose sleep over that because of not even what can happen to the University but what can happen to somebody's lifelong research. It makes me cringe, and yet people are still willing to put things at risk. I don't think they're doing it on purpose. I think they don't understand or don't know the risk associated with it.
“We have contracts with AI vendors,” Lancaster added, “because they have met our terms and conditions and they are indemnifying us and they promise to have what's called a zero data retention rate, which means they're not learning from our information, they're not keeping our information. We are essentially in a locked environment. … We don't want anybody getting into our data and looking at it. We want to have a secure environment for everybody to use. The problem is that people want the newest and the greatest AI models and I get it. … But you have no recourse if that model gets hacked.
“I'm negotiating with Anthropic right now,” she said of the company behind Claude (and that Pitt has an agreement with), “and they're being tough about indemnification. They want to be held harmless if somebody breaks into our AI and takes some data. I'm hoping that we can pull together a group of senior leaders that are willing to collaborate and own the AI decision-making together. I will tell you that the people the level down from senior leaders, those of us who are working on this and up to our elbows in it, we'll do the lift on this, we will do all the research, we will look into things. We will do all the due diligence – we just need senior leaders to come in and help us make a decision. We'll provide them with all the facts and information.”
Nonetheless, Lancaster told the virtual gathering, “I can be very excited about this kind of stuff. I think AI makes things very possible, not just a good Excel spreadsheet. I think it helps us learn, grow and engage in our work. I really believe, particularly for us on the staff side, that AI is about a community and how we use it together. And the more that we use it together, the better we're able to share information.
While the AI familiar to most individuals, such as ChatGPT (from OpenAI), has “lowered the barrier to entry for all of us,” she said, with its use of human-like interfaces, AI is now capable of so much more: “If you don't know how to code, I'm telling you you can learn how to code with artificial intelligence and you will be amazed by it. It's called ‘vibe coding.’ We're going to do a lot of training on that.
“I had Claude draft my job description when I moved to Pitt Digital,” she said. She now uses Claude for PowerPoints and other “low-level” tasks, but notes that Anthropic, Claude’s manufacturer, was not able to guarantee its compliance with HIPAA (the federal rule for health care record privacy). “I'm currently in negotiations with them” to right that situation as well, she said.
In the meantime, Lancaster said she is more often using Microsoft Copilot, which is built into all Microsoft applications, and using Google Gemini as her “third-tier” AI application. It helps that the Microsoft “harness” – its group of AI offerings – is both secure and HIPAA compliant, she said.
The best harness is available through Pitt’s own AI Connect – 24 AI models, at the moment – “the most powerful AI models that we can get,” but it takes some added know-how for access and use. “As a matter of fact, we're going to create trainings on it, if you want to get into some heavy-duty work with AI and you want to do some analytics or complex coding.
“If you want information on what platform to use, I would come to Pitt Digital,” she said.
Her job, Lancaster explained, is to understand each AI company, what current and future platforms they offer “and at what point do we need to retire those and start to look for other things. I'm not exaggerating: AI changes every day. Every day I have a conversation with somebody and I learn something new or something that used to be available last week has been upgraded as a better model, so I'm trying really hard to keep tabs on what's happening and make sure that it meets the needs of the University and is secure.”
Pitt is also “actively trying to establish a subject matter expert to be the point of contact for” specific AI uses. Together, these experts will also be examining each new AI tool, to determine which ones to choose for what task.
She is also working on a certification program to get Pitt AI users past AI basic training to understanding it at a deeper level.
At the moment it is hard to judge our return on investment for each product, she noted. “We are watching people do things and create things that have saved significant amounts of time,” she observed. But that will be even harder in the future because AI companies are moving from offering institutions a “seat license” for use of their software to a “consumption-based” payment model, in which users must buy tokens to spend as they work (or through which they will be allocated a certain number of tokens from their employer). The more complex the task, the faster the tokens will run dry.
“I have talked to Google and Amazon and I said I need some case studies in consumption models for higher education – how much are they spending on things – and nobody knows,” Lancaster said. “It's like the Wild West. However, there's a lot of data in corporate America and we've been able to extrapolate based on similar activities in higher education.”
She added that she has been working very closely with the CFO’s office and with the Provost on multiple questions: “How would we pay for this? Will we offer people a certain amount of tokens per month How much will they get?”
But AI companies will also be creating new tools: including “cowork” tools, which put the “agent” in agentic AI. “Coworks can build agents for you – they’ll burn your tokens but boy the work you can get done is significant,” Lancaster said. The latest trend is toward “multiplayer AI … when multiple people on a team have an [AI] agent and they feed that agent information and requests and that agent is then able to manage multiple people's work. This is really important in a research environment, but it can be used in any kind of team environment.”
Lancaster urged AI users of all levels – even those simply curious – to join the community of practice she has run for several years at Pitt: “We have over 400 people in this community of practice, and we represent every RC in the university.”
She concluded: “There are very few jobs right now that aren't requiring AI and … it's going to redesign workflows. It already has and it will continue to. If you have something that just makes you bonkers crazy – “Why do we do it this way?” – that's a great question to pose to artificial intelligence.
“I want this tool to help us, particularly staff, to do our jobs better, smarter, longer and I want to make sure that you guys have tools that are safe to use and you don't have to worry about.”
Marty Levine is a staff writer for the University Times. Reach him at martyl@pitt.edu or 412-758-4859.
Have a story idea or news to share? Share it with the University Times.
Follow the University Times on Facebook.