By SHANNON O. WELLS
In his presentation to Faculty Assembly about how artificial intelligence (AI) is being incorporated into Pitt infrastructure and learning culture, Mark Henderson stressed that the rapidity with which the technology has developed — along with the rise of its providers — cannot be overemphasized.
“If you take Anthropic as an example, this is an outfit that didn’t exist five years ago,” he noted. “They now are looking at a trillion-dollar valuation right in the eye. You’re going to break a whole lot of eggs in achieving that kind of momentum, and they’ve not been able to hire to keep up with their growth.”
What’s great for the rapid expansion of San Francisco-based Anthropic, however, is not always the best for its clients, including Pitt.
“We’re having troubled in communications. We talk about the left hand and the right hand not knowing what’s going on? The first digit and the second digit don’t know what’s going on at Anthropic. And that plays out almost by the day,” said Henderson, Pitt’s senior vice chancellor and chief information officer. “We’ve had instances where they said we’re going to do this, and then we go to bed breathing a sigh of relief. (The next day) we find that that’s been undone.
“It’s a challenging but exciting environment.”
Henderson, along with Victoria Lancaster, executive director of AI Service Delivery at Pitt Digital, addressed Faculty Assembly at its Oct. 7 meeting held in Posvar Hall on the University’s adoption and strategic implementation of AI and its related tools and software. Pitt is negotiating a contract renewal with Anthropic, known best for its Claude model, while also working with exploratory or “frontier” agencies as well as assistive services like AI Connect. Pitt Digital also offers options including Google Gemini, CoPilot (as a premium service), and Microsoft Chat.
“We are currently in negotiations with Anthropic, looking at that contract that we have,” Lancaster explained, noting it’s “really interesting” to work with o. “We are talking to executive leadership and talking to Anthropic about what is it going to look like for us to renew with them.
“What is it going to cost us? What are they going to offer to us? And is it going to be a safe and accepted technology?”
Since joining Pitt Digital in recent months, one thing Lancaster’s charged with is how AI can be delivered safely to the various facets of the Pitt community. Safe handling of the technology, she emphasized, is her “north star.”
“I’m a huge proponent of administrative artificial intelligence, and I really like it. I want people to have access to the tools, and I don’t want to be an impediment to them, but I really want it to be safe,” she noted. “That’s the thing that we’re working really hard on at Pitt Digital, to make those negotiations and try to understand what’s out there and try to give the best resources we can to the University with help from executive leadership.”
Echoing Henderson’s assessment, Lancaster said elements of the AI universe “are literally changing every day … New models are coming out. Things are happening, and we are constantly keeping an eye on what’s happening at the higher education level and how we can support that.”
Keeping options open
Although Henderson described Anthropic as early “out of the gate” and the first to focus on enterprise level AI technologies with its Claude platform, Pitt Digital is “watching very closely” as all the so-called “frontier labs” are approaching parity.
“There will be some that specialize, in Anthropic’s case, and some scientific applications that we want to use the very best for the challenge at hand,” he said, but multiple tools always will be available. He added that there’s “nothing new about AI. It’s been around for a long time.
“What’s new is that computing horsepower and internet speed and availability of data has allowed AI to reach what had been forecasted (in the 1970s) … It’s a challenging time now.”
Henderson explained that Anthropic’s price advantage proved appealing when Pitt signed a one-year deal, along with the University’s longtime partnership with Amazon Web Services (AWS), which worked with Pitt Digital and Anthropic to help bring Claude to Pitt.
With the improvements all AI labs have made, some may be better than others in specific areas but are approaching parity, he noted. “(We are) re-upping with Anthropic, assuming that we get to a reasonable position with them … As we all know, they are planning to (take the company) public.” All of the labs are having problems “proving a return on the trillions of dollars invested in them.”
“Our legal team is involved, my team is involved, others are involved, really looking at the whole package,” he added. “There may come a time where other tools will become much more attractive, particularly from a cost perspective.”
Safety is No. 1
Responding to a question about “personalized” AI agents such as Muse by Meta and ChatGPT’s Instinct platforms, Henderson said Pitt Digital is “paying very close attention to all developments. We experiment with a lot of things.”
He added that he’s “pretty firm” that whatever company or tool adopted at Pitt “has to be safe, and just because it’s out there and it’s available doesn’t make it safe. It has the potential to exponentially be dangerous to our environment.
“All of these things have to go through a security review.”
The focus is on a single entity because “we cannot secure the homeland if everything is available in our environment …
“There are some folks in in our community who bring in their own (AI program) and load it on their devices, and I’m here to tell you that is a horrible idea,” Henderson stressed. “Everything you put into these systems, it is out there forever and available to anyone who happens to come across it in their prompts.”
“I know that the temptation is great because these things are changing so swiftly, and things in the news are hitting the marketplace almost hourly in some respects. Please don’t give in to temptation, because the downside is monumental.”
Victoria Lancaster mentioned non-frontier AI models that are “widely secure and really good,” such as Copilot and Google. “Talk about parity. They are coming up fast to the frontier models, and they have security to them. We have good contracts, especially (with) Microsoft. We have a deep relationship with them.”
What keeps her up at night regarding all this, she confessed, is “really the security issue of using a frontier model or a beta model in with your private data or private research, especially if you don’t understand AI hygiene,” she said. “Even if you toggle (to) ‘Do not train the LLM,” it doesn’t stop it. We are a contained environment which can protect your data.”
She also emphasized the importance of ethical use of tools, regardless which supplier company works with Pitt.
“I keep reminding myself, ‘We’re not any different from any people. Professors aren’t inherently more ethical than any other people,’” Lancaster said. “So, how should we be thinking about kind of reining in, not only the unsafe uses but the potentially really unethical uses?”
Henderson said those considerations “factored largely” into Pitt’s decision to work with Anthropic.
“They had a constitution as to how they’re going about doing their development and were formed by a group of folks who broke off from OpenAI because of questionable ethics in the development of their models … We started our AI journey really with safety first, so we began by creating appropriate guidelines in the use of AI technologies,” he added. “Because it’s moving so quicky … we try to update them.”
Responding to a question about faculty members buying their own programs when what’s available doesn’t jibe with their needs, Lancaster said she wants faculty to have the best tool available, “but it has to be the safest.”
The original arrangement with Anthropic was thought to include “a little bit more of a secure environment, and it turned out they had a little change in some things, and so we had locked everything down,” she noted. “As we’re negotiating with Anthropic, we are looking at a much broader portfolio.”
She noted the availability of AI Connect, which makes 24 AI coding tools available. “If you really want something (for) advanced calculations, we have something for that … (Also) Microsoft Copilot is coming out with more advanced tools, too, that are safe.
“We know that you guys need different tools and better tools,” Lancaster added. “We’re just trying to get it safely to you.”
Regarding sustainability and the University’s environmental impact from AI use, Henderson said Pitt “probably burn(s) more fossil fuels (or is) less compliant to sustainability through the use of YouTube.
“Our use of AI right now — the massive consumption of power and water and land and building these hyperscale facilities — is almost completely the domain of training the (LLMs) … We do not do that work at (Pitt), so I can’t give you a hard number in terms of our compliance. I would say that we are likely compliant …”
Henderson emphasized that Pitt Digital has endeavored to “communicate with the community as much as we possibly can in a timely fashion around these changes or developments.”
Because of the possibility of not reaching a renegotiation point, Pitt extended the contract deadline with Anthropic from September to December.
“We continue working very hard, and it’s our intention to re-up. But there’s a window now until December that if, heaven forbid, we cannot get to a good place with them for any number of reasons … there’s time to react and time to think about change management, which was one thing that was very important to us …
“We will work with the community alone,” he said, “and be very transparent as this continues to unfold.”
Shannon O. Wells is a writer for the University Times. Reach him at shannonw@pitt.edu.
Have a story idea or news to share? Share it with the University Times.
Follow the University Times on Facebook.