The Pause and Possibility of AI
The artificial Intelligence race heats up as development and access explode.
In the last few months—since OpenAI launched ChatGPT in November 2022—artificial intelligence (AI) developers have been in a race. Though birthed in the 1950s, the field of AI has seemingly exploded overnight, and what started as a friendly jog in the AI park has turned into an all-out sprint to the finish line. Amidst the onslaught, there is an evolution happening that will forever change society … and business.
Taking a Breath
There isn’t an industry or business existing today that won’t benefit from or be touched by AI in some way, says Peter Stone, founder and director of the Learning Agents Research Group (LARG) within the Artificial Intelligence Laboratory in the Department of Computer Science at the University of Texas at Austin. AI has been the focus of Stone’s work for more than 20 years.
Nonetheless, Stone is among a growing number of businesspeople, educators, developers, philanthropists, and more who are saying, not so fast. Many of these, including Stone, signed an open letter released by the Future of Life Institute in March of this year urging a six-month pause in development of AI systems greater than OpenAI’s GPT-4 to allow the industry and the public to take a breath. (At press, the letter had more than 25,000 vetted signatures, with thousands more waiting to be verified.)
“I feel we don’t yet understand the social, political, and economic impacts that are going to arise from the most recent large language models [LLMs],” Stone says. “People are still uncovering the potential ways they can be used by people, the ways they can be misused by people. People are still trying to figure out what the potential startup opportunities will be. We have innovation opportunities. We still don’t understand what the privacy, copyright, and intellectual property implications are and should be.”
To be clear, Stone is not in favor of slowing down AI research in general, he says; he just believes that the specific development and deployment of ever larger LLMs should be done more cautiously. He doesn’t think a pause is realistic nor the letter perfect, but the effort to bring its possibility to light was important to him. He uses the evolution of the automobile to illustrate the importance of taking some time to ease into the rampant application of AI.
“Imagine if we went from the Model T to everybody having a car that can drive 80 miles an hour in an interstate highway system and roads all through all of our cities without yet understanding what kinds of accidents can happen, what traffic laws should be in place, what sort of regulations there should be on emissions, and what kinds of insurance policies should be in place—that couldn’t have happened in six months to a year.”
Regardless, AI technology growth seems to be coming fast and furious.
Why AI
Artificial intelligence is a broad field and has a host of applications. In Switzerland, Firmenich is using it to elevate the taste of its plant-based culinary creations. In the UK, the Ben Kinsella Trust, an organization working to stop knife crimes, partnered with a media company to use machine learning in outdoor screens to listen for ambulance sirens and then post-campaign prompts on the screen. Closer to home, Stone is using AI to develop a soccer team of humanoid robots as a part of the RoboCup initiative, with the hopes of beating the best World Cup soccer team on a real soccer field someday, as well as working on general service robots to help with common household tasks.
There has been significant chatter lately about generative AI (applications such as ChatGPT and others that can write text and code, and create images, illustrations, video, and more), but the application of AI in all its forms holds a host of possible uses for businesses across the gamut.
CEO Carlos Martin specializes in robotic process automation (RPA) and AI at Austin-based Macami.ai, where the team tries to demystify these technologies and create process improvement and optimization for their clients. For companies looking to digitally transform their business, Martin says, there are some important questions to ask first.
“The question is: Why are you
doing a digital transformation? Why are you using this technology?” Martin asks. “The reason behind the ask is always: How can I be more competitive? In my experience, as somebody in technology, you always need to cater to the question behind the question.
“The answer needs to be from the perspective of cash flow, from the perspective of the balance sheets, from the perspective of optimization of the operation of the company.”
IBM Distinguished Engineer and Master Inventor Romelia Flores echoes Martin’s C-suite queries and adds that the executives she works with are also looking for agility. They don’t want to be slaves to old technology paradigms. They’re also looking to retain and grow their talent base to execute that agility, using the latest technology to attract top-notch developers and keep pace with the growing speed of business.
Another concern, Flores says, is mitigating external threats. “Everyone is always scared of AI. I’m probably more scared of the hackers that are out there,” she says. “A lot of people have a lot of time at home, and they’re developing systems that can be harmful to others. They’re looking for ways to tap into financial records, looking into, ‘How do I tap into other systems in the environment so I can try and hack them?’ for fun or for benefit financially.” AI systems need to be prepared to recognize these attacks and behave accordingly to mitigate them. In terms of AI, it is important to also leverage appropriate tools and techniques that enable an organization to monitor their AI models, ensure the data being produced from their models is explainable and that data drift and bias detection is assessed. These tools and techniques will enable an organization to deploy AI with trust and confidence.
“The biggest concerns that I hear a lot of our C-suite executives describe to me are those of agility, skills, and security,” Flores says. Addressing these three key elements is important when considering an AI system.
Data Is Everything
The challenge for businesses that deal with significant amounts of data is being able to use that data in real-time to impact a current business scenario; that’s where machine learning can shrink the timeline so businesses can act.
Striveworks, a machine learning ops company in Austin, is currently helping businesses make sense of their data. Its platform, Chariot, makes it easier for businesses to build models or algorithms quicker and cheaper.
“All of these models, artificial intelligence models, machine learning models, everything else—one of their core underpinning assumptions is that the data they’re trained on, the data that’s used to create the model, has to look like the data that’s being used in the real world against that model as well,” says Striveworks CEO James Rebesco. “When that assumption breaks down, you have a lot of things go bad.”
“Bad” can mean a few things, he says, like when the AI says something like “this pen is a duck” because it wasn’t trained on context. An even more troublesome outcome can be bias or unfairness based on baked-in assumptions.
It’s data in, data out.
“Models don’t exist in a vacuum,” Rebesco says. “They have to be tied to a use case, to an end user who wants to achieve an outcome, and it has to be built for a purpose. It can’t just be built in a vacuum, and you just pull it off the shelf and hope for the best.”
Martin agrees. “To boil it down, it’s all statistics. You have an algorithm, then you have all the data that you feed it, and then you give it the parameters for learning. If anything falls outside those parameters, you have your outliers, and you need to pay attention to those.”
Identifying those outliers, the mistakes that happen in an imperfect system, which Rebesco says they all are, and correcting the data either manually or through a model function feeding correct data back into the system helps it learn.
The big successes of this are the Amazons and Facebooks. “They’ve closed the loop,” Rebesco says.
“The businesses that are really excelling at this are the ones who have figured out ways of capturing that decision and feeding it back into their model development framework,” he adds. The ones that aren’t successful in this are likely experiencing frustration and not seeing the ROI they expected.
Martin, Flores, and Rebesco see immense potential for AI for any business or industry looking to drive decision making through access to real-time aggregated data. Indeed, the aforementioned generative AI is being used by more than a third of US marketing (37 percent) and technology (35 percent) professionals, according to a recent survey conducted by Statista. Healthcare industry adoption came in with the lowest usage rate at 15 percent of the professionals surveyed.
Bad Actors
As with other technologies, those in the field expect there will also be bad actors who use AI for harm. In fact, the AI Index 2023 Report found misuse of AI technology had been on the rise since 2012, with an increase of 26 times in that timeframe.
“Many technologies can be used for both good purposes and nefarious purposes—9/11 was airplanes being used for bad purposes. That is a very real concern with artificial intelligence technologies as well,” Stone says. “So, regulation is a tricky question because I don’t think you can regulate artificial intelligence as a whole.”
In the 2021 report of Stanford University’s One Hundred Year Study on Artificial Intelligence, the committee’s conclusion found that governments have a role to play in regulating influential aspects of AI technology along with creating informed communities, starting with elementary and secondary education, that can live in, work in, and contribute to an AI-infused world.
What that role looks like in practice is not quite clear yet. The Future of Life Institute’s open letter also calls for policymakers and developers to accelerate the development of “robust AI governance systems.”
“Any kind of regulation focus on AI should be targeted by sector,” Stone says. “Thinking about how we should regulate the use of artificial intelligence in cars, how we should regulate the use of artificial intelligence in healthcare, or even more narrow than that in radiology, that’s where we can try to make sure that either the intended or unintended consequences are mostly positive.”
The Next Iterations
Last summer at the YTexas Summit Launch, held at AT&T Stadium in Arlington, a woman’s face watched professionals mix and mingle from big-screen TVs mounted around the event. She smiled and nodded as if she were talking to someone the group couldn’t see. It was hard to make out until YTexas CEO Ed Curtis introduced Sai Bezawada, principal technology leader–healthcare at IBM, and he explained that she was an artificial intelligence program—a digital person—developed through a partnership between IBM and Soul Machines out of San Francisco.
A version of that digital person, Digital Iris, was installed at the Dallas/Fort Worth International Airport (DFW) in terminal B in mid-October 2022. She was the product of Soul Machines’ avatar, and IBM’s Watson Assistant technology, an AI that IBM developed in the aughts that would later compete and win at the game show Jeopardy! Since then, Watson has been deployed by IBM clients around the world and across a range of industries to deliver customer care experiences, such as responding to time-sensitive COVID-19 inquiries, helping citizens get more information on voting procedures, and helping insurers provide more personalized services, among others.
Two years ago, DFW began researching ways to deliver better services to senior travelers and learned they were not as comfortable digging into websites and other technology but rather preferred talking to someone. During this process, the DFW team learned about Iris.
“It was always intended to be a test-and-learn initiative,” says Jodie Brinkerhoff, vice president of Innovation at DFW. “We knew that
the technology was evolving. We knew that digital avatars were going to be improving. What we were really trying to figure out was: Is there a solution in here that, over time, becomes more viable for us? We continued to do the iterative research and the prototyping to bring what was the most recent version of Iris to life.
“The process of rolling her out was designed to collect information and to understand: What is the desirability, feasibility, and the viability of this as a solution to meet the needs of some of our customers?” Brinkerhoff adds.
Earlier this year, the testing phase for the Digital Iris project finished at the airport, and she was taken out of the terminal. However, in that six-month testing phase, thousands of travelers used Iris to find information about the airport. By installing Iris and testing her in real time with real travelers, the team surfaced issues like ambient noise and language diversity among others that needed to be addressed, and that feedback is helping the team evaluate next steps.
“Now we have a really robust set of iterative improvements that we want to tackle,” Brinkerhoff says. “Even in the short six months that we had it up and running, the technologies have continued to develop, and the skill set of our people has continued to develop. So, we’re looking forward to a phase three in working out some of the kinks that surfaced during our pilot.
“Where we have an opportunity is to really think more about all of the systems that will inevitably go into creating a successful, productive environment,” she says. “You really only surface all of those things if you go out and try. … The learning that revealed itself was just a reminder that testing, real testing, in a real environment is so critically valuable to longer-term iterative design, and the voice of customers along the way in your development cycle is critically important.”
For their efforts in being the first airport to roll out a digital concierge, DFW earned an Impact Award from Innovation Leader, and Brinkerhoff recognizes the importance of discovering the work Soul Machines, and IBM had put into the digital persona product they used to achieve it.
A perennial in the AI market race, IBM held 7,343 active patent families in the AI field at the end of 2021, according to Statista, and Romelia Flores holds 84 US patents for IBM. Flores collaborates and co-creates with IBM clients utilizing leading-edge AI, hybrid cloud, and data technology to produce agile solutions. Over the past three years, she has led AI co-creation projects with clients in government, healthcare, banking, and automotive industries.
Flores also worked on the team that brought Iris to life at DFW.
“We are leveraging not just the avatar, but underneath the covers, our Watson Assistant technology, our Watson Discovery technology, to go out and find information,” Flores says. That information could include flight arrival and departure details or airport facilities information. “We pulled all that data, and every time we do that, we were training Iris on something new so that she can then interact with clients on all these things.”
The use of Watson as an underlying technology for Iris is, but one of the exponential AI use examples that Flores lays out: to identify potential credit card applicants as they use a banking app, to monitor system performances to increase loads during peak-use times, to identify imperfections in assembly line products during inspection, to manage electric fleet vehicles, and the list goes on.
“It would surprise me if it’s not being used in just about every industry and every market,” she says. “I’m thinking through all the markets and industries that IBM deals with, and AI is a big play in all of our markets and industries.”
New Opportunities
The emergence of widespread adoption of AI technology, Martin says, is a combination of environmental facets. “We not only have specialized teams in universities and companies doing the research, we have kids in a garage inventing some really cool thing that nobody thought of. It’s a mix of not only the technology, how it’s evolving, but also how connected we have become.”
And employers are looking for those individuals who are rising in this burgeoning field. According to the AI Index report, AI-related job postings rose from 1.7 percent in 2021 to 1.9 percent in 2022.
The first US university to rise to meet the demand in the AI field is the University of Texas at Austin. In late January, it announced a first-of-its-kind AI masters program with a price tag right at $10,000.
In a news release, the university said the Master of Science in Artificial Intelligence will “equip students for an array of potential career opportunities, from engineering to research and development, and product management to consulting.” It will also include formal AI ethics training developed by UT’s Good Systems project, “a research initiative to develop new AI technologies around core values that serve the greater societal good.”
While the AI race has been in progress for the last 70-plus years, the recent uptick of activity and capability is a product of ongoing research, thought leadership, and advances. In addition to the technology itself, the future looks to hold a robust workforce segment that can guide it forward.
