The Era of Generative AI
/
Kate Pospisil, CFE
ACFE Communications Specialist
To echo the sentiments of Joe Rogan and Steve Jobs, Nina Schick’s session at the 2023 ACFE Fraud Conference Europe was “really, really scary and really, really cool.” Schick, author, entrepreneur and widely recognized as one of the first generative AI experts, enthralled her virtual audience during her discussion on the history, growth and outlook of the newest form of artificial intelligence, generative AI. A concept that is less than 10 years old, generative AI has completely moved beyond artificial intelligence as we’ve known it before — it can create data that assists in production of things we’ve historically only associated with human intelligence, such as text, music, images and video.
Generative AI has a few features that set it apart from previous iterations of artificial intelligence:
It can create perfect content.
It can create in any digital medium — images, video, text, audio, etc.
It can be automated.
It is scalable on a level we’ve never seen before.
It is constantly improving at a rapid rate.
Where It Started
Generative AI first came on the scene in 2017 — less than six years ago for those counting — with deepfakes that were generally used to create memes and other funny stuff. (Here’s looking at you, Nicholas Cage Yoda.) The next year, StyleGAN was created, which was trained to create fake photographs that look indistinguishable from human faces. In 2021, we started seeing the advent of foundational models, which were trained on huge data sets, were not task specific, could be prompted with text and requiring minimal fine tuning for a finished product. Stable diffusion, an example of a foundational model, wasn’t the first available, but it was the first open model with millions of people using it within weeks of its release. This product kicked off a generative AI “arms race” of sorts and was used initially for viral TikTok videos of people — namely celebrities — doing and saying strange things. These models are only getting better.
Another type of foundational model is the large language model, such as ChatGPT. As Schick says, “Everyone’s got a ChatGPT story.” ChatGPT reached one million users in five days and 100 million users in two months, which was the fastest adoption of new technology in history. By comparison, the next highest rate was TikTok, which took 10 months to reach 100 million users. This interest in large language models, provable by the usage rates, prompted the big tech companies to stake their own claims in the AI space. Companies like Google, Microsoft, NVIDIA and more are all working to perfect and promote their own version of a large language model. With this in mind, digital content has become the most important medium of human communication, and generative AI means that humans aren’t just consumers of digital content, but also creators.
Where We Are
According to Schick, we are entering a new paradigm where generative AI forms a foundational layer of human communication and of the digital ecosystem. A single creator can now create content that rivals the best Hollywood productions, there is widespread and open availability and businesses are quickly using generative AI to create and sell services that never existed before. There are a wide range of companies that didn’t even exist a matter of months ago that are using generative AI to provide services such as text and code creation, image editing, design, audio and video creation, music creation outside of the human element and more. Machines can now do things we thought only humans could do just by having the right data set. The future of data and privacy in the era of generative AI is one in which AI can be trained to clone people with perfect audio and visual likeness using only a 20-second video, and this clone can be made to do and say anything the creator wants.
Technology is accelerating exponentially, and so is its adoption. For example, it took 50 years for 25% of the U.S. population to have access to electricity, it took less than 5 years for smartphones, and will take less time than that for generative AI to become widespread. Billions of people around the world are interfacing with AI already, shown through the use of ChatGPT alone. Adoption of generative AI technology will be faster than any technology we’ve seen before because it is widely available, and you don’t need specific infrastructure or devices to use it.
The Risks of Generative AI
Schick cited “Rule #34” of internet usage, which is, “If something exists online, then it exists as pornography, no exceptions.” In fact, she went on to say, deepfakes first emerged as AI-generated, non-consensual pornography. All the creators needed was some training data which could be obtained from a victim’s social media presence. So that’s an example of a very serious (and markedly gendered) risk, but we also face huge implications for fraud and cybersecurity as AI technology improves. In 2021, the Federal Bureau of Investigation (FBI) released an alert regarding the risks: “Synthetic content will be leveraged in cyber and foreign influence operations in the next 12-18 months.” Some of the risks are highly believable spear phishing messages and sophisticated social-engineering attacks, especially as the technology can now be used in real time. The personal and political implications of this technology cannot be overstated.
Synthetic identity theft is the fastest growing fraud risk with the real-time capabilities of these programs to use training data to create new, fake identities. Voice deepfakes are also a very legitimate risk, with large frauds already having been committed using the technology. Large language models like ChatGPT have safety layers that are easy to break through in spite of the months of research and development that went into their creation. Where do we go when the digital ecosystem becomes so corroded that it’s completely untrustworthy because of the capabilities and prominence of generative AI? Schick cited a perception of moving from “No trust to Zero trust” — nothing is trusted without verification. The digital and AI realm needs to work on authenticity and transparency in hopes of setting standards for the internet — all content should be in-built with context, which will eventually be the only way we can trust media.
In her conclusion, Schick reminded the audience that we have agency in shaping how generative AI plays out in our society and that we are not helpless on the sidelines. It’s absolutely going to change a lot, and fraud fighters will be on the frontlines of the era of generative AI. Tech is changing what it means to be human, and we must always ensure we are keeping up with it from the human perspective. Collaboration between tech companies and the financial institutions, law enforcement, the traditional media and the average population is vital to ensure we are moving forward in a positive way in this new generative AI world.
