My name is Alison Smith, and I'm a director of Generative AI at Booz Allen. We have some clients asking us, what is it? You know, they want to understand what generative AI is and how it's different from some of its preceding technologies, right? But then you have other clients who are kind of ready and wanting to know what other organizations are already doing. And then finally, you have this other subset of clients, typically ones that are a little bit more advanced, who've done some pretty large kind of deployments of large language models where they might ask, you know, how is this going to change the platform that I already have? You know, what tweaks do I need to do to actually use it? And then lastly, all of them are always asking, you know, all my leadership is super excited about generative AI. How is it different than traditional AI and where does it actually make sense? Because I hear it's very expensive. Generative AI introduces a lot of new challenges that traditional AI didn't really have. The first one that really comes top of mind is interpretability. So the outputs of the generative AI system can be different every time, even if you ask it the same question. And so those outputs, you know, aren't always interpretable. In fact, it's really hard to explain why a generative AI system produced or created the sequence of words that it did. And in use cases where it's really important to explain why a decision was made or why, you know, a report looks the way it does, it's going to be really hard for generative AI system to explain that or for even a human to try to explain it. There are a few other challenges that include the bias in data. Generative AI systems have been trained on a whole lot of data, and you can't even imagine what types of biases are inherent to that data. And that's not something you can just pull out from a foundation model. And so it would be important to think about the guardrails that you need in place and having a human in the loop to make sure that we're not perpetuating especially some harmful biases. Another one would be thinking about the security. So generative AI systems are very expensive to build, and you have these tech companies that are pouring in millions and billions of dollars into building these foundation models. And so you can't expect your average organization to build one for themselves. And so they're going to be using other people's tools and foundation models. And so we really need to think about how to keep that sensitive or proprietary data that these firms have secure when they're using these external tools. And then lastly, we have to think about IP rights. All of that still remains very uncertain. A lot of the training data, some might argue, included some copyrighted or even patented information. And so it's hard to say whether it's okay for a foundation model to have that information. And if it accidentally produces output that was partially copyrighted, then, you know, is it okay to use it? And then lastly, the outputs that it does create, if that's going to be some form of intellectual property for your enterprise or for any organization, you have to think about where policy is going on that and seeing whether you know, you can actually have rights to something that was created not by a human or your organization, but a generative AI system. The government holds itself to a much higher standard, than you have to necessarily for some companies, right? You have companies that can use generative AI for recommendation engines. In terms of you know, what movie to watch next or what clothing to buy. And those are relatively lower risk. And when you're working with the federal government who have missions that really impact people's entire lives, they have to maintain a standard that is much higher in terms of security and accuracy. And so while the government may be slower to act on generative AI, it's going to be paving the way in terms of how it's thinking about policy for AI, how to maintain security and how to think about it from a very responsible and ethical framework. And I think those are all areas where you're going to see the government innovate potentially even faster than industry because they have to. It's sort of limitless at this point, and it's really exciting to be at the precipice of something totally new that's going to create a huge shift in how we do work.
