This week we are diving into the world of artificial intelligence. The term artificial intelligence has become a catch-all term used to describe a variety of related but distinct concepts. AI, machine learning, and generative AI are often lumped together as if they all relate to the same thing, or as if there is a single technology that is in fact AI. These related but distinct technologies differ in scope, capabilities, techniques used, and also in terms of reliability and trustworthiness. But generally speaking, artificial intelligence focuses on techniques and applications that create the appearance of human cognition in or from machines. Machine learning focuses on developing algorithms and statistical models that allow computers to perform tasks without being explicitly programmed for those tasks. And generative AI focuses on creating new data, such as images, music, or writing, that mimics patterns presented in the training data.
Much of what we discuss here addresses the technical problems facing the industry and innovators in this space. Specifically, we discuss the various practical manifestations of artificial intelligence, ML, and generative AI that may exist as of 2024 and into the foreseeable future. We attempt to separate fact from fiction regarding what these technologies can do today, what they are good and bad at, and what the industry can expect going forward.
We will also discuss the thorny issue of hallucinations, the implications of artificial intelligence learning from its own creations, the steps being taken to ensure reliability and accuracy, and how to verify that AI is safe, secure, and trustworthy.
We'll start with Jason Alan Snyder, Global Chief Technology Officer at Momentum Worldwide, part of the Interpublic Group. Jason is a futurist, technologist, inventor, author, and lecturer on the topics of the metaverse, artificial intelligence, machine learning, and cognitive robotic process automation. A long-time friend, Jason is recognized as one of the top 30 experts in the world on the topics of the metaverse, virtual reality, augmented reality, and blockchain.
When asked for his preliminary thoughts on the current state of AI technology, Snyder explained that AI certainly does not have agency and will not take over the world tomorrow. This led me to ask a further question: does saying that AI will not take over the world tomorrow mean that AI will one day: actually Snyder said that AI could take over the world, to which he replied, “Of course, that's possible.” He also said that AI has no agency and doesn't have the same shared context as humans, which he called “intersubjectivity.”
“There's no context. I have no idea what Adolf Hitler means. I have no idea what Elon Musk means,” Snyder explained.That makes sense to us, but it doesn't make sense to the platform. And we forget that every time we sit in front of the platform, because it feels like the platform has agency. The platform has no agency.”
We will also hear from Dr. Marek Ben Salem, Managing Director and Data & AI Leader at Accenture Security, who focuses on security for emerging technologies such as quantum, blockchain, AI and the metaverse. Dr. Ben Salem explained how memory limitations can cause AI to malfunction, exposing sensitive information and technology.
“This is called a divergence attack,” says Ben Salem. “As you continue to interact with the AI, it loses context and has to switch to a completely different context and start generating information. In doing so, there's a risk of leaking sensitive information. So when you're interacting with an AI, of course there's a memory limit. At some point you hit a memory limit and it loses track of the context because it hits a memory limit. You can increase the memory limit, but as you know, at some point you hit a limit in everything. So I think there will always be ways to attack AI.”
To hear the entire conversation, listen wherever you get your podcasts (link here) or visit IPWatchdog Unleashed on Buzzsprout.
