Agentic AI: The Future of Autonomous Decision-Making - Six Five On The Road

Summary

In this episode of "AI and Us," host Dave Nicholson delves into the world of Agentic AI with experts Patty Mays from MIT and John Rose from Dell Technologies. Agentic AI aims to digitize and automate skills, creating autonomous systems that can execute tasks with minimal human intervention. The discussion highlights the evolving role of AI agents in industries, emphasizing the importance of human collaboration and oversight. Despite the potential for increased efficiency, concerns about trust and the pace of technological adoption remain prevalent, urging a cautious yet optimistic approach as industries explore the possibilities of agentic AI applications.

Highlights

• Agentic AI is the new kid on the block, enabling autonomous decisions beyond basic tool usage. 🎉
• Experts argue for a blend of AI autonomy and human oversight to ensure things run smoothly. ✨
• The chat explores industries that benefit from Agentic AI, weighing their readiness to adopt it. 🏭
• Concerns linger around AI's trustworthiness and the speed of advancing tech. ⏩
• Starting small and building trust gradually is key to integrating AI successfully. 🗝️

Key Takeaways

• Agentic AI is all about letting AI take the wheel with tasks, offering autonomy beyond simple tool use. 🤖
• It's not just a tool—Agentic AI acts like a virtual assistant that can handle tasks independently. 🚀
• Human collaboration remains crucial as AI steps up, ensuring trust and effective implementation. 🤝
• Industries are exploring how much trust they can place in AI, especially where there's little regulatory control. ⚖️
• The conversation emphasizes balancing autonomy and control to avoid tech overload. ⚙️

Overview

Welcome to the future of AI where Agentic AI is taking center stage. Imagine AI not just as a tool, but as a digital assistant handling tasks autonomously. Experts Patty Mays and John Rose dive deep into how this shift towards autonomous AI is reshaping industries and interactions. But like all 'new kids on the block,' Agentic AI is not without its challenges, particularly around trust and control.

What's fascinating is the synergy needed between humans and AI. No, it's not about handing over the reins completely. It's about creating a dynamic partnership where AI can take care of mundane tasks, while humans maintain the oversight. John and Patty discuss how different industries are tiptoeing on this AI balance beam—some have the freedom to leap, while others must tread carefully due to regulatory shackles.

The big question remains: how fast should we move towards this agentic journey? The dialogue doesn't shriek 'too fast, too soon,' but rather suggests a thoughtful, measured pace. By starting small, industries can not only assess the benefits but gradually build the trust so critical for AI-human harmony. This episode serves as a call to action, urging stakeholders to position themselves wisely in the relentless march of technological evolution.

Chapters

• 00:00 - 00:30: Music Intro and Episode Introduction The chapter introduces the podcast series 'AI and Us', which explores the future in the age of artificial intelligence. The episode focuses on Agentic AI, highlighting it as a new frontier in AI innovation and aims to clarify this cutting-edge topic for the listeners.
• 00:30 - 01:00: Introduction of Guest Experts This chapter introduces guest experts who discuss how Aentic AI is expected to redefine industries in the future. Dave Nicholson from the Futurum Group hosts the discussion, featuring two prominent figures from Dell Technologies and MIT. Patty Mays, an award-winning author, scientist, and professor in the media arts and sciences program at MIT, is one of the key guests. She founded and directed the MIT Media Labs Fluid Interfaces Group and is known for her contributions to AI and media arts.
• 01:00 - 05:30: Understanding Agentic AI and Its Applications The chapter introduces the topic of agentic AI and its applications, featuring discussions with experts including a professional affiliated with the MIT AI lab, who specializes in human-computer interaction, intelligent interfaces, and ubiquitous computing. Additionally, John Rose, the global CTO and chief AI officer at Dell Technologies, joins the conversation to delve deeper into the subject. The discussion promises intriguing insights into the definition and implications of agentic AI.
• 05:30 - 09:00: The Role of Human-Agent Collaboration The chapter titled "The Role of Human-Agent Collaboration" discusses the concept of 'agentic,' which involves using artificial intelligence to digitize skills. This goes beyond merely creating a tool for humans; it involves transferring work, reasoning, and behaviors into AI systems. By achieving this, it leads to significant implications for collaboration and integration between humans and AI.
• 09:00 - 15:00: Industry Impact and Adoption Challenges The chapter titled 'Industry Impact and Adoption Challenges' introduces the concept of an 'agent' as a software system, often incorporating a large language model, communication tools, a knowledge graph, or datasets for memory and reasoning. The emphasis is on the agent's behaviors, enabling autonomy and task delegation. This exploration is likely set against the backdrop of industry impact and challenges faced in adopting such technologies.
• 15:00 - 25:30: The Future of Agentic Technologies and Human Interaction The chapter discusses the transformative role of agentic technologies in reshaping the nature of human interaction and task management. These technologies are designed to understand and execute tasks based on human intentions, allowing humans to express intent and let the technology carry out the work. This shift challenges the traditional model where humans were required to make all decisions, moving instead to a model where humans are involved in oversight, while the technology performs the tasks.
• 25:30 - 35:00: Concerns and Cautious Approaches to Agentic AI The chapter titled 'Concerns and Cautious Approaches to Agentic AI' explains the emergence of new AI capabilities referred to as agentic AI. These AIs possess the ability to perform 'function serving,' which allows them to interact with various tools to gather new data or influence the real world. In addition to data processing, they have perceptual capabilities that enable them to observe their surroundings through sensors. While other AI tools provide valuable functions, agentic AI stands out by its ability to interact with and adapt to the environment, highlighting the need for careful consideration in its application.
• 35:00 - 41:00: Guidance on Adopting Agentic AI in Business The chapter discusses the concept of integrating agentic AI into business practices. It begins with the importance of viewing AI as part of the business toolset, emphasizing the interaction and collaboration between humans and AI agents. An expert, Patty, shares insights based on her 30 years of experience, on how this collaborative effort has evolved. She suggests that there is a need to rethink how we interact with our AI systems, indicating a shift in paradigms towards a more engaging and useful interface between humans and AI advisers.
• 41:00 - 43:00: Conclusion and Future Outlook In the conclusion and future outlook chapter, the text highlights the increasing responsibilities and busyness of both work and life. It notes that as we have to juggle multiple tasks and manage a plethora of information, autonomous agents are presented as a solution to help handle these demands on our behalf. The chapter emphasizes the importance of these agents in efficiently managing both professional and personal aspects of our lives.

Agentic AI: The Future of Autonomous Decision-Making - Six Five On The Road Transcription

• 00:00 - 00:30 [Music] Welcome to AI and Us, a series where we explore our future together in the age of artificial intelligence. In this episode, we'll be exploring Agentic AI, which represents the new frontier in AI innovation. It's a new technology to many of us. So our goal today is to have a conversation that'll bring clarity to this cutting edge topic, offering
• 00:30 - 01:00 tangible insights into how Aentic AI will redefine industries and evolve in the years ahead. I'm Dave Nicholson with the Futurum Group and I'm joined by two visionaries from Dell Technologies and MIT, both of whom stand at the leading edge of AI. First we have Patty Mays. She's an award-winning author, scientist, and professor in the media arts and sciences program at MIT. She founded and directed the MIT Media Labs Fluid Interfaces Group, and works
• 01:00 - 01:30 regularly with the MIT artificial intelligence lab. Her areas of expertise are human computer interaction, intelligent interfaces, and ubiquitous computing, which I'm fascinated to hear more about. And of course joining us once again on this podcast is John Rose, global chief technology officer and chief AI officer at Dell Technologies. Good to see you again, John. Good to see you, Patty. John, let's let's get straight to it. What is agentic AI? And
• 01:30 - 02:00 I guess more importantly, what isn't it? The world is a combination of digitized information, but it's also a world of skills. And agentic is the approach of using artificial intelligence to actually digitize those skills to not just create a tool for people to use but to actually transfer the actual work the reasoning the behaviors into this machine layer that is now an AI system. Uh now the result of that if you do both of those you you've made a profound
• 02:00 - 02:30 impact. But what is an agent? An agent is the tool to do that second thing. And what it is is just a software system. It probably has a large language model. that provides some communication and other tools to it. It might have a knowledge graph or some data set that it uses to kind of emulate memory and understanding and reasoning. But more importantly, it's really the behaviors of the agent that matters. It is not just a tool for a human to use. It's actually something that you can give autonomy to. You can delegate a task to
• 02:30 - 03:00 it. You can give it an intent. It can figure out how to do that work. It can ultimately accomplish that work. It can interact with you as it does that work if it has a question and ultimately that system introduces an environment where the human is no longer in the loop exclusively meaning making all the decisions. The human is on the loop. It is expressing an intent and letting the technology do the work even if that work is narrow or that work can be very broad. Um more importantly agents tend to have a couple of characteristics that
• 03:00 - 03:30 are new. They have the ability to do what's called function serving which means they can reach out and interact with tools. So, not only do they process data, but they can be told to interact with another tool to get new data or invoke a behavior in the real world. They're also perceptive. Generally, they're able to observe the real world. You can give them access to sensors. And so, they can sit there idally. If your system is self-contained and cannot interact with the world around it in any way, it's probably not an agent. It's probably something else. Now, all those other tools are useful and good, but agents are the new capability. And
• 03:30 - 04:00 imagine, you know, what we can do with this capability as we think about it in part of our tool tool chest. Patty, John mentioned this idea of the kind of interactions and collaboration that happen between humans and agents. That's your area of expertise. What does that look like? Where are we now in this collaborative effort between humans and agents? Yeah. So I've been arguing for uh over 30 years actually that we should change the way we interact with our devices and uh give our advisers a
• 04:00 - 04:30 little bit more autonomy to take care of tasks on our behalf. work and life are ever getting more uh busy and demanding. And whether it is in a work context or uh in our uh private lives, we all sort of need to multitask and take care of more and more uh information and process it and keep track of a lot of things. And so agents are a way to do that. they
• 04:30 - 05:00 can help us with managing the overload of information that we want to keep an eye on and uh the multitude of tasks that we are sort of uh uh dealing with in parallel. So I think uh that increasingly again in in private lives and in work life people will be managing uh a collection of agents that help them with all of these tasks. Um now I would say that uh these agents especially in
• 05:00 - 05:30 the beginning are not likely to fully automate tasks um because you want to make sure that the agents actually uh do things the way you really want them to be done. AI still makes mistakes. We all know that LLMs which are sort of the back end for today's agents um still can hallucinate and don't quite understand the world or or uh still make reasoning mistakes. So it is important in my
• 05:30 - 06:00 opinion that humans are still in the loop and that these agents um sort of either are audited and report back on their behavior and that people check that especially uh initially when they adopt uh agents into their work or or private life processes uh to make sure that that things all go smoothly. So it sounds like there's a sort of a continuum of of of a of a journey towards autonomous behavior here. Um so
• 06:00 - 06:30 definitely a collaboration between humans and agents. John, are there specific sectors of the economy or industries that you see um having specific challenges with this? And then conversely uh maybe which are more well suited for this this type of collaboration. And Patty, I want to get your feedback after after gone. The analogy I give will they will have giant training wheels on them. They will be
• 06:30 - 07:00 constrained in a way where only the things that we're comfortable them doing are they're allowed to do autonomously. Now the nice thing is because the technology foundationally is quite strong. It's just a question of its behavior and its data that we got to get comfortable with. We can relax those constraints over time. You know, your first project might only allow it to do a very very specific thing with some degree of autonomy and everything else you're involved. And once you get comfortable with that, you might give it a little bit more capability and you're just relaxing the constraints, giving it
• 07:00 - 07:30 a bit more data, changing the prompt. Those are things that are very likely to be the pattern of deployment regardless of industry. Now, if we get into industries, what we discover is the risk profile of industries matter. For instance, in some cases, we will see an industry that is heavily regulated and that industry will fundamentally require uh a much slower graduation of autonomy in places like software development, funny enough, which is a very important space, but there's not a lot of regulation on it. We're seeing much more aggressive use of agents as long as we can predict the behavior and the
• 07:30 - 08:00 outcome. So I I think uh there isn't an industry that won't be impacted but the the path depending on what industry you're in will be an industry in terms of how much risk is resultant from transferring trust granting agency to a technology to do things on your on your part. By the way to date we have found no industries where there isn't a starting point where there isn't something that they're incredibly comfortable handing to an AI to do on their behalf if it's constrained if it's got the guardrails the training wheels and they can actually trust it. And that
• 08:00 - 08:30 once you get down that path, relaxing the constraints is far easier than building the first agent. Well, Patty, do you have a different view based on the human equation? Yeah. Well, I do think that there's a huge role for agents definitely. uh but we have to thread carefully and um I think uh starting with tasks that are well defined constrained um and like John said giving the agent uh gradually more autonomy uh is really
• 08:30 - 09:00 the way to go um but I think ultimately people always still want to be in charge of their agents and uh know what they are doing uh know what they have been doing so get reports reports uh audit them regularly. We don't want to lose complete uh control sort of of our processes uh and delegate them to agents without knowing what these agents exactly um are doing. So I am very
• 09:00 - 09:30 interested in that problem sort of of human agent interaction and uh making sure that we can trust them and that we still ultimately um the agency or the control still lies uh with the person. You know Patty it's a great point because um you know we think about it's it's it's more complex than you interacting with a bunch of agents. The agents interact with each other. There's almost no scenario where AI operates in
• 09:30 - 10:00 a complete vacuum. In the agentic world, you may find that you are ultimately the instigator of whatever you want to happen. It's a great example would be if you had a concierge agent that was very personalized, was running on your local device. You completely trusted it. It understood your your priorities and then you delegated and said, "Your job is to book my vacation." Well, it may be the thing talking to the travel agency that has an agent and the airline that has an agent, different roles, but to your point, the ultimate measure of success is does the human being trust that it is
• 10:00 - 10:30 doing the right thing in the right way on their behalf. If that is not true, this technology will not be successful. If it is, it will be wildly successful. What are some of the emerging technologies that are going to uh that that we're all going to have to keep up with as humans that are going on right now? The biggest one that we're really excited about is how do you make these things work with each other. You know, you cannot do that by just hoping they work. You have to develop interworking standards. And today we we have a lot of work around things like agent to agent
• 10:30 - 11:00 and MCP and agency. And there's a bunch of initiatives that are all coming together at MIT that really exciting about kind of dealing with registries. So we're going to develop interworking protocols, orchestration models for these systems. And all of those things are necessary. The exciting thing is most of them didn't even exist six months ago and now they're at a point that we can actually code to them. We can implement them and we can play with them. And so that tells us if you ask me what the important technologies for agents are today, I can kind of answer. If you ask me in two years, I have no
• 11:00 - 11:30 idea. This is a fastmoving space. We're going to invent all kinds of things, but it's going to keep moving. What about this December, Patty? Where are we going to be in this agentic journey by the end of this year? And remember this is being recorded. So yeah, uh personally I think we'll see uh early experiments uh in businesses uh especially with uh tasks or domains where maybe the cost of a mistake
• 11:30 - 12:00 happening here or there is not too high. uh so I think uh businesses will have to be careful about uh what kinds of uh tasks and processes they first sort of try to uh automate or delegate uh to agents. I think in the um sort of space of human computer interaction, I think increasingly we will be talking to an agent um that is sort of like our
• 12:00 - 12:30 concierge agent as John called it. uh that knows us really well, knows what we care about, how we like things done, and that can really sort of on our behalf keep track of a lot of things and automate a lot of things and answer questions that are easy to answer, etc. So that we have an easier time sort of uh staying on top of our uh say huge inboxes and busy calendars and lives. uh
• 12:30 - 13:00 and and all of the things sort of that we're trying to fit uh into our lives. Yeah. John, John, when you tell us, you know, where where you think we'll be by the end of the year? Um also, maybe you could help us understand where we should start. A lot of folks are asking the question, how do I either save money or make money with AI, let alone just agents, but but so help us understand where are we going to be by the end of the year? And and where where should
• 13:00 - 13:30 people be starting today? How do you get started? And what my prediction is by the end of this year, the first places where you will see Agentic materialize and it is the first place where it should materialize. So if you're asking where you should start is don't pick an entirely new AI workflow. It turns out that if you started to tackle the process of finding customers and selling to them like we did, you found that your first problem was, hey, in that process, maybe I can make content management better by using generative AI to create
• 13:30 - 14:00 one source of truth and one interface. Great. That's perfect. But now you already have a process. And so say ask yourself, well, if I have agents, what would I do to make that even better? And we realized that boy, you know, as you're going through that process, there's moments in time where you have to leave the process and go to like a financial system and get a price or go to an inventory system and get an answer. Now, you could hardcode that in and do it the old way. Perfect agentic use cases. An agent whose only job is when the salesperson needs to get a
• 14:00 - 14:30 price, it does it for them. That is a perfect agentic use case and it complements the path of getting from finding a customer to selling to a customer. And so same thing with software development. Don't throw out your coding assistants and decide to go to agents orthogonally. Look at the things within your coding process where there's a bit too much human intervention. It's characterizable. You could incorporate an agent. So by the end of this year, I think the manifestation of agents in the real world is not in an entirely new domain.
• 14:30 - 15:00 It's complementaryary to the first generation AI workflows that we've actually already put into place. Patty, from from the kind of that back to that human technology partnership, do you see the starting place for individuals being a little a little different? I'm thinking of using agents to do things for me personally or as an individual contributor to the business making me more efficient. Where would you recommend people start that journey? Uh the personal journey. Um I
• 15:00 - 15:30 Yeah. So I think again increasingly I mean especially the young generation that I uh work with at MIT the younger students they use uh today's chat bots um all the time like at least two hours a day for solving or all sorts of uh problems or questions uh answering all sorts of questions that they have and I think that especially that generation will also be very quick in adopting ing
• 15:30 - 16:00 um agents systems that go be above uh just providing information but actually automate some simple things on their behalf. Whether it is um uh answering to someone uh are you available at this time on that date or uh simple things like that that where the cost of a mistake isn't too high. Uh maybe it is ordering pizza from your favorite pizza place etc. But I think they'll still be
• 16:00 - 16:30 fairly constrained in terms of what kinds of things uh they are allowed to do. So baby steps, but the baby is stepping at 60 miles an hour is sort of what I'm hearing. It makes it makes sense to have approach, but it's all going to happen quickly. Well, by by the way, even in industry and companies, we see the same pattern. are younger in career people are more likely to embrace a different way of working, a different
• 16:30 - 17:00 tool. And so it doesn't mean that, you know, those of us been doing this a long time aren't open to it. It's just if you're not stuck stuck in your ways and somebody gives you a tool that can actually change the way you work in a really positive way, you're likely going to lean into it. Uh but but our data does show that even the people that are most resistant as you roll these technologies out, if they work, within six months, everybody's using it. So it's just a question who goes first. It's not a question of it doesn't get adopted. Patty, I'm I'm fascinated by something. Uh you've you've been at this a long time. We think, you know, we're
• 17:00 - 17:30 talking about Agentic AI, which you could argue is sort of the leading edge of what we're doing with AI today. Um but you've been involved in this study of the thoughts around artificial intelligence for literally decades, decades, and decades. What does that look like? Have have the things that you thought about 30 years ago, dare I say, uh how many of those things have come true? What sort of surprises have you seen? Walk us through that. Yeah. So, um
• 17:30 - 18:00 after getting a PhD in uh AI, I actually uh shifted my focus towards what I call IIA or intelligence augmentation. And I sort of had a um sort of realization that I didn't really want to make computers and robots smarter, but I wanted people to uh become smarter and more capable etc. And that's actually uh why I started talking about software agents in
• 18:00 - 18:30 1994 and uh arguing that we should change the way we interact with our computers so they actually do much more of the uh work on our behalf and help us manage our busy lives. Of course back then we were prototyping some of these agents but we had to carefully craft them by hand. And of course what has happened now is um that agents can rely on these foundation models, these large language models that know a lot about
• 18:30 - 19:00 every task out there and uh what the different steps are that are involved if you want to uh get pizza on your uh um kitchen table tonight, etc. Um so we don't we no longer have to build these systems by hand. we can rely on LLMs and that is what has opened up this whole discussion again I believe around agents have you been surprised by the pace of
• 19:00 - 19:30 change either way has it often when we're really close to a subject we assume that things will happen at an accelerated rate but in this case over over the decades what does that look like are we are we lagging behind your vision well it took 30 years but uh definitely I barely can keep the last couple of years in terms of how fast the world of AI is um moving and most likely that's because there's now so much investment uh that goes into it
• 19:30 - 20:00 and a lot of the um uh development is happening in industry of course it used to be largely the domain of uh academia before that and uh um that was a little bit more of a manageable pace I think so what I worry about these days is that maybe we're going a little bit too fast sometimes. We're a little bit too eager maybe to uh um adopt these technologies without sort of truly understanding how
• 20:00 - 20:30 they work or what their limitations are um how what happens when we have all these agents talking to each other where the agents are not 100% say trustworthy etc. Um, so that's definitely one of the things that uh I'm a little bit concerned about. Yeah. Do you are you more concerned today than maybe you would have been 10 years ago? I am definitely. I mean uh for the last couple of years um the philosophy just
• 20:30 - 21:00 has been like if we can build it, we build it and then we throw it out there and encourage everybody to use it. And we're sort of um uh conducting a um experiment in the real world as opposed to letting researchers sort of conduct smaller scale experiments uh to understand what uh some possible problems may be. That's also why I recommend that businesses for example really think carefully about what processes in their business um uh are
• 21:00 - 21:30 not too critical and are very repetitive in nature. etc. so that um uh they are maybe a better uh sort of opportunity to deploy agents that are carefully constrained and so on uh so that nothing uh can go wrong. So John, you you talk to CIOS and CTO's probably every day and you you and I both know that the big question is um you know it's not it's
• 21:30 - 22:00 not just where do I start, but it's how do I start? What do I do? How do you begin to approach that? Are people even asking the right question? Where I can imagine within Dell, your kind of uh professional services and solutions engineering folks are really being pushed to the forefront because so much of what you just described sounds like that mapping is absolutely critical. Yeah, that's a great point. In fact, our our PS organizations and our business
• 22:00 - 22:30 transformation folks spend way more of their time on the nontechnical aspects of AI. It's helping customers figure out what process where to apply this, how to safely apply it. You still have to do technical work that you can't miss that I is a technology and ultimately has to be implemented. But the order of operations matters and I I you know I will tell you it's insufficient to just do the consulting work because it has to ultimately get built and implemented. But it's also unwise to do the technical work without knowing what problem you're
• 22:30 - 23:00 solving and how that problem should be best solved. So, you know, I think that this is just another example of when you do modern technology, it's always an ecosystem. It's always multi-dimensional. It always has a human component to it. This is just like every other technology at this time. It just happens to be faster, more impactful, and more interesting. So, do they need to understand any of the tech? When we are evaluating or creating a path to do agentic, there is always a technologist in the room. This technology is not unbounded. It is not infinitely capable.
• 23:00 - 23:30 It has risk and if you're developing even if you're picking the process, you should do it with a conscious understanding that there is a technology implementing this and it has boundaries. We view that you cannot hand off a purely intellectual concept into a technology ecosystem without ever connecting those two. And the best way to avoid it is have the technologists and the business people at the table at the same time for the entire process. It will just work better. Well, it's interesting. It feels like uh a lot of
• 23:30 - 24:00 this keeps going back to uh Patty's area of of domain expertise which is this intersection between technology and people because you're describing it kind of from from a different direction but you know it there is no such thing as just business people figuring this out. There is no such thing as just technologists figuring this out. Would you agree? incredibly likely that within the next two years, whatever you do today, whatever your work is defined as,
• 24:00 - 24:30 the amount of effort that you have to expend to do it changes profoundly. And a huge portion of that work that quite frankly could be automated, could be delegated, but can't be because there's no one to delegate it to, suddenly is. And you fundamentally scale. Increasingly, our entire lives will be mediated by AI. The way we work, the way we learn, the way we take care of our health. We will have personal AI systems
• 24:30 - 25:00 that know about our goals, that know what we care about, and that help us by automating a lot of the actions and the tasks on our behalf and keeping track of a lot of things on our behalf. Yeah, absolutely. And I think all of us here will be embracing the agentic revolution as we move forward. John and Patty, this has been a delight. As always, this is a conversation that could go on for many, many hours. Uh we thank everyone for
• 25:00 - 25:30 joining us in this conversation. This is a conversation about artificial intelligence. And when it really comes down to it, this is all about AI and us. I'm Dave Nicholson for the Futurum Group. Thanks for joining us. Stay tuned for further exploration into this subject in the future.
• 25:30 - 26:00 [Music]