Leadership vs. Technician Priorities in IT, AI and Healthcare: Adoption Rates and Future Trends

[00:00:00] It's Wednesday, April 17th, 2024. I'm Dave Sobel. Welcome to the Business of Tech Lounge.

[00:00:18] Today on the show, AI and healthcare with some insight on adoption rates, a dose of the future

[00:00:25] infused with AI as the holodeck arrives, frameworks and policies with John Gillum,

[00:00:31] a preview of this weekend's interview on the nuclear event in cybersecurity and

[00:00:36] your questions and comments. Now I want to thank SalesBuilder, our Patreon sponsor,

[00:00:43] whose support makes this show possible. Focus on your IT sales force workflow with power of

[00:00:49] automation and visit them at salesbuilder.com. That's B-U-I-L-D-R dot com. Want to get your logo

[00:00:57] right here? Vendors, you can do so with our vendor Patreon program. It's a simple monthly

[00:01:03] subscription and visit patreon.com slash MSP radio to sign up. I can't do the show without

[00:01:09] support and thanks again to SalesBuilder for theirs. Now I do take questions and

[00:01:15] comments throughout the show, so make sure to put them in the chat. We'll have a dedicated

[00:01:20] question section in the show later with all those listeners submitted questions,

[00:01:24] but we'll throw those chat comments up anytime. But first, our top story. Now I want to dive

[00:01:32] into Auvik's recent IT trends 2024 industry report. The report identifies a persistent

[00:01:40] shortage of skilled workers as the primary challenge for IT teams in 2024. As a consequence,

[00:01:48] there's an increased investment in IT with budgets expanding to accommodate needs for

[00:01:53] automation and outsourcing. There's a notable increase in investments in automation technologies.

[00:02:00] 24% more IT professionals plan to invest in automation in 2024 compared to 2023.

[00:02:08] Almost 96% of IT professionals, so almost all, are using at least one AI or ML tool to improve

[00:02:17] efficiency. A significant portion of IT teams are outsourcing tasks, particularly network-related

[00:02:24] functions, to manage workforce shortages. Nearly three out of four respondents indicated that

[00:02:30] at least some network-related tasks are being outsourced. Nearly 58% of C-suite executives

[00:02:38] expressed high confidence that their organization's network toolset meets the needs of remote workers,

[00:02:44] whereas only 35% of IT technicians shared this confidence. This disparity suggests that

[00:02:51] upper management may not be fully aware of the operational challenges faced by their IT teams,

[00:02:58] potentially leading to decisions that do not align with the technical realities and needs.

[00:03:04] There's also a notable difference in how customer satisfaction is prioritized across

[00:03:10] organizational levels. About 46% of C-suite executives list customer satisfaction as the

[00:03:16] most important metrics, compared to only 26% of technicians. This suggests that while executives

[00:03:23] focus on customer outcomes as a measure of success, technicians may be more concerned with

[00:03:28] resolving immediate technical issues, perhaps due to being overburdened with tasks.

[00:03:34] The majority at 86% of respondents reported increased budgets in 2024, with nearly 50%

[00:03:41] saying they expect to see an increase of at least 20% from 2023. In terms of specific

[00:03:48] investments, 48% of survey respondents shared that they're investing most heavily in SaaS management

[00:03:54] and monitoring tools, 46% shared their investing in Wi-Fi management, and another 46% shared their

[00:04:00] investing in cloud monitoring and in management. So why do we care? It's this disparity. It's

[00:04:10] always the disparity. When you look at the differences between leadership and those doing

[00:04:15] the work, that misalignment causes the most friction. And in fact, I think this is the critical

[00:04:21] area to focus upon, particularly when you look at the difference between priorities of C-suite

[00:04:26] executives focusing on customer sat and technicians not doing the same or that disconnect

[00:04:33] between focusing on business outcomes where their technicians are not doing the same.

[00:04:38] And in fact, it's a very similar gap. We're seeing roughly 20% difference between the two.

[00:04:43] And in that 20%, that's the key opportunity. Those that are able to execute here and outperform

[00:04:50] are the ones that are going to do far better. It's why we care. I don't want you to focus

[00:04:55] so much on the tool sets or where they're necessarily investing. I want you to focus

[00:05:00] on the cultural gaps between leadership and those that are doing work. And it's important

[00:05:06] for you to make sure internally as an IT service provider and as a managed services

[00:05:10] provider that you're teaching your people the right skills to make sure they can communicate

[00:05:15] effectively with your customer base. That's the takeaway here that I think is most interesting.

[00:05:22] I think there's areas of growth for both providers and for their customers to focus on

[00:05:28] that interdependency. And by the way, I want to note that 96% of IT professionals saying they're

[00:05:34] using at least one AI or ML tool. I'm not surprised by that, but it's how effectively they're

[00:05:40] using it that we're going to want to dig into further. Now, if you got a question or a comment,

[00:05:45] make sure to put it in the chat. If you're watching live, we will take any questions during

[00:05:50] our question section later. Now, you know, I love a good use case. So let's dive into

[00:05:57] tech crunches look at generative AI making its way into healthcare. And not everyone is

[00:06:02] convinced of its readiness. While their investments and startups focused on generative

[00:06:08] AI in healthcare concerns remain about its limitations and efficacy. Studies have shown

[00:06:14] that generative AI struggles with complex medical queries and can produce incorrect

[00:06:18] diagnoses. Additionally, there's worries about perpetuating stereotypes and biases.

[00:06:25] So let's quote directly from the article quote only about half at 53% of US consumers said

[00:06:33] they thought generative AI could improve healthcare, for example by making it more accessible

[00:06:38] or shortening appointment wait times. Fewer than half said they expected generative AI to make

[00:06:43] medical care more affordable. Andrew Bakorsky, chief AI officer at the VA Sunshine Healthcare

[00:06:50] Network, the US Department of Veterans Affairs largest health system doesn't think that the

[00:06:56] cynicism is unwarranted. Borosky warned that generative AI's deployment could be premature

[00:07:02] due to its significant limitations and the concerns about its efficacy. Well, one of the key

[00:07:09] issues with generative AI is its inability to handle complex medical queries or emergency,

[00:07:15] he told tech crunch. It's finite knowledge base that is the absence of up-to-date clinical

[00:07:20] information and the lack of human expertise make it unsuitable for providing comprehensive

[00:07:25] medical advice or treatment recommendations. Several studies suggest there's credence to these

[00:07:31] points. In a paper in the journal Jama Pediatrics, open AI's generative AI chatbot, chat GPT,

[00:07:38] which some healthcare organizations have piloted for limited use cases, was diagnosing pediatric

[00:07:44] diseases 83% of the time. And an open AI's GPT-4 is a diagnostic assistant. Physicians at

[00:07:51] Beth Israel Deaconess Medical Center in Boston observed that the model ranked the wrong

[00:07:56] diagnosis as its top answer nearly two times out of three. And this section, quote, Arun Thurvasascu,

[00:08:04] a clinical research fellow at the University of Oxford, said there's nothing unique about

[00:08:09] generative AI precluding its deployment in healthcare settings, quote, more mundane

[00:08:14] applications of generative AI technology are feasible the short and long term and include

[00:08:19] text correction, automatic documentation of notes and letters and improve search

[00:08:23] features to optimize electronic patient records, he said. There's no reason why

[00:08:27] generative AI technology, if effective, couldn't be deployed these sorts of roles immediately.

[00:08:33] But while generative AI shows promise in specific narrow areas of medicine, experts like

[00:08:39] Borkowski point to the technical and compliance road box that must be overcome before

[00:08:44] generative AI can be used and trusted as an all around assistive healthcare tool.

[00:08:50] Significant privacy and security concerns surround using generative AI in healthcare,

[00:08:54] Borkowski said. The sensitive nature of medical data and the potential for misuse or unauthorized

[00:09:00] access pose severe risks to patient confidentiality and trust in the healthcare system.

[00:09:06] Furthermore, the regulatory and legal landscape surrounding the use of generative AI

[00:09:10] in healthcare is still evolving with questions around liability, data protection,

[00:09:16] and the practice of medicine by non human entities still needing to be solved.

[00:09:21] Another thought leader, bullish as he is around generative AI in healthcare says there needs to

[00:09:26] be rigorous science behind tools that are patient facing. So why do we care? It's the studies,

[00:09:36] it's the thinking strategically around what the right use cases are. I don't necessarily

[00:09:42] think that everything is going to be effective in all cases right away, but there are obvious easy

[00:09:48] places to find those initial investments. Note taking, text correction, automatic documentation,

[00:09:55] summarization of information, obvious first use cases that will get adoption quickly.

[00:10:01] Beyond that, you're going to be looking to find out what's more results are coming

[00:10:06] out of these kinds of studies to help guide your clients through the kinds of investments

[00:10:11] they're going to be making. It's not going to happen overnight, and that's a good thing because

[00:10:16] we want to have thought through those various pieces. And you can use those first rounds of

[00:10:22] implementations to help guide with policy and procedure to make sure that you're not putting

[00:10:27] confidential information into untrustworthy agents or making sure that your people understand

[00:10:33] the correct use. And ultimately, the thing that I'm finding most consistent across all

[00:10:39] of these use cases is the places where humans remain in the loop are the most effective.

[00:10:46] So that's why we care. We continue to look for use cases that are effective and are working out

[00:10:51] well for customers in order to leverage them further in expansive service offerings.

[00:10:56] Now reminder, we continue to take questions and I'll watch that chat. If you've got something

[00:11:01] to say, we do appreciate it. And by the way, throw that list of who liked the videos.

[00:11:06] Thank you very much. It's much appreciated. Now, last weekend, there was an interview that came out

[00:11:12] with John Gillum whose company focuses on AI detection. He joined me for an interview that

[00:11:18] released last weekend. Here's a segment of that interview.

[00:11:22] I think that leads me to then to where my next question is like, how do you

[00:11:26] tell people to best use AI detectors to be effective?

[00:11:30] Yeah. So most of our use case is within the world of marketing. So where we have people that

[00:11:37] are have writers on staff that are creating content, they're happy to pay those writers $100,

[00:11:44] $1000 or an article, whatever it might be. They're not super happy when they find out

[00:11:48] that that content was popping and pasted a chat, GPT in five seconds. That doesn't

[00:11:52] feel like a fair value exchange. And so the way we recommend our tech to be used is

[00:11:58] within the copy editor. So whoever's receiving the content from writers,

[00:12:03] use a tool as a spot check and build up a live back history with each writer to understand,

[00:12:09] okay, this person has always had 0% AI detected. They had one article that was a 50%

[00:12:16] sort of a flip of a coin. The classifier wasn't unclear. And it was back to 0%.

[00:12:21] So probably that's the situation where the classifier or detector got it wrong.

[00:12:26] Let that go. There's another situation where you would see a trend on a writer that used to get 0%

[00:12:32] is now getting 90%, 100%, 100%. Pretty good sign that that writer is now using AI and if that's

[00:12:37] not within your policies, then you should not be using that content.

[00:12:43] Gotcha. So how are you then advising organizations to put their policies together?

[00:12:48] What are the questions that you think are most important to answer?

[00:12:52] Yeah, this comes back to that first question that you asked around understanding

[00:12:55] within the organization where the risk associated with AI having input into content that you might

[00:13:02] not want exists. And so if you have a part of your organization where absolutely no content,

[00:13:08] no AI usage is allowed, some law firms are taking that approach, some medical

[00:13:12] organizations are taking that approach where we want no chance that AI is being used.

[00:13:17] We want whoever's writing this to have 100% responsibility on the words that have been

[00:13:23] used. And in those situations, then the policy will be different than say the marketing team where

[00:13:28] there's a little bit of sort of let time build up, let some data build up where there's a sort

[00:13:34] of a stronger, more binary approach potentially within law firms. So I think it again, it comes

[00:13:40] back to sort of a nuanced approach but making it clear to the editors who are then in charge of

[00:13:47] policing whatever policy gets enforced, but creating a policy that accurately reflects the risks at

[00:13:53] that portion of the organization.

[00:13:57] So ultimately we have to ask why do we care? I love this conversation with John because it was

[00:14:03] the application of practical tools in a way that helped enforce policy and it is a great

[00:14:10] example of human in the loop. Really, if you've missed this, I would encourage you to

[00:14:14] dive in here because he walks through the various ways that the organizations that leverage his

[00:14:19] toolkit use the product to be thoughtful. And one of the things is much more about finding

[00:14:26] trends and patterns than it is around specific implementation. Here what John was talking

[00:14:32] about, he's engaged when you're working with a marketing firm who's hired an organization

[00:14:37] for content management and they want to look at each lawyer or each writer, what they're

[00:14:42] doing over time is they're running it through the detector and they're getting a sense of how that

[00:14:47] person works over time. They are able to then look at those exceptions and find out and make a

[00:14:53] determination whether or not that's actually an exception or look for a larger pattern of

[00:14:58] increasing usage of AI to generate tools and break the relationship they have. I found it

[00:15:04] to be a really insightful set of uses and the way we could think about applying policies and

[00:15:10] procedures in our day-to-day operations. Encourage you to listen to it. It is on the YouTube feed

[00:15:16] and the podcast feed now. Now, remember we're continuing to take questions and you can submit

[00:15:21] them anytime in the chat window or if you're listening to the recording, go ahead and submit

[00:15:26] for next week and send to question at MSBradio.com. Now I've got one more story before we dive

[00:15:33] into questions and it's a use case I just had to write up.

[00:15:38] Holodeck, a system for generating interactive 3D environments has been developed to train robots

[00:15:44] in navigating real-world environments. It's leveraging large language models and Holodeck

[00:15:48] can generate a wide range of indoor environments based on user requests. The system has been

[00:15:54] found to outpace human created tools in terms of realism and accuracy and has shown

[00:15:59] positive effects on the agent's ability to navigate new spaces. Holodeck addresses the

[00:16:05] challenge of efficiently generating diverse environments for robot training, expanding

[00:16:10] the scope of research beyond residential spaces. Quoting the article, named for its Star Trek

[00:16:17] forebearer, Holodeck generates a virtually limitless range of indoor environments using AI

[00:16:23] to interpret users' requests. What we can use language to control it says, Yang, you can

[00:16:28] easily describe whatever environment you want and train the embodied AI agents. Holodeck leverages

[00:16:35] the knowledge embedded in large language models, the system's underlying chat GPD and other chat bots.

[00:16:42] Language is a very concise representation of the entire world, says Yang. Indeed,

[00:16:46] LLMs turn out to have a surprisingly high degree of knowledge about the design of spaces

[00:16:52] thanks to the vast amounts of text they ingest during training. In essence,

[00:16:57] Holodeck works by engaging an LLM in conversation using a carefully structured series of hidden

[00:17:02] queries to break down user requests into specific parameters. And just like Captain Picard might

[00:17:08] ask Star Trek's Holodeck to simulate a speakeasy, researchers can ask Penn's Holodeck to create

[00:17:14] a 1B1B apartment of a researcher who has a cat. The system executes this query by dividing it

[00:17:20] into multiple steps. First, the floor and walls are created, then the doorway and windows.

[00:17:26] Next, Holodeck searches the objectverse, a vast library of pre-made digital objects for the

[00:17:32] sort of furnishings you might expect in such a space. A coffee table, a cat tower, and so on.

[00:17:38] Finally, Holodeck queries a layout module which the researchers designed to constrain

[00:17:43] the placement of objects so you don't wind up with a toilet extending horizontally from the

[00:17:48] wall. Now, why do we care? I love pulling interesting use cases again over and over,

[00:17:57] and we can always borrow from science fiction to get a sense of what's coming. Think this through,

[00:18:02] they're using the Unreal Engine to build virtual environments on demand for testing scenarios.

[00:18:08] It's incredibly powerful. If you need an infinite test variety to test robots in,

[00:18:14] you can now do so with this simulation. They can actually generate the graphical representations

[00:18:22] of full environments fully rendered using Unity. This is exactly the kind of thinking that we want

[00:18:27] to do. You might want to do it from a designer perspective for a single environment, but some

[00:18:33] use cases are required at infinite numbers and that's where the recursive use of these kinds

[00:18:38] of tools are particularly interesting. Now, we care because we want to extend the application

[00:18:44] and thinking again to our own customers. Are there scenarios where we can leverage these kinds of

[00:18:51] technologies to create multiples and scenarios we haven't before or be able to create data sets that

[00:18:58] we might not have been able to test against prior? That's the kind of thinking we want to use

[00:19:02] when considering these tools. Now, we're coming up on those questions. Make sure to put them

[00:19:07] in chat if you've got any last minute ones. I do love taking them. Focus first, always on

[00:19:13] those user submitted ones. Remember, bring your questions live and you'll get a live response.

[00:19:18] This is a great part of the show to get involved with our ongoing discussions.

[00:19:24] Let's take our first submitted question. What are the potential impacts of cybersecurity

[00:19:30] company acquisitions on the service quality and client relationships of MSPs?

[00:19:37] The acquisition question. You know, this one comes around every so often for me to get

[00:19:42] my thoughts on what happens when companies are acquired. I start with my basic advice.

[00:19:49] Anytime a company is acquired, do not panic and understand what your contractual obligations are

[00:19:55] because ultimately the company wants to continue to engage and receive your customer

[00:20:00] engagement and receive your money on an ongoing basis. So they're going to protect that

[00:20:05] contract as best as they can. And they are obligated to do that continuing based

[00:20:10] on the contract that you have with the previous organization. And remember, post acquisition,

[00:20:16] there's always 12, 18 or 24 months of merger as the two organizations come together.

[00:20:24] Oftentimes when you consider changes in organizations, they don't come until after

[00:20:29] that process is concluded. So the time to think about that and be engaged is not at

[00:20:35] the time of transaction. It's either two years down the road when they are in a position to

[00:20:41] start making changes or on your contract end. So from my perspective, I think about it from

[00:20:49] that rhythm and I advise people to not worry on day one. Now here's the other thing that I

[00:20:55] think is important specific to the solicitor's question around cybersecurity. There are,

[00:21:00] frankly, too many cybersecurity companies out there to support the market. I'm anticipating

[00:21:06] we're going to see consolidation of those vendors into a lesser set. Either companies will get

[00:21:14] acquired for talent or technology and get rolled up into more robust organizations. So you should

[00:21:21] be in a position where you're expecting more acquisitions in the cybersecurity space. It's

[00:21:28] been heavily funded by VCs and private equity over the past number of years, meaning those companies

[00:21:35] are looking for ways to extract that value out. And they're going to look to do that with mergers

[00:21:41] and acquisitions. So you should expect that there are going to be more of them to have less

[00:21:45] cybersecurity companies over time, frankly, because we have too many. It's great taking

[00:21:51] questions. I love doing it. Make sure you send them in for next week. If you didn't get a

[00:21:56] chance for yours to be heard this week. Now I want to highlight an upcoming interview,

[00:22:02] something that's coming up this coming weekend, an interview with Rodrigo Larrero on cybersecurity

[00:22:07] defense. Here's a clip from that interview. You've also offered that this, that AI may be part of

[00:22:14] the defense. How do you think that that fits in as a countermeasure? So more than just that,

[00:22:22] I think AI is the only defense that we can have against this. I mean, it's essentially an arms

[00:22:31] race, right? You don't bring a knife to a gunfight and the attackers are going to use AI. There's

[00:22:39] nothing you can do about that. So if on our side, on the defense side, you are hesitant about

[00:22:48] about the use of AI, you don't want your people to use AI, you are essentially putting your cyber

[00:22:55] defenses with knives, arming them with knives to fight off gunslingers. So that's why I think that

[00:23:04] we must use AI. One of the most obvious applications of AI is in terms of a cybersecurity

[00:23:12] analyst. Our industry is desperately lacking qualified cybersecurity analysts. So AI can

[00:23:23] contribute to increase defectiveness and the productivity of those cybersecurity analysts,

[00:23:29] because right now, the few that we have are already overwhelmed with responding to incidents in

[00:23:36] order to increase and protect against these heightened attackers. The only way we can do that is

[00:23:46] through the use of AI. Rodrigo has some bold points. In fact, he thinks AI may be a nuclear level

[00:23:54] event in cyber. I do push back in the interview and I encourage for you all to listen to that on

[00:23:59] this coming weekend and we'll discuss again next week in the live show. A reminder for

[00:24:04] listeners that my Patreon supporters already have this video and you can get all my interview

[00:24:10] content early as a supporter, visit patreon.com.com. through MSP radio to sign up now. I want to

[00:24:17] thank Salesforce Builder, our Patreon sponsor who support made this show possible. Focus on your

[00:24:22] IT sales workflow with the power of automation and visit them at salesbuilder.com. That's

[00:24:30] And vendors, you too can get your name mentioned on the live show.

[00:24:35] It's a simple monthly subscription.

[00:24:38] Visit patreon.com slash MSP radio for more.

[00:24:41] And listeners, thank you for all your support.

[00:24:44] And here's what you can do to help continue with the show.

[00:24:47] Make sure to like share and follow on all your favorite platforms and share it with

[00:24:52] your friends and encourage them to listen as well.

[00:24:55] Or you can support directly on Patreon with our give what you want model.

[00:25:00] You set what you think the content is worth.

[00:25:02] And if you have a question and are listening to the recording, send it in at question at

[00:25:07] MSPradio.com.

[00:25:09] Next week, the live show is on Wednesday, so catch us live on Wednesday, 3pm Eastern.

[00:25:15] Thanks for joining me for the Business of Tech Lounge, and I will see you next time.