AI Is Dumber Than You Think, But Here's Where It's Getting Good 🤖📈

Hey, I'm Jason Heiner with ZDNet. We're here at Spice World 2024. I'm with Jim Raposa from Aberdeen Research. And Jim, you just did a session on AI, the topic of the moment, and we got a really interesting reception in that because you are both an IT skeptic and an IT enthusiast. So tell us a little bit about why. It's true. Like everybody, when you see some AI stuff, it's kind of mind-blowing. The first time you create an image with AI or you remove a photo bomber from a photo and it's like, wow, they just disappeared and you kept your background and it looks great. Or you create something. So all of that is awesome with AI. But I'm also highly skeptical. I find AI is often wrong. I find the use cases stupid. I find AI itself often stupid. And that doesn't mean it's not effective, but the ideas that people have of AI that it's, you know, that we're going to be at this moment where Skynet becomes aware, not likely anytime soon. Not even close. Yeah. You talked about there's some different opinions about how smart AI is and how dumb AI is. There's a wide range from really what are considered experts here, right? Yeah. Well, there's some, and I won't name names, but some prominent AI businesses have their leaders have come out and said, you know, within thousands of days that AI will, you'll have super intelligent AI, like basically Skynet or the matrix or whatever you want to use as an example. You know, so that's one opinion. A very prominent Turing Award winner, you know, Meta's chief AI scientist, Yann LeCun, said current AIs are not as smart as a house cat. Probably not even close. House cats are barely smart. I think of a house cat as smart. I do not think of AI even close as smart enough. And he doesn't expect it to ever get, you know, to get smarter than a cat anytime soon. Maybe like 20 plus years. And interestingly, some of the companies that were promoting super intelligent have come back and said, well, current large language models have probably reached their limit. And the idea that we can throw more data into them or throw more compute at them and they'll continue to improve is probably not accurate. Yeah. So. Yeah. You've also talked to a lot of businesses that are using AI about how they're using AI, what they're using. And there's a big difference or there's an interest, some interesting differences between the size of businesses as well. Talk a little bit about that. Yeah. Well, as you expect, you know, enterprises are more likely to have like some made some dedicated investments in AI and that, I mean, not just using copilot and things that come in, but, you know, we're trying to like build our own models. We're trying to use our own data. We're trying to customize AI to meet our own needs. So over half of large businesses, enterprises are doing some of that already. But still over 30% of small businesses are doing that. And, you know, and it's, it's interesting, you know, medium sized businesses are also there as well, but we often do see this sort of dichotomy where both large businesses and small businesses sometimes are more innovative than the businesses in the middle. It's just a resource thing. Yeah. Large enterprises have more resources. Very small businesses are sometimes like agile startups. So they're just super nimble, experimental, right? Yeah. So there's also a number of, of issues with the things that people are choosing to use AI for in business. You've uncovered and talked to them about, you know, what are the use cases for it? What are, what are some of the things you found? It's really promising for businesses today because the cynical part of me thinks maybe they're all going to be, oh, we're doing chatbots and we're creating AI powered assistance. But they're, they're taking a very practical approach, often not even really gen AI, really what we think of as classic AI machine learning. They're using AI to, you know, improve analytics, to improve self-service in like IT support, to put more knowledge in the hand of IT workers, to, you know, improve predictive analytics, to help with setup of management of like infrastructures and things like that, to optimize your network, things like that. Yeah. Things AI is really good at, to the way you can say, here's a bunch of data, either summarize this or remove things that I shouldn't have to care about. Things that might've taken an IT person days or at least several hours to do. Yeah. And AI can do pretty effectively very quickly. And these are very sensible, maybe not sexy, but useful ways to use AI right now. So like the things that AI is very good at is there are clear, you have really good data sets. There are very sort of clear parameters. It can crunch that stuff and do it really well. Where it's not so good is where you need to make a judgment call, where there's, you know, gray areas where there are large continuums. Yeah. Is that fair? You never want AI, especially a gen AI, to be the final source of truth. Never, ever, ever. And that's something a lot of times people talk about AI. It's like, oh, well you can use AI to start coding. You know, it'll generate a base and you can go in and clean it. You can use AI to create a summary of like a valuable report or a security report. You know, you can use AI to basically create some kinds of code. And these are all true, but the truth is not everyone is going to do that. They are going to use the final output from the gen AI without checking to see if it's actually true or even close to true. Or if it made things up, you know? Yeah. So you need to have a human filter, you know, a human editor for lack of a better word, afterwards going through a check. And humans being who they are, that's not always going to happen. Yeah. Fair. So for the average person, gen AI may just end up being a better, easier, smarter search tool. Yeah. Yeah. Gen AI is actually really good at searching. You know, that's some of the ways you can use like, oh, where in my company is this document that I need? Where is our, you know, compliance report, things like that. It's really good at that. It can go through a lot of stuff. Yeah. And that's, you know, that's why nobody's talking about it that way because it's not sexy. You know, they want gen AI to be this answer of all questions, to be the Oracle on the mountains, and it's not. But if you want it to like query your employee manual to find like, what's the parental leave policy? How long do I get? Well, it's great for that. It can parse that with the data that it has. Right. So one of the things you said, you know, to sort of wrap this all up, you know, you made a convincing case that you think we are likely in the next few years going to say generative AI failed. Right. Because there's all this hype. It's going to be very difficult to live up to it. And, and yet, we may still end up using it every day, it may have failed, and we still may be using it for more and more things that just are not very flashy. Like a lot of previously overhyped technologies, we've seen it before, because it's not going to do the things that everybody's talking about. It's not, you know, going to replace workers, it's not going to write, you know, Oscar winning movie scripts, it's not gonna, it's not, it's not going to do any of that stuff. Sometimes it'll do cool things, but it's not going to do those things. And so people say, like, oh, AI failed, it didn't become super intelligent in 1000s of days, it didn't do all that. But at the same time, it's going to be running everything underneath. Like, a lot of our software is going to be improved, optimized by like common sense and sensible applications of AI. Very good. Well, Jim Raposa, thanks for your time. For more on this topic on AI, you can visit Aberdeen Research and ZDNet and find lots of good information, usable data on how AI can help you be smarter and do better stuff. Thanks for tuning in.