Integrating AI into Your Existing Systems: A technology perspective.
[00:00:00]
[00:00:09] Nichola: Thank you for joining us, Lee. So we're going to be talking about AI as it's such a hot topic at the moment. So businesses with existing systems, how should they even start when integrating AI?
[00:00:18] Lee: I think the really important thing is to start small. if you're just dipping your toe in the water, so to speak, you want to find an area of your business that can benefit from the power that AI can provide.
[00:00:32] But something that's not going to be missed if it doesn't work too well to start with because you're going to spend some time kind of tweaking it and, evolving it as it becomes more and more, useful to your business, and by starting really small and just picking out without one little service or that one little thing you're minimizing your risk as well.
[00:00:56] Is there any examples of any sort of automations that might help?
[00:01:01] Lee: Yeah, I think so. You could have something that just, Let's take a support desk system.
[00:01:06] As an example, you might have a piece of AI tech that's going to sort those support tickets in order of priority for your human agents to then deal with. And in that way, they can provide a better service to your customers because they're dealing with the important things first and your customers don't even know that AI has been involved in the process.
[00:01:28] Nichola: How does it even do that? How does it, prioritize? What is behind it.
[00:01:32] Lee: There's a lot of different mechanisms in AI. a lot of people right now when they talk about AI, they think about, what are actually large language models, things like chat, GPT and Claude and the others.
[00:01:43] And that's only a very small part of AI, but in this scenario, that's the kind of thing you would use.
[00:01:49] You might use, some sentiment analysis on it. So you might get an idea of is the customer angry or are they upset. Are they happy? and use that as one of the metrics that you might prioritize by.
[00:02:01] But equally, you could have another little bit of AI that's looking at the ticket and trying to determine, how important is this thing?
[00:02:10] is this problem? The customer's got something that's breaking their business, or is it just a question?
[00:02:18] being able to then use lots of like different data points, in that sense to provide an overall priority score.
[00:02:26] It will help you then answer the questions in a good order so that the people with the biggest problems or the having the hardest time I guess would get the answers quicker.
[00:02:38] Nichola: Oh wow. Is there any companies that use that currently? Do you know?
[00:02:41] Lee: I think it might be behind the scenes in a lot of different companies at the moment, but the beauty of it is you don't know. As a customer, you wouldn't even know that's happening behind the scenes.
[00:02:51] And that's one of the best ways to use AI at the moment, is to augment your existing processes rather than getting customers to interact directly with AI, which they can always tell, and it can be frustrating sometimes.
[00:03:06] By using it behind the scenes like that, you're still getting value out of it, you're still doing something worthwhile with the technology.
[00:03:15] And it's helping your business and helping your customers at the same time, providing a better customer experience. But you're not getting your customers to directly interact with an AI.
[00:03:24] Nichola: Are there any misconceptions in AI in business that you've seen?
[00:03:29] Lee: Yeah, quite a lot to be fair. and similar to what I said earlier that a lot of people when you talk about AI right now, they are thinking chat GPT, Open AI, Claude, all those large language models, the chatbots conversational.
[00:03:44] Style AIs. And that's actually only a very small part of the AI umbrella.
[00:03:50] If you like, AI is a big kind of overarching term that covers a lot of different areas of machine learning. And some of the more useful areas of machine learning actually are things where it can sort things into categories for you.
[00:04:05] So a good example of that would be a content recommendation system on a website. For example, you might have, where you ask the AI to provide a recommendation of five pieces of content for a particular user and in doing so it's got to look at what that user's already looked at on your website and under the hood comparing it or comparing that user to other users who've looked at similar things and it's grouping those users together and trying to find similarities, but also what have those other users looked at that this user hasn't yet?
[00:04:41] And those become the recommendations because you're like these other users. So you might like the same things. And in doing that, you can get recommendations for content without, needing your editors to tag content with specific categories, without needing your users to actually go into their account and enter what their interests are and things like that.
[00:05:02] You can do it seamlessly and again, invisibly to the end user. You're not, they're not interacting directly with the AI again in that scenario. So it's just providing those recommendations.
[00:05:13] Nichola: Do you think AI is limitless? Can you do everything? Or is it not there yet?
[00:05:19] Lee: I don't think it's quite there yet. I think that it's got a lot of potential. And the main, Thing it gives you is that it can make decisions and do this categorization and deal with data far faster than a human being can.
[00:05:35] And that's really its power. it can do so many more computations on whatever it is than a human being can in that same time span. The quality you get out of it right now may not be the same. there's always that human intuition that you can't really put into a machine at the moment it may get there, it may eventually become that.
[00:06:00] But right now, I think there's a, there's always an argument to have a human oversee what's coming out of an AI in some senses and be able to review it.
[00:06:12] And that in where AI is in use at the moment. you hear a lot about algorithms in social media. They're AI under the hood, it's machine learning.
[00:06:21] They're learning what you look at. So when you're scrolling down your Instagram feed or your Facebook feed and you're looking at different images, different posts, it's timing how long you look at them for.
[00:06:34] It's seeing what you click like on it, seeing what you open and it's building a picture of you.
[00:06:41] in, in the AI and it's going to group you with all the similar people, like the content recommendation algorithm we talked about,
[00:06:47] and then it's going to do that. It's going to put more of the same things on your feed. And it turns social media into a bit of an echo chamber in some respects, but you don't necessarily realize that say, I under the hood when you're using it, but you do have the ability within the app to go, I don't like that.
[00:07:06] Or take that off my feed, hide that, or whatever it might be. And that's also feeding back into the AI and going,
[00:07:15] okay, they said explicitly they don't like that thing, so we'll show less of that. and it might take three or four don't likes to get there, but it becomes more and more accurate over time.
[00:07:29] Nichola: Yeah, it's constantly learning.
[00:07:30] Lee: Yeah.
[00:07:31] Nichola: With, I know AI, we said AI is a hot topic at the moment, but how long has it actually been around?
[00:07:37] Lee: Oh, decades. 20 odd years, easily. the main reason it's become such a hot topic lately is partly because of ChatGPT and their marketing machine. And partly because of the massive increase in processing power available today.
[00:07:52] as more and more processing power becomes available because the it's only going to become more powerful as time goes on, the use of AI in particular, large language models and these, much larger kind of neural net AI's, they are going to become More and more human as we go forwards.
[00:08:16] I think there's always going to be some tells. I think there's always going to be those little idiosyncrasies where they can hallucinate for a start and give you the wrong answers. And it's very difficult within that model to eliminate that because the premise of a large language model is that it's a chat, it's a conversation.
[00:08:38] And you could almost argue that giving wrong answers is very human.
[00:08:43] Nichola: Yeah, exactly.
[00:08:44] So if you want to integrate AI into one of your systems, is there a lot of rework?
[00:08:49] Can you just add it on? So what is that process?
[00:08:53] Lee: if you're working with a microservice architecture, then you could almost slot it in, as another service. And every service in that kind of architecture has an input and an output. So you would. Give it some data input and take the output and then act on it in another service, but in More monolithic systems.
[00:09:13] It might be a little bit more difficult to slot in You might have to do a bit of work to actually embed it into the system, but ultimately I would say To anyone that's considering AI into their systems or products Stop and think about it first and think, are we doing this because AI can actually improve or add value to our product?
[00:09:36] Or are we doing it just to slap that sticker on and say, we're powered by AI. And if it's the former, then you should already have a really good business case for why you want that AI in that product. And you should, you'll know how it's going to fit in with the rest of your system. If it's the latter, you'll struggle more.
[00:09:55] Nichola: Thank you for that, Lee.
[00:09:56] Lee: Yeah, I've got a question for you. What do you think about AI?
[00:09:58] Nichola: I think AI has positives and negatives. I think the positive side of it is that it can speed up processes. I think the negative side of it is that it could, by speeding up those processes, you're probably making other roles slightly a bit more redundant.
[00:10:14] But I think it's a case of building on AI and using it for your benefit and speeding up your processes so that you can then concentrate on other things and put more effort into more important things.
[00:10:26] Lee: Yeah, it's interesting that there's some parallels there with what happened in the automotive industry when they started adding robots to the assembly lines, made a lot of people redundant, but it also generated a lot of new jobs as well.
[00:10:39] So it's a pros and cons,
[00:10:42] Yeah, definitely. I think if you can harbour it and use it in a really positive way to speed up what you do,
[00:10:44] Nichola: Okay. Thank you for that, Lee. It was, yeah, really interesting.
We are the home of unreasonable progress. We are distinction.
