Transcript
Transcript: Trends in Technology-Driven Change: Creating Value with Generative AI
[00:00:00 Text appears onscreen that reads "Trends In Technology-Driven Change".]
[00:00:06 The screen fades to Chris Howard.]
Chris Howard: Hi, I'm Chris Howard. I'm the Global Chief of Research at Gartner. Thanks for taking some time to listen to the advice that we have around A.I. and related subjects. I hope you find it interesting.
[00:00:17 Text appears onscreen that reads "Creating Value with Generative A.I.".]
[00:00:21 A slide is shown that reads:
"-GenAI value is closely tied to industry + role
-GenAI value is increased in human + A.I. collaboration services (vs. either alone)
-Text-only GenAI (e.g., ChatGPT) is less transformative than multimodal GenAI (image + voice + text)
-GenAI alone is less transformative than GenAI + other A.I. capabilities: this is called composite A.I.
-GenAI requires modern data architecture and governance to produce high value
-Greater value is often generated in an ecosystem where composite A.I. capabilities from connected partners work together".]
Okay, so generative A.I., what we've learned here is that there are a number of things that create more value. It's closely tied to industry, and well, what do I mean by that? It's that there are use cases in every industry but they feel very different from one another. So, use case for government, a lot of it has to do with creating better services for citizens of understanding complicated policy environments. I used it when I was here last, and some of you have heard the story. I was learning the A.I. bill and (inaudible) stuff from the government. I just fed the PDFs into ChatPDF and had a conversation with them. So, I didn't have to read the 140-page bill. I was actually able to ask about ramifications and dependencies and all those kinds of things. So, it's a way of kind of parsing through complex stuff, and you sit on tons of complex stuff.
When you combine humans working with A.I., you get better results than either one on their own. And so, you're getting a higher-level result, especially a more human result. What I mean by this is that well-designed artificial intelligence provides cognitive offload for humans so that they can do things that humans are well-designed to do. I'll give you a very specific example. So, there are a number of countries that have implemented a bot that sits on 9-1-1 calls and is assisting the human who's also on the call. And so, that bot may be trained to watch for very specific indicators, patterns of text, sound in the room, that kind of thing, maybe to look for congestive heart failure, and its role is to spot that with enough accuracy to dispatch the emergency services. Well, the person's role is to calm the other person down and get better information from them, higher fidelity information, as opposed to having to think about all of this other stuff and dispatch and so on.
So, the two of them working together, the results are so much better than either one of their own, okay? A machine by its own, a little bit better than the human, but together, their ten points, percentage points, better in terms of accuracy, and you can take that across any domain and that remains true. So, I'd be looking for ways to combine this thing, which is really… what it's for is to create better decision fidelity, help you make better decisions, whether that's differential diagnosis or helping with a case that's open or dispatching an ambulance in a health care situation. That's where you get it.
Text-only generative A.I. is less transformative than when you start to add pictures and video and audio and other things, schematics and so on. What I worry about is that for a lot of people, ChatGPT feels like fancy search, which it is, right? But its real value is much more creative than that, and something I've been thinking about a lot lately is that the way that you make… the way that you keep it from writing your obituary is to ground it with your own data. So, actually, you're telling it, use only my data to create the generated response to try to make it more accurate, but getting to 100% accuracy is really, really, really hard and may never be possible. And so, what I've been thinking is that, actually, generative A.I. is less about accuracy and more about creativity.
And so, where are those places where you need to kind of shake people up or change the logjam in their mind to give them different options of the ways of doing things, some of which may be wrong, and then let the human decide kind of what are the better choices in the midst of that? Because if what we really want is accuracy, there are better ways to do that, things like knowledge graphs or vector embedding in databases or those types of data techniques. If you want accuracy, business (inaudible), like those are the things you use to get accuracy. It's the combination of the two that gives you an insight that you might have been missing before. So, that's something really important to remember.
When you combine generative A.I. with other types of A.I., that's called composite A.I. So, if you see that term, that's what that means. It requires data investment. And so, this is… I hear a lot about this when I come to Ottawa and work with the departments, is that there's a lot of stuff in there that's 60 years old. A lot of it's still on paper and spread across multiple systems. Those processes have calcified around them, right? It's very, very, very hard to change. To take true advantage of these systems, you need to actually invest in that data environment to get it to work more effectively. This is probably the hardest problem that you have, the legacy environments. When it comes to legacy code, some interesting things are happening, that you can use these tools to convert from one language to another.
So, for example, if you're trying to get off an old COBOL language or something like that, you can actually do conversions of it into new languages like Python, with a caveat. All it does is translate it. It doesn't make it better. So, it'll take crappy code and make it crappy in a new language, right? Which gets you part of the way there, but if you think that's the end, then… but it's a great place to start. If you actually want to get it into a more modern environment, these generative A.I. tools can help you to do that, and then generate it in an ecosystem.
So, this often comes up also when I'm working with governments, sort of, how does it change the nature of information processing, like what you would get from the private sector that would be useful in a public sector setting and vice versa? And so, understanding the scope of where you want to create insights on your data to share it out with the private sector and vice versa, because this is not cheap yet. This is really quite expensive to do. And so, to be thinking about, where is it so important that it's actually worth the investment for us to do it on the data that we have versus getting it from the financial services sector or the health care sector or somewhere else. We'll re-forge those relationships.
Thanks for watching. And again, I hope you found this useful and interesting for the work that you're doing in Canada.
[00:05:48 The CSPS logo appears onscreen.]
[00:05:55 The Government of Canada logo appears onscreen.]
