Are We Still Repeating the Same Mistakes with AI? [AI Today Podcast] AI transcript and summary - episode of podcast AI Today Podcast: Artificial Intelligence Insights, Experts, and Opinion

Go to PodExtra AI's episode page (Are We Still Repeating the Same Mistakes with AI? [AI Today Podcast]) to play and view complete AI-processed content: summary, mindmap, topics, takeaways, transcript, keywords and highlights.

Go to PodExtra AI's podcast page (AI Today Podcast: Artificial Intelligence Insights, Experts, and Opinion) to view the AI-processed content of all episodes of this podcast.

View full AI transcripts and summaries of all podcast episodes on the blog: AI Today Podcast: Artificial Intelligence Insights, Experts, and Opinion

Episode: Are We Still Repeating the Same Mistakes with AI? [AI Today Podcast]

Author: AI & Data Today
Duration: 00:11:03

Episode Shownotes

Artificial intelligence has been on the horizon for over seventy years. In fact, the term AI was officially coined in 1956. So, why does it seem so close but also so unattainable? In this episode of AI Today hosts Kathleen Walch and Ron Schmelzer discuss the question: Are we still

repeating the same mistakes with AI? Continue reading Are We Still Repeating the Same Mistakes with AI? [AI Today Podcast] at Cognilytica.

Full Transcript

00:00:01 Speaker_00
The AI Today podcast, produced by Cognolitica, cuts through the hype and noise to identify what is really happening now in the world of artificial intelligence.

00:00:10 Speaker_00
Learn about emerging AI trends, technologies, and use cases from Cognolitica analysts and guest experts.

00:00:22 Speaker_01
Hello and welcome to the AI Today podcast. I'm your host, Kathleen Walsh.

00:00:26 Speaker_02
And I'm your host, Ron Schmelzer. You know the interesting thing about AI is that it's like the oldest new technology or the newest old technology, whatever you want to call it. AI has been around as long, if not longer than computers.

00:00:40 Speaker_02
We all know that the history of AI, that the term AI was coined in 1956, just like the term. But obviously you can't coin a term if you haven't been talking about it before. It's not like you all get together and like,

00:00:52 Speaker_02
say, let's make up some new idea, and then let's come up with a term for it, and now we'll stick with it. Clearly, people have been doing AI research for decades before the term was coined.

00:01:01 Speaker_02
And it was under many different ideas, of course, the ideas of making machines intelligence. But the thing is, if AI has been around for so long, why does it seem like it's always just around the corner but never there.

00:01:16 Speaker_02
Every single wave of innovation that we've had with computers, with technology, we feel like surely in this wave we will have licked all the problems of artificial intelligence and we are just one or two innovations away from the intelligent super machine doing everything we want that we see in science fiction.

00:01:34 Speaker_02
It's like this elusive goal. Maybe we're much farther than we seem. And part of that is because we seem to be constantly making the same mistakes with AI over and over, of course, in different forms.

00:01:48 Speaker_02
But it just seems like we repeat history quite a bit when it comes to AI.

00:01:53 Speaker_01
Exactly. And so if you've listened to our podcast for a while now, you know that we've had podcasts on the AI winters. And so we won't go into them in great detail. I'll link to it in the show notes.

00:02:03 Speaker_01
And you can listen to those podcasts to hear, you know, the history of the AI winters. We also have some articles on it and also in our cognitive project management for AI training and certification, the CPM AI certification.

00:02:15 Speaker_01
we do go into AI winters as well. It's important to understand, and so we'll give a high-level overview on this podcast, because that helps frame why we still see these problems.

00:02:25 Speaker_01
So Ron said the term was officially coined in 1956, and you may go, wow, if it was coined in 1956, why do I feel like I'm just now hearing about it, right? Like, that's older than maybe many of our listeners on this podcast.

00:02:37 Speaker_01
And so you go, well, why hasn't it continued to evolve, and why haven't we seen things happen? We fall into these traps with AI, and that's what brings us to these winners.

00:02:46 Speaker_01
And the winners are a period of decline in investment, decline in research, decline in adoption by organizations. And so people just, you know, do things a different way.

00:02:57 Speaker_01
And because people aren't, you know, really investing and using artificial intelligence, then it goes into this kind of dormant period. And that's why we call it a winter.

00:03:07 Speaker_01
So, in 1956, the term was officially coined, and there was a lot of interest with, you know, what we could do with artificial intelligence. There was a lot of AI research.

00:03:17 Speaker_01
There was these burgeoning fields of computer science and, you know, neuropsychology, brain science, linguistics, all these different related fields. And one thing that's really great about AI is it does bring in all these different fields.

00:03:30 Speaker_01
That can also cause a lot of terminology issues, but that's for another podcast. So, basically, we had all of this different research, and governments were investing heavily in artificial intelligence.

00:03:43 Speaker_01
But around the 1970s, we fell into our first AI winter, which is this period of decline in investment. And basically, funders of AI just realized that it wasn't, you know, we didn't, we didn't have have the data that we have today.

00:03:58 Speaker_01
We didn't have the compute power that we have today. We didn't have machines that were, you know, the processing power that were capable of what we have today.

00:04:08 Speaker_01
And there were a few other things, the Lighthill report in particular, which we can link to, really kind of killed that first period of, you know, AI excitement. And so we fell into this decline of interest and funding and research.

00:04:23 Speaker_01
And so we fell into our first AI winter. So you can go, okay, well, how did we get from our first AI winter to here? Because you wait, there was a second AI winter.

00:04:32 Speaker_01
So back in the 1980s, we started to have desktops on machines and people said, well, we have all this compute power now, what can we do with this? Organization said, how can we leverage this to our advantage?

00:04:45 Speaker_01
And so this is where expert systems really kind of had their heyday. And so this was supposed to mimic those experts at your organization.

00:04:55 Speaker_01
But what people found is that they were brittle and they were really hard to work with, and we still didn't really have what we have today, right? We still didn't have the data that we have today. We didn't have ways to manage that big data.

00:05:11 Speaker_01
Computing was still incredibly expensive. Even though we had more desktops than we did in the past, and we had more compute power than we did in the past, really didn't have what we needed. And so this is what brought us into our second AI winter.

00:05:25 Speaker_01
And so, you know, maybe in the in the early 90s, late mid 90s. Now, in this wave, we did see some incredible things. This is when IBM's Deep Blue beat chess grandmaster Garry Kasparov. And so, you know, we had some exciting things going on here.

00:05:41 Speaker_01
But again, we still had this big, overall overarching theme that we were just over promising and under delivering on what we could do. do when it comes to AI systems. Now, you can say, why were we over-promising? And why were we under-delivering?

00:05:55 Speaker_01
We talked about some of those issues. But one thing about AI that's really different than other technologies, like mobile, even the internet, is that science fiction, Hollywood hypes it up so much. They get so excited with what can happen.

00:06:08 Speaker_01
We think about these machines that act and think and behave like humans. And so we conjure up all these different ideas. And the technology just really wasn't there.

00:06:20 Speaker_02
Yeah, and I think one of the things is when we think about the AI winters is not just as a history, and it's true, there's history here and some of you may be really interested in the history of AI, but it's really to think about the lessons that we learned from those failures and that first wave of AI, which came from basically the invention of the digital computer combined with

00:06:40 Speaker_02
all these things that we thought that we could do with a smart computer, we had some pretty remarkable, really cool demonstrations, if you will, of this technology. Claude Shannon, if you know him, he's an informaticist.

00:06:54 Speaker_02
He basically developed a lot of the core concepts around information science and all that sort of stuff. He demonstrated a mouse that can navigate a maze autonomously. And there was no big data. There was no internet. There was no supercomputer.

00:07:11 Speaker_02
There was barely any computer and nothing that would fit inside of a mouse. So it's all electromechanical. And clearly, something was happening there, right? We even had things like Gray Walter's tortoises, which were

00:07:23 Speaker_02
can navigate the world around them, find their own docking station, plug into them. If you look at them, they're very simple machines that really leverage cybernetics, this whole feedback systems approach.

00:07:34 Speaker_02
Not a single line of Python code in any of these things. If you think about it, in some ways, those innovations are much more impressive than the ones we have today, where we've trained monster GPUs on

00:07:48 Speaker_02
hundreds of millions of dollars worth of GPUs on petabytes upon petabytes of data, so they can basically recite poetry and generate images and do all sorts of NLP tools, but it can't do anything else.

00:08:00 Speaker_02
Is that more impressive than Grey Walter's tortoises? I don't know. I mean, you could argue that that first step was much more impressive in many ways. We even had our first image recognition systems, a class of binary classification system.

00:08:14 Speaker_02
You might have seen the early perceptrons where you could take images and put them under very simple photo cells, and it would do very basic categorization, gender-based categorization. That was the thing that we were doing back then, right?

00:08:27 Speaker_02
just based on an image, and that was impressive. So what happened? Why didn't all of a sudden we just went and went and went and went? Well, it's because we realized the limitations of those systems.

00:08:38 Speaker_02
There's only so much that we could do with these pattern recognition systems. And at some point, we realized that our problems couldn't be solved. We couldn't solve those problems with these AI systems.

00:08:49 Speaker_02
We had to solve them in other ways that were not AI. We basically invested in non-AI systems, and those things took off.

00:08:55 Speaker_02
We realized that what was more important, more transformative, the computer was more transformative as a business tool, as a personal computing tool, than it was as some artificial intelligence tool. So, PCs took off.

00:09:08 Speaker_02
Business computers took off and AI didn't. It declined right about the same time, right? Mid-70s is when the PCs started taking off, the Apple and all those devices. And of course, people just weren't interested in AI anymore, right?

00:09:20 Speaker_02
Then, of course, we had the late 80s, early 90s, and now we had networked systems, we had desktop computers, the deep blue, all this sort of stuff raised our interest. But then what happened? The decline happened not too much longer after that.

00:09:32 Speaker_02
The second wave of AI was much shorter lived. in many ways than that first wave. And that's, again, because we realized limitations of expert systems-based approaches. They didn't really work very well. They were very brittle, as Kathleen mentioned.

00:09:46 Speaker_02
But then what was taking off at the end of the 1990s? The internet, right? The internet took off, and the internet got people's attention.

00:09:53 Speaker_02
People were like, wow, I can gain more benefit from spending my time and effort building all this stuff, all this dot-com stuff that was ramping up.

00:10:02 Speaker_02
The AI stuff is just a lot of promises and not a lot of delivery and people just no longer interested. So this latest wave of AI, powered by big data, powered by computing resources, powered by deep learning, of course, now neural nets,

00:10:17 Speaker_02
has been a much longer wave. I'd say like the wave started probably in the mid 2000s, 2005, 2006 relative time frame. We are now 2024. So, you know, when this podcast is recorded, 18 years is probably about as long as that first wave.

00:10:34 Speaker_02
So are we finally here to stay? Are we not going to hit any winter? The answer is no, because people's behavior is the same way. They're like, oh my goodness, this generative AI LLM can generate some really awesome text, image, music, whatever.

00:10:49 Speaker_02
Therefore, I have super intelligence, clearly just down the road. And I'm not really sure how they're making these connections here. I'm like, yeah, but your LLM can't even do math. for various reasons, we explain.

00:11:01 Speaker_02
The LLM is coming up with things that it clearly can't understand. Yes, your image system will still generate six fingers if you don't moderate it properly. It'll come up with text that's clearly hallucinating.

00:11:12 Speaker_02
So how you can jump to the conclusion that we are just inches away from AGI, general intelligence systems, is kind of beyond me. This is people over-promising. This is the classic over-promise.

00:11:26 Speaker_02
You could say that we are probably, you know, if the history is a teacher, we may learn our lesson that we may repeat the past.

00:11:35 Speaker_02
Now, we actually talked a little bit about this in one of our previous podcasts, that part of the reason why we're not going to gain superintelligence is because a lot of people are missing the U layer and the D-I-K-U-W pyramid.

00:11:45 Speaker_02
They're not even thinking about it. They think all they got to do to go from

00:11:49 Speaker_02
knowledge of patterns to basically know what the patterns are doing to all of a sudden being able to make these wise decisions to know a superintelligence knows how to apply them, I can just somehow go from one to the other.

00:12:00 Speaker_02
We're like, well, you're at least one layer away from that, which we haven't even done any real... Just one layer, not hard or anything. Yeah. So we're here again, right?

00:12:13 Speaker_01
Exactly. And so we are excited. So we say, you know, we are in, you know, this thawing, this AI spring, we like to call it. And we are here for a number of different reasons. And, you know, we've had some incredible things happen in this wave.

00:12:26 Speaker_01
And I think that now, you know, a lot of our listeners, a lot of people that we talk to really have been excited. I mean, we have talked about, you know, Will there be another AI winner?

00:12:35 Speaker_01
If you've listened to our podcast for a number of years, you know, we bring this up every once and again, and we really do think about this because we say the number one reason, we talk about how there still is a lot of failure when it comes to AI projects, anywhere from about 70% to 80% of AI projects are failing.

00:12:52 Speaker_01
So this isn't like, you know, we're totally hitting it out of the park right now and that everybody's kind of moving forward and succeeding. Well, when that happens, you have to look at, you know, money, time and resources for these AI projects.

00:13:04 Speaker_01
And they're not free, right? They're very data intensive. You know, we say data is the heart of AI, and there's a real cost to all of this. So as we are looking at these AI projects, we say, all right, This is great.

00:13:17 Speaker_01
And now with generative AI, I think a lot of people are super jazzed up about what they can do, really feel that it's at their fingertips.

00:13:24 Speaker_01
The difference between this wave two and the previous waves was that in the first wave, it really was just governments that were investing in AI, Because again, it was so incredibly expensive.

00:13:35 Speaker_01
In the second wave, it was organizations that were also investing in AI. We saw the rise of venture capitalists. And so organizations had money and they were investing in this. But now in this third wave, it really is in the consumer's hand.

00:13:50 Speaker_01
And so you yourself can

00:13:52 Speaker_01
you know, are using and interacting with AI every day, whether or not you realize it, you know, helping you write better emails, it's helping you with navigating, you know, from point A to point B. But also now it is with generative AI really helping you write better emails or write blog posts or help with marketing or help create images.

00:14:14 Speaker_01
And, you know, kind of be that companion. We always talk about augmented intelligence, right? And so it really is helping you maybe do your job or task or role better, especially specific jobs or tasks or roles.

00:14:25 Speaker_01
And so that's what's different about this wave, where that didn't happen in previous waves. You know, the average everyday person wasn't interacting with these AI systems. So Is it here to stay?

00:14:36 Speaker_01
We're really hopeful that it's here to stay, but we always, always say you have to look at why we went into AI winners in the past. And the main reason is over-promising and under-delivering on what these systems can do. That still can happen.

00:14:50 Speaker_01
You know, generative AI only takes us so far. Ron, Ron jokes a lot on these podcasts about how you're going to go through a drive-through window and start asking whatever the ordering system to help with your math problems or your homework.

00:15:04 Speaker_01
And that's not the right use case for that technology. So you have to understand its limitations. Or with, you know, ChatGPT, people go, wow, it was able to create this image from just a little prompt.

00:15:16 Speaker_01
Oh, wow, it was able to create this blog post because I, you know, put in this prompt. Help me with this math problem. Help me do this. Help me do that. And it's like, You still have to learn how to use it, right?

00:15:27 Speaker_01
So we have a whole podcast series on prompt engineering and best practices for that, learning about prompt patterns, learning how to interact with these systems. And also we talk a lot about the soft skills of AI, right?

00:15:41 Speaker_01
And you need to collaborate and you need to, you know, share your prompts with others and really learn from others. We're still at the beginning of this.

00:15:49 Speaker_01
And we still can continue to over-promise and then under-deliver on what it can do, because people get so excited, they really see what's happening. And now, because it is in the hands of everybody, it's a good and a bad thing, right?

00:16:00 Speaker_01
They can see the power behind this, but then also maybe start using it in ways that they shouldn't, they're misusing it, and then they're getting frustrated because it's not bringing the results that they want. And then they're like, you know what?

00:16:11 Speaker_01
We're just gonna drop this for a different way that we know will give us the results that we want. And that is that over-promising and under-delivering, and then falling back into an AI winner.

00:16:22 Speaker_02
Yeah. So, you know, part of the reason why we keep talking about this is, again, this is people-processed technology. Technology is not going to get you out of this hole. Technology will just, it's just technology. It's a tool.

00:16:32 Speaker_02
Just like every other tool we've ever had in the history of human civilization. It's like, it's up to you to figure out what to do with it. And there's no such thing as one-size-fits-all. One tool does everything. Jack of all trades, master of none.

00:16:43 Speaker_02
But I would say, I would say search in your organization. And think about situations where there might be over-promising going on. And you should heavily understand that there's most likely going to be under-delivering going on.

00:16:56 Speaker_02
Someone's like, let's put this great interactive chatbot on our site and can ask every question about blah, blah, blah, and it'll be great. It's like, OK, it's not a bad idea. Not a bad idea. So that's the think big part. OK.

00:17:07 Speaker_02
But let's be a little cautious about our promises as to what this can do. Let's do a little start small activity, see if it actually will give us good responses, even in the most limited of cases. And if the answer is yes, then okay, great.

00:17:21 Speaker_02
Maybe it is a good tool to solve that problem. But then what's happening is what you're really doing I want to give you a little bit of insight. We never actually really haven't talked about this. Think big, start small.

00:17:30 Speaker_02
What we really mean by starting small is limiting your expectations. We might say starting small is about finding the best ROI fit, that's good use of the tool, finding the right pattern, that's a good fit. All that stuff is true.

00:17:44 Speaker_02
But really, what starting small is, is saying, okay, think big is, give me your big idea. Start small is, let's rein in those expectations.

00:17:53 Speaker_02
Let's sort of lower, not lower the expectations, let's just moderate those expectations to something that's more realistic. Like, I get your big picture. Your big picture's fantastic. I love that big idea.

00:18:05 Speaker_02
But starting small is all about, one, scope management and figuring out the best appropriate thing and doing things in the right order, but it's also about Reducing your expectations. Lowering expectations? I don't know. Reducing your expectations.

00:18:18 Speaker_02
Limiting your expectations. Because if you can do that, then the chance of over-promise and under-deliver is much lower. We're not over-promising. We're sort of right-sizing that promise, and we're right-sizing the delivery.

00:18:28 Speaker_02
Maybe the over-promise is still there in the big idea, the think big, but maybe we don't have to commit to that thing. This is where we get into problems. It's when people say, think big, now commit to that big idea. I want a self-flying plane.

00:18:40 Speaker_02
I want a self-driving car. I want AI systems that can diagnose everything in the radiology imagery. I want AI systems that can generate the best music, that can rival this and rival that. When you hear that, you're thinking, OK, that's your plan.

00:18:53 Speaker_02
That's your think big vision. But what is your start small execution? And if you don't have a start small execution, of course, that's a recipe for failure. And we have recipes for success. And it's never about overpromising. at all.

00:19:05 Speaker_01
Exactly. And I think that when you, you know, follow a step-by-step approach, you really understand what needs to go into this.

00:19:11 Speaker_01
We also say when you are figuring out, you know, if AI is good, is a good fit for your problem or not, look at the seven patterns of AI, which we will link to in the show notes. We have a lot of podcasts on this as well.

00:19:26 Speaker_01
because it helps you understand where AI is a good fit, where it's not.

00:19:30 Speaker_01
That's also in phase one of the CPM AI methodology, which is the Cognitive Project Management for AI methodology, which is that framework, that step-by-step approach for how to run and manage AI projects.

00:19:41 Speaker_01
It walks you through six different phases for each iteration and also helps you answer a number of different questions and go through things in a logical approach that you're not kind of jumping around and say, well, everything got done, but it maybe didn't all get done.

00:19:55 Speaker_01
in the right order or the right time frame. Because we've seen a lot of AI projects that one iteration can take up to a year. And we're like, wow, do you know what happens in a year? A lot. A lot changes in a year. So this should not be taking a year.

00:20:11 Speaker_01
That's also not starting small. That's like thinking big and starting big and overscoping and you are not going to succeed that way.

00:20:19 Speaker_01
So if you'd like to learn more about the CPM AI methodology and certification, you can go to Cognolitica.com slash CPM AI. And I will also link to it in the show notes. A lot of our listeners have become CPM AI certified.

00:20:32 Speaker_01
And so we would love for additional listeners to join our, you know, very growing community, global, you know, community of folks that are looking to run AI projects with a best practices approach. This really helps your career.

00:20:46 Speaker_01
It helps your knowledge. of AI and the entire space and really how to run and manage AI projects. Many project professionals have been taking this to enhance their career, you know, help get them maybe employed at other organizations.

00:20:59 Speaker_01
So definitely check it out. It's Cognolitica.com slash CPMAI and I'll link to that in the show notes as well. Also subscribe to AI Today if you haven't done so already and reach out to us. We love to hear from our listeners.

00:21:08 Speaker_01
We also love when you review our podcasts and we love five-star reviews. Like this episode and want to hear more? With hundreds of episodes and over 3 million downloads, check out more AI Today podcasts at aitoday.live.

00:21:22 Speaker_01
Make sure to subscribe to AI Today if you haven't already on Apple Podcasts, Spotify, Stitcher, Google, Amazon, or your favorite podcast platform. Want to dive deeper and get resources to drive your AI efforts further?

00:21:35 Speaker_01
We've put together a carefully curated collection of resources and tools. Handcrafted for you, our listeners, to expand your knowledge, dive deeper into the world of AI, and provide you with the essential resources you need.

00:21:47 Speaker_01
Check it out at aitoday.live slash list. This sound recording and its contents are copyright by Cognolitica. All rights reserved. Music by Matsu Gravas. As always, thanks for listening to AI Today, and we'll catch you at the next podcast.