AI Mismatch: Today’s Hype vs. Business Impact

Playback speed

Share post at current time

Share from 0:00

0:00

Transcript

AI Mismatch: Today’s Hype vs. Business Impact

What the history of machine vision can tell us about the future of LLMs

David Rogers

Jun 06, 2024

Transcript

It’s been a busy month of travel!

I look forward to sharing insights with you from my recent conversations with CEOs, startup founders, and government ministers in Dubai, Madison, Tokyo, and New York.

But for this week’s issue…

I wanted to share a recent conversation on AI that I had with Tarja Huuskonen, on her “Leader’s Agenda” show.

As happens, the topic of AI came up as we discussed digital transformation and leadership.

In the video, Tarja and I discuss:

A brief history of artificial intelligence and the feedback loop of machine learning
The evolution of predictive AI, from spotting muffins to solving real business problems
The gap between where the hype and the economic value of AI lie today
How smart leaders are thinking about Generative AI
How to take a “test and learn” approach to AI within your own business

☝️ Click the video above to start watching.

TRANSCRIPT:

[00:00:00] Tarja Huuskonen: What do you say about AI? Is it changing everything? What do you make out of all the…

[00:00:11] David Rogers: Artificial intelligence is a very broad term, and it started as sort of an academic category and a field of study. But it's now become, really particularly in the last dozen years, or maybe even 15, because of the development, because of the breakthrough…

[00:00:30] And people have been thinking about learning systems, and developing versions of them for decades, right?

But what's happened is one approach, very broadly, to artificial intelligence, to creating systems that simulate human decision making and cognition and so forth, is “machine learning.” And that is creating things like an algorithm that actually trains itself.

[00:00:52] Rather than… Classic example: a mathematician creates a model to forecast consumer demand for certain products, or some other thing in your industry. Then, what happens? The next year goes by. You get new data. The model doesn't quite perform. And so a human statistician, probably the one who created the model, looks at the latest data. And they say, “Oh, you know what, we should tweak the weightings” Or “We should add this other variable that would have been a little more predictive this year, if we'd done that.”

With a machine learning system, the idea is the model–the algorithm, the formula–trains itself. So actually, you feed the data back in: “Well, that was the prediction. Here's what actually happened" with crop production or consumer demand.” And so, it tweaks its own weightings.

[00:01:37] So that idea has been around for decades. But we had this breakthrough in what's called artificial neural networks.

They are not neural networks to be clear; it's a metaphor. But it is a technical way of creating computing systems, using these different layers, and it turned out to be extremely powerful.

[00:01:58] So we had this breakthrough, and that has led to lots of different technologies and approaches that have built on this artificial neural network approach to machine learning. So right now, we're at this really interesting place.

So first we had, if you look back 12 years or so, we had companies announcing, “Wow, this new AI can play chess” or “It can play Jeopardy,” or “It can identify what is a mouse and what is a muffin in photographs from the internet,” or “It can spot cats in YouTube videos.”

[00:02:30] And that was great from a scientific point of view. It's like, “Wow, it's breathtaking that the system can do this!” But okay, so what? That was sort of the “gee, wow” phase.

Now we've had a dozen years. These systems are being used in every industry. This is what I would broadly call “predictive AI.”

[00:02:50] So you're looking at data and you're making predictions and anticipating what's next. What's the right forecast? What's the message that's going to create a greater response, higher response rate from the stakeholders we are emailing? There are all kinds of applications.

[00:03:10] Predicting… identifying from images: “machine vision.” We've got drones that fly over public infrastructure and identify what's a crack you should worry about and what's one you don't need to.

There's a million applications being used. And you talked about drug development, incredible applications in the science, and gene folding, and a lot of other applications using these kinds of models.

[00:03:28] So at this point, it's all kind of in the background, it's all very enterprise. It's not on your phone, so most people don't think about it. It's been having a big impact every year, bigger and bigger economically.

Now along comes the next breakthrough. It's, broadly speaking, called “generative AI.” At the bottom level, we're still using artificial neural networks, right? (An approach to machine learning.)

[00:03:48] But, a really interesting novel approach, this idea of large language models. And it's really exciting and fascinating and hard to get our heads around why it works sometimes and completely fails in others. But it is emerging all sorts of capabilities.

[00:04:09] But right now, there's this weird gap between the hype and the reality. So right now, 99% of the hype about AI is about generative AI. Everything's about the chat bot, and the Dall-E, and the image generator, and there's now a YouTube channel on that's creating news every day, but it's all LLM-generated, it's all just generative AI images and avatars reading you the news, and so forth.

[00:04:38] The real business value is like 99% from the prior wave of AI, from the “predictive” artificial intelligence.

[With generative AI], in almost every industry, we are still playing around with it, which is great. The “test and learn” phase.

[00:04:57] Like, “How would we use this generative AI in my job?” Not like a cool demonstration. Not the equivalent of “show me the cat in the video.”

But, “Okay, how can I actually take one of the things I have to do every week in my job and do it much better and faster with this as a tool? We're figuring that out.

[00:05:15] I have no doubt that within the next couple of years, we're going to start to see big value created in many different sectors. But I know that in almost every case, we don't know yet what that's going to be.

So, it's a really interesting time. And I'm very interested to see what comes next, but that's my take on where we are right now in the era of artificial intelligence.

[00:05:39] Tarja Huuskonen: Yeah. So what do you tell to the business leader who wants to be ready for what comes next in AI?

[00:05:47] David Rogers: Well, I actually wrote an article on my Substack newsletter, David Rogers on Digital. My website is www.davidrogers.digital, and you can click there to subscribe. It's a free newsletter. I write every week.

And I wrote one recently called, “ChatGPT is not a strategy.”

[00:06:00] And it started with that point: that it's not a strategy.

Focus on the problems you want to solve, right? That's where I see the companies who are doing this effectively…

They take preexisting problems: customer, stakeholder, internal issues. Something that doesn't work, that takes too much time, that's ineffective, that's unpredictable, whatever.

[00:06:26] And they ask, “Could we use one of these new tools to improve this?”

And the advice I lay out in that article is actually – How do you do that as a process. How do you set up people inside your company right now to be experimenting? Really low stakes, low cost experiments.

This year, 2024 should be the year of learning for generative AI in almost every organization.

[00:06:48] Don't be thinking, “Well, we're going to get a big financial payback or windfall in 2024.”

But it's a great opportunity right now to start figuring out “Where could this really start to fit into our business, and really make more and more of a powerful difference in the months and few years ahead?”