Jacob Ridley, senior hardware editor
This week I was at CES 2024: A week of stomping around Vegas makes my feet hurt, but it was worth it to see so many new products for the first time. This year saw an overwhelming number of products with AI capabilities, and I'm wondering what to make of it all.
Another CES, another buzzword. In recent years it has been blockchain, metaverse and putting the word 'smart' in front of everything, but you don't win anything if you guess what it is for 2024. Artificial intelligence is the talk of the town and there was almost no escape from it during my travels to the tech show in Las Vegas.
It's easy to understand why AI is taking the world by storm. The rapid success of ChatGPT and image generation has the whole world watching. Now most major tech companies are thinking about how to integrate AI into their products, such as Microsoft and Copilot for Windows or Intel with the latest Meteor Lake Core Ultra processors.
There is money to be made with AI. Nvidia is a trillion-dollar company thanks to its GPUs' propensity to boost AI. ChatGPT's creator, OpenAI, has received billions in funding. These companies are bringing something truly new to the table, but not all AI features are so obviously brand new.
Some of what I saw advertised as AI at CES 2024 looks like it could have been called a smart feature before the AI hype peaked, including sleep monitoring, automated timers, head tracking, and automatic mode switching.
Many of these features existed before the AI boom. Is there something going on beneath the surface that makes any of them more deserving of the AI slogan, or is this tricky marketing?
What distinguishes AI from other, somewhat intelligent functions is in many cases not entirely clear.
With some AI products you'll find explicit references to training and learning, often with integration into a cloud-based system that does most of the actual AI processing. Adobe, for example, offers AI algorithms that handle functions that regular smart algorithms can't, such as creating generative images. Admittedly, I still think that the standard magic eraser is often more effective than the newer background removal, but I'm sure this will change over time as the AI features can be improved through the deployment of new models that are trained based on new examples and feedback. It is this type of training and feedback that is beginning to separate AI from traditional code consisting of if statements and logic loops.
The story continues
The MSI monitor pictured above will also come with an app that can train the software to work with games other than League of Legends.
In some other cases, I'm struggling to define what it is about the AI products of 2024 that are supposedly different from pre-existing smart features. The exact definition of artificial intelligence varies, especially in an age of AI and as new models and systems emerge. Although there are a few important terms to note.
The first is artificial general intelligence. This is defined as a system that is generally as smart or smarter than a human; it can achieve similar results to a human, if not better, in a wide range of tasks. That's different from how a bot designed to play Go could surpass a human in its capabilities today, as the bot is highly specialized. Most consider a true AI to have the ability to outsmart a human in many ways, through self-learning and decision-making.
We're still nowhere near common information, although you'll hear companies like OpenAI talk about their attempts to achieve this often. Once a system claims to be as intelligent as a human, there is a good chance that this form of AI will become the only form of AI worthy of the name.
That raises the interesting question of whether what we consider AI is changing as AI becomes smarter. I think that's true to some extent. The various chatbots launched before ChatGPT already seem quite formulaic compared to newer versions, and I doubt if/when we reach general intelligence for AI, we will still classify the current ChatGPT bot as artificial intelligence. Still, today it's probably the best we have, at least among the commonly available models.
Stanford University's Institute for Human-Centered Artificial Intelligence publishes a list of AI definitions [PDF warning] including one by emeritus professor John McCarthy from 1955. McCarthy defines AI as 'the science and engineering of creating intelligent machines'. With this definition it is possible to define all kinds of technologies from the past decades that promote the advancement of AI. That's true, as AI wouldn't exist without decades of computer science, but Stanford also adds its own definition to clear things up.
"A lot of research has people programming machines to behave in smart ways, like playing chess, but today we're emphasizing machines that can learn, at least somewhat like humans do."
What we are looking for is a machine that can somehow match and learn with a human.
The human part of that definition doesn't quite apply to the AI PCs and devices at CES 2024. These operate under a slightly different definition. One that refers to learning and automation - what many would describe as machine learning.
Most definitions consider machine learning to be a subset of AI, and while they are related, they are crucially different. Machine learning is about analyzing data or images, recognizing patterns and using them to improve itself. The kinds of things you'll see applied to AI noise reduction software.
As Microsoft outlines, machine learning and AI go together to create a truly useful machine. This machine learning lays the foundation for understanding a device, and the AI uses that to make decisions. You can see these kinds of symbiotic relationships in computer vision, where machine learning is used to teach a car about the dangers on the road, but the work of braking before crashing into a cow - imitating a human response - amounts to AI.
The difficulty lies in how these definitions are then applied to products and how we apply them historically.
Take the MSI Prestige 16 AI Evo for example. This is a laptop equipped with MSI's "AI Engine," which "detects user scenarios and automatically adjusts hardware settings to achieve the best performance." My question is: how does this AI compare to, for example, Nvidia's GPU Boost algorithm? GPU boost uses various sensors on a graphics card to automatically improve performance where thermal and power constraints allow, but that feature has been around since 2012 and no one calls it AI, and for good reason.
There's certainly some overlap between intelligent features that respond to human input and features that learn from it, and I'd be interested to know if MSI's profile switching feature actually learns from a user's actions to improve over time to improve and that is what defines it as AI. That seems to be the best way to distinguish between artificially intelligent functions and regular smart functions.
Your next machine
Best gaming PC: The best pre-built machines.
Best gaming laptop: Great devices for mobile gaming.
Moreover, AI is not entirely a local technology. These AI functions are largely enabled by massive data centers, and you could say that the AI revolution is riding the wave of cloud computing. We're seeing more AI processing power locally, like on your own PC - Nvidia GPUs come with tons of AI processing cores and Intel's latest Core Ultra chips feature an AI-accelerating NPU - but many of today's AI features require an Internet connection. So that further muddies the waters of what is actually an AI product and what is simply tapping into an AI service elsewhere.
The various subsections of AI are difficult to define by their nature, but make it even more difficult to pinpoint the true use cases of AI. The only practical conclusion I can draw is that stamping the all-encompassing term "AI" on a product may as well be meaningless for the many and varied applications of what is today classified as AI and machine learning.
If you're in the market for a new laptop, processor, toaster, or pillow, try to ignore the noise and pay attention to the really effective features you'll actually use. The really must-have AI and machine learning features should be pretty obvious.
