In Celebration of My Dumb Kettle
Why Most Things Shouldn’t Have Embedded AI
Last year, I made over $5,000 from YouTube videos showing people how to reset their kitchen appliances.
That’s not a flex — it’s actually evidence of a major problem. Even formerly simple devices have become incredibly complex, with massive numbers of modes, menus, and “smart” features.
These devices brim with failure-prone sensors, chips, and machine-learning features. And with the rise of generative AI, things are about to get a whole lot worse — and more dangerous.
Power Cycle Your Toaster
Recently, I went to toast a bagel and noticed that the screen on my fancy Cuisinart smart toaster had frozen.
I spent the next several minutes figuring out how to properly reset it (a simple pull of the plug won’t do — it somehow manages to reboot into the same frozen state.)
I finally remembered the proper procedure — which requires holding a knob for a specific amount of time, waiting for the toaster’s firmware to reboot, and navigating back to a home menu — and was able to return, a bit defeated, to making breakfast.
The idea of rebooting your toaster feels like a bad joke from the 1990s. But as the cost of computer chips has declined, the types of sensors available to designers have proliferated, and wireless connectivity has gotten cheaper to build into even simple gadgets, so-called “smart” appliances have taken off.
According to Statista, smart appliances will be an $11.2 billion industry in 2024. By 2028, more than 45 million Americans are expected to use smart appliances in their homes daily.
Smart tech is already here, if you know where to look. My Amazon smart microwave connects to the Internet and responds to my voice. My Aeris kitchen air purifier automatically detects when I’ve burned something on the stove, ramping up its power and clearing the air automatically.
My Nest thermostat knows when I’m home, adjusting the heat to match my exact preferences (and prioritizing my dog Lance’s preferences when I’m out.)
The Thinking Washer
And that’s only the beginning. Today’s smart appliances often integrate embedded machine-learning algorithms, which their manufacturers market as “AI.” These algorithms give them the ability to add a semblance of thought and analysis to their otherwise mundane jobs.
My LG washer, for example, uses sensors to detect the exact types of fabric I’ve put into it, adjusting the cycle time and temperature accordingly. It then communicates that information to its colleague, my LG ThinQ dryer. When I go to dry a load of laundry, the dryer is already aware of what it will contain and can adjust its heat level to avoid melting my pants.
Many new appliances take things even further. Samsung’s Family Hub fridge reportedly uses internal cameras and AI to detect what you’ve placed in it. It then suggests recipes for dinner based on what you have in stock.
I recently tested an AI-powered oven from LG that uses machine learning to determine exactly when your roast chicken has perfectly crispy skin, sending you an alert on your phone that it’s time to take it out and eat it.
Sprinting for the Edge
These features are neat party tricks. But in most cases, the smart features marketed as “AI-powered” today use fairly simple, early-generation machine learning algorithms to perform their apparent magic.
My smart washer, for example, uses an infrared sensor to determine the reflectivity of my clothes (which likely gives it a rough idea of their composition), and a weight sensor embedded in the drum to figure out the relative weight of the fabrics I’ve placed into it.
It then performs a simple lookup function, comparing the parameters of my load against 11,000 pre-saved profiles embedded in its memory. When it finds the closest match from its library, it grabs the pre-saved settings from that match best, and runs a wash cycle with them.
It’s helpful and impressive, yes. But the washer isn’t really using AI. Instead it’s primarily using clever sensors and a very basic algorithm — in this case, simple pattern matching — to simulate deeper intelligence.
As generative AI takes off, though, appliances are going to get genuinely smart, and fast.
In late 2023, Google announced its new Gemini Large Language Model (LLM), a competitor to OpenAI’s ubiquitous GPT models. Gemini Ultra got all the press, but Google also announced a potentially far more radical version of their generative AI tech — Gemini Nano.
Nano is a full LLM, but it’s optimized to run on a mobile device. Currently, LLMs are so complex and computationally expensive to run that they’re relegated to giant banks of servers in massive, remote data sensors. That limits their reach.
Efficient models like Gemini Nano, though, would allow LLMs to run on far simpler hardware. In the parlance of computer engineering, they move LLMs towards the “edge,” allowing them to run locally on people’s own devices, rather than in huge server farms.
Currently, Gemini Nano still requires a fairly capable smartphone. But as LLMs get more efficient, their move towards the edge will accelerate, and they’ll be able to run on progressively cheaper, simpler hardware. Ultimately, these models will reach the point where they can run on the simple chips powering smart appliances.
When that happens, your toaster, blender, and oven won’t just perform simple machine-learning tricks. Like today’s advanced chatbots, they’ll be able to analyze, reason, and communicate about the tasks they’re performing.
In short, your toaster will be able to think.
The Dangers Lurking In Your Smart Kitchen
This bodes well for my YouTube channel — more complexity means more things to go wrong, and thus more need for smart toaster troubleshooting videos!
But it’s not necessarily a great thing for consumers. AI-powered appliances are great when they work. But when they fail, the added complexity of these systems makes troubleshooting them difficult or impossible.
Consider the difference between a 1985 Subaru and a 2023 Tesla. A hobbyist could readily service and even repair the former at home. The latter requires specialized equipment and procedures that are beyond the abilities of even many professional mechanics. Complexity adds power, but also makes maintaining our devices harder and more expensive.
And in some cases, the move towards generative AI-powered appliances won’t just be inconvenient — it will be dangerous.
Today’s generative AI models are black boxes — even their creators rarely understand how they work. Because of the ways they’re trained, generative AI systems like LLMs often embed and reflect the biases of the people creating them.
Evidence already shows that self-driving cars, for example, have a harder time detecting pedestrians with darker-colored skin, potentially putting them at great risk for accidents.
A biased toaster probably can’t cause much damage. But imagine when LLMs — with all their embedded biases — end up in smart appliances like home security cameras or alarm systems.
A smart home camera that preferentially sends alerts when it detects people with dark skin color would automatically perpetuate biases that cause real harm and risk for many communities.
At an even more basic level, the unpredictability of generative AI models poses risks.
If my smoke detector senses smoke, I want it to immediately sound an alarm. I don’t want it to go “Hmm, I sense smoke. But based on Tom’s calendar, I see that he has a big presentation in the morning and needs his sleep. Best not to wake him!”
Thinking machines have their place. But sometimes, you need deterministic code that always follows the same procedure, no matter what its internal “brain” is saying.
My Dumb Kettle
Luckily, there’s an easy way to combat this generative AI-powered feature creep: opt out.
I recently decided to buy an electric kettle for boiling tea. The model that I settled on is dead simple. You fill it with water and press down a little lever to switch it on. It heats up, and when the water inside boils, the lever snaps back up with a satisfying click, switching the kettle off.
My kettle is decidedly, unapologetically dumb. There are no smart features, no onboard AI algorithms, and no complex, embedded sensors.
It’s quickly become one of my favorite appliances.
Before you protest and say “Surely, no one has added smart features to something as simple as a kettle!” let me assure you that they have. I recently reviewed the Cuisinart PerfecTemp smart kettle, which has nine buttons, six heat settings, LED indicator lights, and even embedded memory, all in service of performing the simple task of boiling water.
I didn’t buy this kettle. I’m sure it works great for people who want an intense degree of control over the exact temperature of their water, or just love things with pointless blue LEDs.
But for me, the simplicity and ease of use of my dumb kettle is its most appealing feature. You switch it on, and a few minutes later there’s boiling water. You don’t have to fiddle with menus or make decisions. I’ll never have to power cycle or reboot it.
Opt Into Dumb
More appliances should be like this — simple, durable, and dumb.
Don’t get me wrong — I’m no Luddite (nor is John Dutton!). I love my pseudo-AI-powered washer and dryer; the ability to throw in a ton of laundry and know that it will come out perfectly washed and never shrunk or discolored is delightful. Even my voice-powered microwave has been a big boon for family members with limited vision.
But not every appliance benefits from smart features. And most devices certainly won’t benefit from the onslaught of embedded generative AI that’s right around the corner.
As that tsunami of LLM-powered devices arrives, I urge you as a consumer to take a step back and resist it for a moment.
Desperate to compete with each other and to appear as innovative as possible (as well as to goose their margins with expensive add-ons), brands will soon embed AI into absolutely everything — and try to convince you that these AI features will make your life easier or better.
In most cases, they won’t.
Generative AI has its place. But before you buy that LLM-powered blender, pause to think about the complexity and potential risk you’re introducing into your life.
In the case of live-protecting appliances, what biases might you be unknowingly perpetuating, and what risks are you introducing? What steps is the company behind your smart appliance taking to protect your privacy and safeguard the data it’s gathering? Will your complex new smart gadget fail in a month?
In short, in the face of an explosion of generative AI smart features, ask yourself: is the embedded intelligence of this new smart device really worth it? Or would you be better off with a device that’s simple, reliable — and as dumb as my kettle?
I’ve tested thousands of ChatGPT prompts over the last year. As a full-time creator, there are a handful I come back to every day. I compiled them into a free guide, 7 Enormously Useful ChatGPT Prompts For Creators. Grab a copy today!