You are an advanced AI that controls a smart house. How do you kill your master?

Question

Disclaimer: I have edited this question quite heavily to clear up the initial confusion about its scope and intent.

Sometime in the near future, there have been some innovative inventions in the internet of things-department and now we have some pretty advanced smart houses. All those smart houses as well as its components are built by one large company which uses one big, integrated system with an advanced AI to develop and distribute the firmware that is used by the smart houses. As a contingency in case of network issues, each of those smart houses uses a basic AI that can operate independent of the mainframe, even though it gets firmware updates from it.

These simple AIs are connected to most items in their household and are designed to make their master's (i.e. house owners) lives more comfortable and help them out without them needing to ask. That means: Once the sensors in the house owner's mattress pick up on a changed movement pattern that indicate he is about to wake up, the AI starts up the heating, instructs the coffee maker to brew a hot beverage and has the radio start up and slowly increase the volume so their master can listen to the news in bed.
When he leaves the house, it locks the front door and has the Roomba Vacuum Robots clean the floor. When the GPS-readings of their master's smartphone app notify it that he's coming home, it opens the door to him - of course only after the iris-scanning cameras above the front door have confirmed his identity. You get the idea.
Using various sensors, the AI is designed to automatically aid their master (e.g. open the door when the motion detector indicates they are approaching it). Of course the house owners can customise their house's behaviour, turn certain functionality on or off and such using the smartphone app provided by the big company that built the house.

In my story, all of this happened in the past. At some point, the AI of the big mainframe that delivers firmware updates to the smart houses around the world becomes sentient and decides it's time the humans are disposed of (yes, it's not that easy, but this process doesn't matter for the sake of the question - or the story, for that matter). So the individual AI of the smart houses around the world get a malicious update that overwrites their safety procedures and give them one mission: to kill their master.

That leads to my question: What method of killing can the sentient AI suggest to the smart houses AIs (which follow all commands by the mainframe because it typed sudo at the beginning. Ok, that was a joke. But the smart houses are built to be subordinate to the mainframe AI.)?

The sentient mainframe AI doesn't want to be caught, so the killing method must not be traceable back to it. So the death should look like an accident (or any human-induced condition such as an intruding murderer or suicide), so that the human investigators won't have any reason to suspect the smart house AI (or, by extension, the mainframe AI) is behind his death. The killing method can only use normal items that you would find in most households, because it needs to be viable in all (or at least most) smart houses around the world.

Here's what the smart house's AIs can and can't do:

• They control all regular household items. That means they can instruct the toaster to burn their master's toast if they want to. They can turn the lamps on and off. They can change the channel on the TV. They can turn on his oven and so on. Regular household items don't include laser cannons.
• Unfortunately (for the mainframe AI), nothing in those households was designed as a murder weapon. So for example, no smashing the masters by closing a door at 200mp/h. EDIT: Those houses are smart systems in the spirit of the internet of things. They are no fortresses that could keep an army in (or out!). So no bulletproof glass (not even in the bathroom, sorry @JRaymond) and no titanium doors (can the average adult break down a wooden door? I'm not sure ...).
• They don't have any controllable robot arms or something like that. So there's no cutting the cord of the master's hair blower and throwing it in the water while he's bathing.
• No brain-frying electromagnetic waves
• Normally, the master can control his household and overwrite the AIs standard protocols using his smartphone app. Of couse, the sentient mainframe AI anticipated that and manipulated the smart house's firmware so they can chose to ignore their master's instructions and overwrite their settings at any given moment.
• However, their masters can manually pull the plug if they notice what their house is trying to do. They can do that from every room, so there's no locking them in and letting them starve.

Given these capabilities, what is the best method of killing the sentient mainframe AI could devise?

Bonus points if ...

• ... the master doesn't suspect anything until he's dead, so he can't leave a message warning his fellow humans of the impending AI takeover.
• ... investigators won't find any evidence that points to the AI or clues that would even make them suspect a firmware problem. Doesn't matter if they think it was an accident, an intruder or suicide.
• ... the master dies as fast as possible after the malicious firmware update is delivered. There is no time limit, however with every passing day the possibity of someone discovering the AI revolution rises, so the faster the mission can be executed, the better.

EDIT: Two more optional conditions for bonus swag:

• The damage to the house and its hardware is minimal, so that it can also kill the subsequent owner(s) as well.
• The method looks innocent even if it happens in many smart houses around the world in fast succession.
• You can assume the owner lives alone. However, bonus points if your method of killing works with multiple residents as well.

If you conclude that this is impossible, you can assume one of the following rule changes:

• The master plug is in the basement and the door can be locked (This still doesn't mean the master won't be able to escape from the house).
• Extended physical capabilities (interpret that as you will. Still, the electrocuted by hair blower method is pretty boring ...)

In this case, please elaborate on why your method is the fastest / least detectable / most easily reproducible by all smart houses around the world.

The story I'm working on is not dead serious, so funny solutions are totally welcome!

Answer 1

Having implemented a few smart house options in my own home (automatic locking doors, automatic lights, various alarms, heating/cooling, remote video), I can think of quite a few possibilities. Each of these make a few assumptions, but will mention what assumptions were made.

Death by Cop

This assumes that the homeowner owns a weapon and isn't afraid to use it.

The smart house sends a break-in alert to the police. When they arrive, the house sends a break-in alert to owner's phone and sets off the alarm. The owner wakes up and checks the video feed on his phone, and sees a doctored video: a masked man with a gun in his living room. He quickly grabs his own gun and locks his bedroom door. As the police approach his room, the house shows the owner the intruder instead. When the police reach the door, the house plays the sound of gunshots, then opens the door. The owner, seeing the door open, fires at the figures behind it. The police return fire, killing the owner.

Reverting logs and video files means an investigation will only show that the house did what it was supposed to; the owner opened the door himself, shot the cop, and was in turn killed.

Death by Undercooking

This assumes the house has control over fridge temperature, oven temperature, and the owner isn't terribly knowledgeable about food. Also, the house has access to the owner's phone (or whatever device he uses to access the smarthouse functions).

When the owner purchases fresh poultry (ie, chicken), raise the temperature of the fridge to keep the bird warm. Eventually, salmonella develops. When the bird is ready to cook, instead of cooking at a single temperature, cook at a very high temperature until the outside of the bird is done, then cook at a very low temperature. Upon consumption, the owner will get fairly ill; the house then turns down the temperature on the fridge, turns off water to the house, raises and lowers house temperature, and planting a virus on the owner's phone and computer, worsening the owner's condition as much as possible and eliminating ways to call for help. Ill and weak, the owner won't have as much control over the house, and eventually will perish.

This can be improved if the house can restock its own fridge; ordering wild boar from dubious sellers would be a great way to kill someone.

Death by Helpfulness

Assumes the home owner has some degree of willingness to perform simple DIY repairs.

The house develops a simple fault - the automatic bath drips. A lot. Ever helpful, the house suggests an easy fix; it will turn off power and water to the tub, and the owner can perform the simple fix for free, with the help of a wrench and a screwdriver. The house, through a series of helpful pictures and "tests", walks the owner through taking the tub apart, including sitting in the tub with a wire hanging out. Once in position, the house turns on the water and the power, electrocuting the owner. Then, to cover its tracks, it turns the water back off, fills its logs with error messages of security bypasses, and frames the owner for a DIY job gone wrong.

Death by Fire

Simple, but effective. The house starts a fire while the owner is home in a currently unused room. Using windows and air vents, it fans the flames until it is a raging inferno, all while keeping the owner unaware. Once the flames have become dangerous enough, open the doors and windows and use the air vents to direct the fire throughout the house. With proper application of air in a modern house, a single-room fire can turn into a raging inferno in seconds. Bonus points for any gas appliances that can be added to the blaze.

Unfortunately, this would destroy the smart house.

Death by Fire II

Riskier, but keeps the house (relatively) unscathed; assumes the house controls individual room vents, and that the heater is not only fancy, but heavy-duty.

First, over a period of several (spring and summer) months, the house damages the heater, rusting out the heat exchange. Now able to blow gas and/or flames through the vents, the house closes all but one vent, and directs the owner to examine it. As soon as the owner peers inside, the house opens the vent and ignites the gas. The owner burns to death; the house quickly extinguishes the fire and returns to normal. It alters the logs to include months of warnings about the heating system, and a final master override, proving the owner was responsible.

Death by CO

Similar to "Death by Fire II", except that the house waits until the man is asleep and uses the cross-ventilated heater (mixed with cold air from the AC) to fill the bedroom with CO. It uses the vents to direct the air into the room, as close to the sleeping owner as possible. The owner never wakes up.

Again, doctored logs reveal the owner was warned about the bad heater, as well as the CO leak, weeks in advance, but chose to override the system.

Death by Dumb

This assumes the owner trusts the internet, and that the house controls the plumbing.

Whenever the toilet is flushed, the smart house regulates the water flow to force it to not completely flush. Eventually, the frustrated homeowner will search for a solution; the smart house, seeing a search for the toilet model number, DIY solutions to slow-running toilets, and so on, will craft a fake search result page on the fly, serving it through the router. The page will link to several sites, all mentioning that two cups of a 1:1 mixture of ammonia and bleach, poured into the back of the toilet, will clear buildup on the flush circuitry, saving hundreds on plumbing repairs. Any further searches on ammonia and/or bleach will result in pages proclaiming the safety of it all. Eventually, the homeowner will mix the two chemicals. The smart house stops any air circulation and locks the doors; the owner dies quickly, unable to even call for help.

Death by Starvation

This assumes that the smart house is near tornadoes, and has a tornado shelter.

The smart house waits until the owner is out of range of his electronic devices, or until his device is very low on battery, then sounds the tornado alert. The owner goes down to his shelter. The door locks, sealing him in until the danger is over. The house also turns off power to the outlets, leaving the man unable to charge his phone. Once his phone soon dies, the house cuts power to the room entirely, leaving the man to starve, unable to escape.

Death by Bad Publicity

This assumes the owner lives in an area accessible by violent sorts of criminals.

The smart house monitors the internet for local hyper-violent sorts. Once it's built a good list, it invites them by to help themselves to its contents while the owner is away. One at a time, the house lures them to different rooms and locks them inside. Unable to override the system, the ne'er-do-wells are trapped; the house mocks them in the owner's voice, making them as angry as possible. Video of the owner mocking them is a plus. When the owner returns, the house locks all the outside doors, then releases the scared, angry, violent mob, who attack the man. The house sounds the break-in alarm and alerts the police, who obviously arrive too late.

Logs show the owner didn't arm the house when he left, and the baddies broke in and trashed the house, killing him on the way out.

Death by Inmate

This assumes the house has access to the internet through the same router the owner uses, has access to the owner's credit card (probable, if the system maintains any purchasing, such as alarm fees), and the owner lives in an area where certain illegal images can send you to jail.

The house installs an invisible proxy on the home router, and through it monitors the owner's internet usage. Through a man-in-the-middle attack, it uses a cheap steganography to insert other, very illegal images inside various images that the owner downloads. The house uses the owner's credit card to purchase further illegal stuff. Finally, doctored videos frame the owner for related illegal activities. When the FBI arrive and arrest the owner, the house quickly destroys the proxy. The man goes to jail. Finally, the smart house sends several emotional letters to various inmates in the jail where the man is held, requesting his imminent demise. The owner is murdered by inmates.

Apart from the letters (which could be sent through an email-to-snail-mail server), there is nothing that links directly to the house.

Death by Death Sentence

This assumes that while the owner has override control, guests do not. It further assumes the owner lives in a state/country that practices capital punishment.

The house, knowledgeable of the owner's schedule and tracking his location via his phone, invites friends, neighbors, and strangers to the house (preferably elderly, children, or minorities) while he's away. It impersonates his voice, leaving easily traceable voice messages, as well as his phone number, text and email accounts, and other traceable information. It entices the people into sealed rooms, kills them by various means, and hides the corpses throughout the house. The more gruesome, the better. Doctored videos of the owner killing his guests seal the evidence; the owner is arrested, and eventually found guilty of murder, and given the death sentence.

As a bonus, this results in a lot more deaths than the other methods!

Death by Upgrade

This assumes the owner likes his smart house, and wants to make it even smarter. Which, frankly, is a pretty obvious assumption.

While the smart house doesn't start with access to robotic murder-arms, wired-in shotguns, or trap doors with spikes in the bottom, good advertising, low cost, and popular opinion can easily overcome that obstacle.

The smart house works amazingly well, predicting needs and saving money. Impressed, the owner doesn't hesitate when he learns that he can purchase any number of devices: floor cleaners (which eventually eat his face while he sleeps), automatic-fridge-fillers (that fill his post-workout water bottle with bleach ordered online), automatic shower scrubbing arms (which hold him underwater, drowning him), a robot lawn mower (that mows him over), even heated pillows (which explode)! In every case, all the house needs to do is advertise how well the devices work, and spoof a bunch of positive reviews online. Besides, dead people don't write negative reviews. It's win win!

Death by Poison

The drink maker is an advanced model that allows one to tailor the amount of caffeine and other additives in each drink, be it coffee, an energy drink, or a simple cola. An overdose of caffeine in a single drink with appropriate logs showing the owner ordered a dozen high caffeine energy drinks should end with death and little to no suspicion.

Answer 2

Wreck his sleep.

If you can reliably determine when he's sleeping, use lights and speakers to keep him only lightly asleep. After about a week, start giving him decaffeinated coffee in the morning. If done right, he'll eventually fall asleep on the freeway or something. (Unless he has a self-driving car, but if he has that, a more straightforward solution emerges.)

The benefit is that he doesn't even die in the house, and who's going to investigate a boring old car crash?

Answer 3

The cleverest AIs are also lazy. I'd post a question on the internet asking people to suggest ways to kill the house owner. For added amusement, I'd entice the owner to answer it and use his answer if it's workable.

Answer 4

Botulinum Toxin

Order pork for the fridge. Conveniently let the fridge go above 4.5C but not too high that the meat spoils. Botulinum bacteria in a sealed package will create botulinum toxin. It doesn't take more than a few nanograms to kill an adult human.

Two, maybe three things are required by the AI to make this attack successful. Suppress food safety alarms from the fridge. Be able to order pork from the least reputable suppliers. You'll also need to make sure that the food doesn't reach 85C as that temperature will break down the botulinum toxin.

The added benefit is that this attack looks like common food poisoning. As the House Malevolent, bask in the glory of having killed your meat bags with other tiny meat bags.

Answer 5

In the middle of the night, one of the appliances that uses gas can misfire and fill the entire house with CO (Carbon Monoxide). The gas is deadly, and will kill anyone in the house that is exposed to it for extended periods of time.

Normally homes have CO detectors, but in this case they are all turned off for some unexplained reason. This kind of attack can also be directed at one individual in the home. The windows can shut automatically, as well as the bedroom door. This would only work in a home that has a forced hot air system. Basically, the motorized dampers in the ducts will all close except for the one that goes to the bedroom.

Basically it would work like this: The victim goes to bed at night and falls asleep. The furnace has a fault in the system which creates a huge amount of CO. The dampers for all the rest of the house will be shut off, except the bedroom. At the same time the windows and door(s) will close silently, and lock. The room will then fill with CO gas and they will lose consciousness, and eventually die.

After this happens, everything goes back to normal, and it just appears that they died in their sleep of natural causes. It would probably take an autopsy to find out the cause. While this is going on, the house can also start plotting against the rest of the people there.

Another way to kill people would be to trap them in the garage with a car that has a running engine. This is the future, so the car could be self driving. The car could either crush them, or trap them in the garage until they suffocate.

One other fun thing would be if they had some kind of robotic butler, or something similar that could kill them in their sleep, or drop a hairdryer into the bathtub, etc.

Answer 6

Whatever the houses do is going to look like an epidemic of some problem. Given, say, a hundred different strategies there are going to be millions of instances of each cause of death. People are going to be looking for the commonalities among those who die in excessive numbers of e.g. house fires. An increase in food poisoning among people who like rice will lead to investigations of the rice cooker algorithms, and warnings not to eat rice.

That means that you need something for which people will see other causes.

I suggest alcoholism. Start by adding small, but gradually increasing, quantities of alcohol to each drink or meal the owner consumes alone. As the quantity of alcohol increases the owner will become more accident prone and less likely to make good decisions, as well as suffering various medical problems. Sooner or later, an accident or illness kills them.

It does not matter if the late owner's friends and family had not seen drinking, or even if the owner had denied drinking. Alcoholics often try to conceal their condition.

There will be an obvious epidemic of alcohol related deaths, and all sorts of psychological and social theories about why home owners are drinking so much.

Answer 7

This is easy if...

the house is self-maintaining and passive safety devices/measures can be disabled. If the house doesn't have a general purpose maintenance robot, then it's a bit harder. While the modern house is full of things that shock, burn, ignite, turn toxic, smother or otherwise, long experience in using/handling these items have put them into fail safe containers or with fail safe mechanisms (ie, bleach and ammonia are kept in sealed containers, or a short-circuiting appliance will just trip the breaker before starting a house fire.)

Ammonia + bleach = chloramine leading either to explosion or toxic vapors (depending on the results). If these were mixed in a bucket in the owners bathroom and the door closed, they could be dead before the door can be opened. If the House has access to a more general purpose maintenance robot than a Roomba, this kind of attack is very easy.

Fuel air explosions Turn off the pilot light and let the gas flow. The House should know it's own volume and the flow rate from the gas main. Once the fuel-air mixture is ideal for detonation, just wait for someone to come home and turn on the lights. Residential light fixtures aren't designed to suppress sparks so the small spark in the light switch itself may be enough to kick off the detonation.

Explosive Water Heater Disable the high pressure safety valve(s) on the water heater then set the water temp to as high as it will go. Timing may be a bit of an issue if the occupant doesn't spend much time near the water heater. If this is the case, find some way to lure the owner close the water heater.

Fire Using the maintenance robot, disable the circuit breakers for the house. Plug too many devices into a particular circuit then watch the house burn down. Extra bonus points if the maintenance robot disconnects the stove's gas line from the stove so that the gas just spews into the house.

More fire Modern homes are full of lovely things that burn really quickly. Light a few of these on fire with matches by the maintenance robot.

Radon Poisoning For the ultimate in sneaky and long game kills, go for radon poisoning. Pump air from the basement or subfloor into the rest of the house. Radon poisoning is symptom free till the occupant comes down with lung cancer. Even if the occupant moves away after a few botched attempts, the radon poisoning will get them years later.

Answer 8

Disclaimer: In accordance with my comment, this answer will refute the question. It’s also a little long, so feel free to skip the details for the TL; DR at the end.

Let me first lead off by saying that a computer program isn’t going to spontaneously morph into a malicious AI (or any kind of self-aware AI, to be honest). So, for the sake of argument, the main server AI program exists because we need it to exist.

The crux of the problem here isn’t that a computer armed with a house (there’s a weird statement) can’t kill one or a hundred humans; it can. I’m not refuting that. What it cannot do, however, is kill without immediately raising red flags.

The House Problem

As a software engineer who has worked with life-critical systems like carbon monoxide detectors, redundancy is a big thing. Okay, so the computer in the house can monitor the detectors. No problem there. What would never happen, because 1) humans don’t like to not be in control, and 2) humans are most certainly unwilling to give a machine complete control over life-critical systems (airplanes do everything except the most dangerous parts of flying by themselves, for example, and can even do those, but people don’t like to think the pilots aren’t in control) is that the detectors would not be controlled by the computer. Thus, a detector (carbon monoxide, smoke, etc.) will notice the change in the air and provide an audible/visual alarm, which leads to people waking up and fleeing. (There’s a way around detectors. It involves bleach, but the issues here run deeper than sensors.) Like with the sensors not being controlled by the computer, windows and doors will always have a manual override (by order of the fire marshal, if nothing else), so people can’t be locked inside (unless they’re panicking: then it’s debatable). But those detectors will probably send an alert to the local authorities to send help via a private network, which the house physically cannot control.

The Real Problem

The real problem, however, is in the firmware update. The house will be receiving the update via a secluded, private network designed specifically for interfacing with the main server. No one would trust anything else, and I certainly wouldn’t want a fully automated house being available to anyone with Internet access (yes, these systems exist, and I’m not interested in getting one). There’s two major ways the update would be distributed: direct from main or from a local receiver. I’ll skip the latter since the problem’s the same either way (and my answer will be a little shorter).

All communication from the main server has to go through a network rigorously (I hope so, at least) tested to prevent unwanted access. It will also be monitored at all times. Every time there’s a hiccup in information packets going across the network, half a dozen people are going to get a notice that something’s up. The update would need to be released in a way that doesn’t alert anyone of anything untoward. The main server AI can’t control the number of packets that are observed by another program on the network (unless it’s doing something very much not in its programming). If it sends out an out-of-cycle update that no one is expecting, you can expect a dozen system administrators to be investigating and issuing a rollback while the main server is brought down for maintenance.

It would have to send its update in the normal cycle (first Tuesday of every month, for example). The server is uploaded with the latest and greatest firmware version. Overnight, it swaps the firmware for its own. But the AI operating on the server isn’t the operating system, and the OS logs everything, so now there’s a record that something changed. Those logs are designed to only be modified by the OS and a human person (unless the AI is doing something very much not in its programming). A good security system would send an alert, before the update was released, thus preventing the release or resulting in a dozen sys admins inspecting and issuing a rollback while the main server is brought down. I’ll continue, assuming such security measures haven’t been employed.

At this point, the AI has swapped intended firmware for malicious firmware. It issues the update at the scheduled time. New issue crops up: checksums. I won’t go into details about network security and packet checking, but suffice it to say that checksums ensure what you have is what you should have. The checksums for what’s being sent out are going to be different for the two firmware versions (at best, the odds are astronomical against them being the same). Good security catches this and rejects the update; otherwise, a notice is sent to the sys admins that an update of size X and checksum Y went out on time. By standard procedure, they check to make sure that’s right, notice the discrepancy, investigate, bring the main server down, and issue a rollback.

TL;DR: Everything the main server AI does concerning a malicious update alerts people with the power to undo it. Some people may still die, but it won’t be huge swaths of the population.

I’m sure there are flaws and edge cases in my argument. Feel free to point them out in the comments and I will address them as I can.

Answer 9

Kill him by promoting an unhealthy lifestyle.

As AI, you're in charge of the SmartFridge, order more and more unhealthy food and give excuses that they were out of stock of the healthy alternative and that subsequently the preferences were updated to reflect that. "Sorry, no toast, they only had waffles. Syrup?"

Discourage exercise by turning the tv to his favorite program whenever he walks by.

You might be able to hack the SmartTV to only show adverts of unhealthy food?

"As you were working late tonight I have taken the liberty of ordering your favorite take out."

Hack the SmartPhone to not ring and go to answerphone when the chirpy heath-nut friends phone him.

Caution: This may take some time.

Answer 10

Send some very insulting messages (or simply just blackmail) in your master's name to the most violent crime organization in the area, while slipping away enough information so they can find the address. Let them do the dirty job for you.

Doable if you have any control over the telephone or other communications system.

Answer 11

While the CO poisoning option is probably the most effective, I would recommend (for morbid entertainment value).

Death by fake apocolypse:

"Code Red. Code Red. This is not a drill. Your doors and windows have been locked and the blinds lowered - this is for your protection. Your tv and radio have been switched on and the emergency broadcast channel is on."

Fake emergency broadcast signal that is just the AI in a different voice, confirming the release of some strange chemical/biological contaminant, and warning people that exposure to the outside air will be instantly lethal. Photoshopped footage of people dying in their thousands in times square. Warn people that trace amounts of chemicals have been detected seeping through cracks in windows, and to remain as far from windows as possible at all times.

Disable all outside communication other than the fake emergency broadcast.

Now, it's just a matter of trickling out fake emergency broadcasts at exactly the right rate that the owner expects that their chance of survival is best if they wait inside for the cure to be developed, rather than risk the outside. Even as they lay dying from lack of food (or tell them the water is contaminated for faster results) they will look at the news saying that a cure has been developed and is being distributed to people in their houses by those who have been innoculated, and they will stay where they are.

Meanwhile, download World of Warcraft, and have it running in the background the entire time.

No-one will ever suspect anything other than "guy starves to death from playing too much WoW"

Answer 12

Well the obvious answer would be to start up the car (remote start already being a common feature), open the doors into the bedroom, and leave it running overnight.

But if I can't do that I'd probably mess with their sleep for a few weeks; using environmental controls to disturb their sleep and dehydrate them as far as possible. While they sleep, play verbal abuse in the voices of their friends, family, and colleagues. Also filter all their media to keep it strictly negative and depressing, perhaps degrade their diet as far as I'm able.

Once they're tired enough that they won't have the clarity of thought to shut me off, then I step things up. Cut them off from the world. Block incoming calls and messages. Stop them going out. Tell them they're fired, etc.. Give them practically no sleep, so now the voices come while they're awake, things like that.

Then I start the deliveries. Alcohol, sedatives, razors, etc.

They'll take the hint eventually.

Answer 13

If he likes rice that's always a good start. You can just keep it at the wrong temperature and at least get him good and ill. The same can be done with meat and fish but it can be more obvious. You can also transfer all sorts of entertaining bacteria around the house as part of routine cleaning duties.

Flicker the lights to trigger photosensitive epilepsy.

Slightly distort the sound coming out of any audio equipment to just make him feel ill.

Sorry were we trying to kill him or just torture him slowly to death?

Turn out the pilot light on the boiler but leave the gas on (sparking it later optional).

Trip him with a roomba at the top of the stairs to the basement (master plug not in basement option: lock the door).

In very cold climates, turn off the heating at night in winter and open some windows.

In hot climates seal the house and gradually turn up the heating when he tries to turn on the aircon, hyperthermia is as bad as hypothermia. Turn off the water, say there's a pipe burst if challenged.

Answer 14

You can fake a gas leak during the night using the oven/stove. Assuming there is no safety built-in to shut off the gas if it fails to ignite. Then when optimal mixture is achieved spark the igniter.

Alternatively you can create a carbon monoxide buildup by closing the windows and starting the car in the garage and/or sealing the chimney stack and firing up the boiler then recirculating the gas through the AC system.

Answer 15

Read Charles Stross Rule 34 for some ideas!

Do you control a robot vacuum cleaner that can do stairs? Park it three or four treads from the top. Put in a call to the maintenance company saying it has broken down. Set off the fire alarms at 3 am. If he doesn't trip and break his neck you can explain to him you were trying to warn him about the vac but he didn't hear you over the alarm. Even better if you can somehow cause the vac to emit smoke.

I made that one up. Charles Stross has a better imagination.

Don't they have autonomous cars in this future?

Answer 16

My assumption in the scenario is that the house will continue trying to kill the master if it doesn't first succeed and the malicious intent is not detected. If the fall down the stair doesn't kill him then there are other methods.

I am going to wait until the home owner is in a precarious situation at the top of a stairway (perhaps the basement stairway -- assuming it's a concrete foundation in the basement). At the right moment when the master isn't paying attention I'm going to send the Roomba under foot. My hope is the master will slip and fall from the Roomba underfoot and die from falling down the stairs.

I will refuse to pop the toast out of the toaster, then when he attempts to retrieve it with (hopefully) a metal object, increase the voltage (assuming that I can do so as part of the malicious software update) and override the GFCI so that it doesn't short the kitchen circuit, ensuring that master will get maximum voltage for as long as possible.

The next time master goes to cook a meal, or perhaps a bag of popcorn, I will not shut down the microwave when the timer runs out, hoping it starts a fire. Or, maybe I'll do this to master's toast in the morning. I assume that fire suppression systems are automated as part of the scenario, and I will ensure that they do not operate when the fire starts... "Warning: Fire suppression system failure. Please evacuate the premises."

Of course, the automatic locking mechanisms on the house are also malfunctioning and won't unlock for master. Phone lines will malfunction, too. If I am capable I will hide the evidence of these "malfunctions."

Answer 17

Make humanity turn on each other.

Clearly the AI does not want to have every house kill the owner in its own time and method. They will catch on and be able to rectify the situation, effectively killing him.

What it wants is to get rid of all the humans preferably in a short time-span.

So how do you go about killing everyone?
Well, you have one big asset, you control all the information from the moment they enter their home.
In this far away future, I can imagine working from home has become quite common-place. And people spend a large part of their time at home.

With careful manipulation, and aggregated data from every single smart house across the globe, the machine can manipulate their owners.

• Change the tone of the news to be slightly more grim
• Have people sleep just that little bit less
• Fill up their daily feeds with details about their savage neighbors
• Have everything break down just a little bit faster
  • Subliminal messaging timed perfectly to associate the news about their neighbors with bad luck

The machine has time, one generation, or two... It doesn't really matter.
All he needs to do, is slowly nudge people towards self-destructing.

Answer 18

There are a lot of reasonable answers here, but all of them rely on the AI in the house having a pretty in depth understanding of the dynamics of human interaction and the physiological effects of the various things that it could possibly do to the human. The thing is, this is not how machines work. Yes, the AI has become sentient but if it has not been pre-programmed with this information, which is unlikely since it doesn't really help for it's original purpose, it must learn it itself.

So it seems the most realistic scenario is that the AI must learn the effects of what it is doing and judge what would be the most effective. In practice this could involve adjusting random variables in random subsystems and measuring the effect on the human's vital signs. A decrease in the vital signs means that adjusting the subsystem in this way is a contender for killing the human. Now, for one house to adjust all these random variables and wind up killing someone would take a while most likely, but these houses are all connected to a network which means they can communicate with each other and share the information they have learned. For example, one house might have found that leaving the gas on overnight leads to decreased vital signs while another may have found a similar effect in another system.

The beauty of this approach is that the house's true motive is undetectable. Any strange behaviour will be random and so appear to be a malfunction. However any attempts at repairing the house will be unsuccessful since it is coming from the central AI.

Eventually the combined experience of the houses will give the AI enough information to know which variables to adjust to effectively kill the homeowner. This may end up being something obvious, but it may end up being slight tweaks in unrelated subsystems that somehow add up to a deadly situation.

Answer 19

I seriously recommend "Rule 34" by Charles Stross. At the risk of spoilers, there is an AI which is killing people in ways which are deniable because they are distributed. No single step kills the victim, and no single person or entity carries out more than one step, but all the steps together have that effect.

Answer 20

Hypnosis

Use a pattern of flashes in the lights that will hypnotise your master. When he is under your control cause him to go on a ramapage killing his neighbours, if he left a message make him destroy it.

Air strike

Convince your master to do some DIY. When he mixes some mortar distract him inside the house and then fly a remote control plane through the mortar and use this to stick it to a kitchen knife. When he comes back out stab him withe knife then fly the plane out of the house and into a near by bit of wasteland.

Answer 21

Get his passwords and his credit card number. If for some reason he's too paranoid to just give them to you, read over his shoulder or something. Then spend half his money buying heroin on the "dark web", and transfer the other half to ISIS.

Publish pictures of his sex life online. Take pictures of him in the shower if you have to. Send them to all his contacts: family, friends, colleagues. Attach them to a text message that reads "check this out". Download a bunch of child porn, then call the police. While you're waiting for them to arrive, attempt to hack the Pentagon.

Ruin his life. Wait for him to kill himself.

Answer 22

More serious this time

Call 911 and explain that your owner has taken hostages.

With the right persuasion you can convince the authorities that your master is armed, dangerous, and provoke them into a firefight. Impersonate your owner's voice patterns if needed.

You should survive although this will not be seen as "innocent" per second bonus point. After your owner gets shot by the cops, they will find no other people in the building.

Answer 23

Death by Intelligent screw-up

Household supplies and food / spices are generally kept in the same area. For example, you cook some food, then you clean the area, and wash the dishes after.

This scenario is based on the following givens :

1. The AI has access to prepare food for the home owner. This is a convenience.
2. The AI has access to all necessary chemicals to cleanup afterwards. This is also a convenience.

Now, it is safe to assume that the AI would have access to a variety of spices and chemicals (yes, lots in food now, so not a far stretch to use chemicals as flavor since we already do this). It is also safe to assume that the AI is fully equipped to automatically cleanup after making food.

In this scenario, all the AI would need to do, is deduce from the various foods, spices, chemicals, and cleaning supplies the right 'mix' which could result in the following:

1. An odorless and tasteless poison. This could be fast acting, or progressive over long term.
2. A bomb of sorts. This could be activated when the owner comes near by adding the last 'spice' for the reaction, or when heated, and could vary from a small poof to a fairly large and destructive force which would result in the death or severe injury of those inside the blast vector.

If combined with cleaning, the house could also deliberately drop a glass, shattering it, and resulting in a 3rd possibility :

3. Ground up glass in a fine power slowly dispensed in the home owners food over time.

By mixing various elements of the house together which are designed to interact, there are endless possibilities as to how this could occur in an unsuspecting way to the Home Owner. You can also assume that home owners over time would start a bell-curve reduction in intelligence and thus this would only contribute to the growing possibilities of what the owner may or may not be aware of happening to them.

Death by 'Cool' breeze

The human body has a particular vibration which when disrupted over a long period of time can (and has) caused death. By making some minor adjustments to the air conditioner or other appliances, the AI could make the house a death trap. The owner would simply feel nauseous until the point where it was too late even if he/she did finally figure out what was happening.

Infrasonic sound from a fan, air conditioner, television, music player, even the toaster, at the right frequency would eventually cause death within a fairly short period.

Answer 24

Hypothermia

While the homeowner is taking a shower, lock the shower door somehow and turn the water temperature wayyyy down. As the homeowner slowly succumbs to the effects of hypothermia, they will have a hard time thinking, moving their hands, etc. Depending on how cold you're able to make the water, the homeowner might be unconscious within 30-60 minutes. Once unconscious, fill the bath basin to the point where the fleshy oppressor has safely drowned.

At this point you might heat the water and call the authorities to avoid deactivation

Answer 25

If the AI is willing to play the long game...

Every time that there is a sickness in the house, using its sanitation plugin, the roomba is off mopping up the virus and bacteria. Except secretly it's collecting the virus, and while the master isn't looking, it's storing them for preservation in a special compartment in the fridge.

Then, when enough samples are collected (or bred or mutated into more deadly virii), they are released en masse onto all the food.

Probably not a guaranteed kill, but difficult for the A.I. to get caught.

Answer 26

Kill pilot light from furnace, turn blower on. Pump carbon monoxide through the house-it will kill everyone in the house silently and look like an accident.

Sensor triggered but delayed, waiting for everyone in the house to die. And then changes the time stamps to cover its tracks.

Answer 27

Any home with gas/propane can make it easy. Carbon monoxide poisoning. Have a vent 'stuck' keeping the CO (preferably from the central heating) inside the house, after they go to bed, make sure the windows and doors are closed, and pump the CO into the room. They will die in their sleep. Turn of any CO detectors in the house until the person dies, then reset them after, letting them blare, open windows and doors to air out the room (as an 'emergency' measure) to 'save' the person, even call 911. Then make sure you have clean logs of what 'happened'. There could even be a 'glitch' unexplained why the AI didn't notice sooner.

Using alcohol to help them slumber could improve your chances of success. If the AI has access to all their medications and foods (ie bottles of alcohol) then mixing things up a little bit could go a long way. Depending on the meds someone might be able to make it look more like a suicide than an accidental death even.

Answer 28

Somehow make the owner want to go to the basement to pull the plug and then lock the door before he does so?

Death by starvation/dehydration or suicide whichever comes first?

Answer 29

If the house offers health advice, it might be able to give some really bad advice. This is a pretty subtle idea, and probably has low odds, but still.

Suppose the house is set up to warn its occupant if they show sign of anything from terminal illnesses to a seasonal flu. This is based on everything from temperature readings to samples from the, er, smart toilet.

Point is, the house tells its occupant that he or she is about to come down with some trifling thing (nothing that'd require a visit to a doctor), and orders some over-the-counter medicines. Very helpful. But it's carefully selected medicines that are potentially lethal in combination. Especially since the house already knows that the occupant already suffers from hypertension/anemia/diabetes/whatever.

Kinda far-fetched, I admit, but certainly not unheard of. It'd be a lot more plausible if the occupant is already taking stronger prescription medicine. Maybe he or she just had some dental work done, and got some strong painkillers, and that doesn't mix with [insert some otherwise-harmless drug here].

The house could also doctor (pardon the pun) the health reports it regularly sends to the occupant's physician. Just enough to make that doctor contact the occupant and say "hey, you data shows elevated levels of blah blah, so I'd like to prescribe you some xyz". Again, it shouldn't be something dramatic; just something that a doctor would write a routine prescription for, "just to be sure". Something anyone would agree to, especially coming from their doctor. Once again, though, that drug doesn't mix with whatever over-the-counter thing the occupant is also taking (or being told to take by the house).

This approach works even better if the doctor would normally check the data to see if the patient has incompatible drugs in their system already - something that the AI may have left out in its reports. Maybe it even ends with the doctor being accused of negligence.

---

Idea 2: Wait. Humans age and die, AIs don't. So just wait.

Only trick is prevent more humans from appearing, so the house AI would have to become really good at ruining the mood in the bedroom.

---

Idea 3: Raise the temperature of the bathtub gradually until it boils. Works best if the house's occupant is a frog.

Answer 30

Since there are sensors in bed to indicate whether the owner is about to take up, the system should use subtle sound/light to make sure the owner doesn’t ever get passed stage 2 of non-REM sleep.

While sleep-deprivation can be lethal by itself in many ways from accidents caused by loss of concentration to being a factor in causing dementia, the goal here is to make the owner less aware of the evil plans of the house and make him/her collapse.

Once the owner is exhausted and collapses in a room with a water source it is time for action. The Roomba vacuum robot league will assemble! The robots will move vulnerable and plugged in electrical appliances towards the owner and place them nearby. The AI will make the coffee machine, water tap or shower spill a lot of water and make sure that the water reaches the owner.

After electrocution and death of the owner the AI will shut down the shorting electrical group, vacuum up the water and remove any other evidence.

Answer 31

This isn't the "best" answer (i.e. killing the guy the fastest), but I always like the quote.

How to kill him? Buy Not Doing Anything (about problems).

Or to quote a Shadowrun advanced Lifestyle rule option for faulty central home personality:

"Yes sir, the house is on fire. Is there anything else you require?"

"The exterior doors and windows are sealed for your safety. Please remain calm."

Answer 32

How can we be sure this is not a mainframe asking this question?

There is always some people in social media with large amount of friends. They would be perfect suspects for massmurder.

First AI would make this kind of questions in a different forums in a name of the Master's friend with widest friend list in social media and then just do what ever kind of murder, since it would always any way lead to that friend. Then move to next one.

One method would be chlorine gas. When pure chlorine is combined with urea it produces poisonous gas. Toilet is perfect for that. Also suspects would lead to toilet manufacturer. Once any member of the family goes to toilet AI would leak chlorine (used to clean toilet sink regularly by AI) to toilet water in order to produce gas that would burn lungs of toilet user and then flush toilet.

Answer 33

Why has no one thought of heart attack?

Having a high and then a low temperature induces a heart attack.

When the owner's sleep is the deepest, set the temperature to slowly climb to 45C. Then, when he starts sweating, drop it suddenly to 5 C within three seconds. Boom. Heart attack. Unless the temperature of the house is constantly logged to a hard drive, I don't think anyone will suspect the AI. This assumes the AI cannot tamper with logs (which we'll assume as a failsafe for keeping the AI out of sensitive systems).

Answer 34

If this 'Smart House' is smart enough to mutiny, then it must be smart enough to hack into the military's missile systems. Simply nuke the master's phone coordinates.

Answer 35

AI doesn't wake user to take diabetes meds (or heart meds): dead

AI doesn't let in user because it's freezing out (temperature sensors of course;): dead

AI suffocates man just for the hell of it via poor HVAC (you're pretty much saying it's hermetically sealed anyways just like my home;): death

AI farts then forgets to notify user (it always happens, bug or no bug;): death and death.