Google Duplex uses several different AI techniques in order to create a fascinating new capability: a digital assistant that can make reservations for you.
Give it a listen.
It uses a combination of different AI backed techniques: voice synthesis, voice recognition, and natural language processing.
This opens a brave new world in which each of us has a digital assistant that we assign tasks and it takes care of them for us. Including making phone calls. Such capability also has the ability to do frightening manipulation automatically too.
Handy technique for saving texture space! There are various tools to do this, but the basic scheme is to split up a large texture into much smaller chunks, individually trimming these chunks, and seamlessly reconstructing the sprite in the viewport.
I went to a recent UX Book Club Meetup (Calagator) given by Amber Case. She spent two years as a fellow at MIT’s Center for Civic Media and Harvard’s Berkman Klein Center for Internet & Society studying the interaction between humans and computers.
First off, the talk was given in a really cool spot I’d not been before. A converted 1900’s fireboat house turned into the Boathouse Microcinema. It’s located under a waterfront overpass heavily cross-crossed with active rail lines and inside a non-descript red door in a sketchy industrial neighborhood surrounded by homeless camps.
Inside were the artist spaces for a number of film, music, and art productions.
This talk, however, was focused on her recent co-published book on the use of sound in modern appliances and technology.
She covered a lot of interesting topics.
Most of design in public is visual. How do we do public design for sound?
How do you feel sound? Low frequency can be felt, but requires high volume. Previously this has been limited by tech that distorts too much. But now we’re finally able to push high volume, low frequency sound without the high distortion.
You cannot have an uber-solution or norm for using sound in all devices. People try again and again, but there is no one size fits all.
Sound isn’t like your eyes. With eyes you can close your eyes or look away – you can’t just close your ears when you hear something. This makes the sounds our devices/world makes much more important to pay attention too when designing things. NYC is a loud cacophony.
megaphonics of the late 18-1900’s – art to make loudest sounds possible.
One experiment of having 5 marching bands playing towards each other at the same time.
Corvallis group is re-doing that experiment
Open office spaces are bad! Open concrete restaurants are stylish – but bad!
Service workers are exposed to 90+ decibels for hours
Loudness fatigue is a thing. Tiredness and irritability rise. Documented as ‘alarm fatigue’ or ‘alert fatigue’
Fellow did a podcast that said when he was diagnosed with cancer, he’d go listen to this hold music to feel better.
Most phone audio compression algorithms are tuned for the best voice response. This means almost all ‘normal’ music sounds horrible when run through those compression algorithms. That’s why orchestral music sounded terrible.
Cisco hold music, however, used lots of electronic tones that turn out to map very well to those algorithms. Done by accident, but actually sounds better than orchestral or live bands due to the phone compression.
With advent of modern cities, these places came into being.
Places where you enter and have no identity. Stores went from personal encounters for goods with people to places where you are an anonymous shopper.
industrialization created dehumanizing non-spaces everywhere
“Two Trains” inequality in NYC subway. A project that tries to ‘sonify’ data.
Tips for better design:
Use better equipment when reproducing sounds on your devices. Better speakers, better recordings, etc. Sawtooths/etc are bad.
Particle physicists – they sonify their data to know when something interesting happened in a particle collision. Added little chimes and upsweep sounds to indicate desired types of particles were found.
Incomplete chord progressions get people’s attention. We look up to see why the upwards/downwards sequence stopped before completing.
Notification chime that listens to db rating around it and chimes just above the ambient noise it finds itself in. That way it’s not too loud/quiet for wrong environment.
We have all-or-nothing privacy settings. This is bad. We need a middle-ground. Example: for gps you might want exact location. For many purposes, however, just a city or country is sufficient. We need to make ways for us to share only what is needed and no more.
Remove sounds – motors, HVAC, hair driers, etc – quality is communicated sometimes as the lack of sounds
Dyson devices (fans/hair driers) show people will pay BIG premiums for less sound.
We put foam in car doors to deaden sound
Pay close attention to component fit – poor fitting = rattling
Use containerization to absorb sound. All the blenders in Jamba Juice (Blendtek) have shell containers to seal in the sound. And they charge an absolute fortune for them and get it to protect workers hearing.
High quality things are category breakers – not killers due to high price. People will pay for these devices.
Apartment floors? Can we soundproof them better?
You’ll make a mint if you can fix super-loud leaf blowers
Allow sounds to be turned off or changed into a different sense
For phones, we turn off beep and use buzz/vibrates instead. Much better
Tiny homes sound great – until she tried it. They were WAY too noisy. Everything is in one space and makes noise. This is why Japanese appliances stress silence so much.
Tech can communicate, but it doesn’t need to actually speak. Voice speaking is tricky
Don’t do ‘universal design’. One size does not fit all. Nor does it fit all cultures/contexts. Thinking you can come up with one motif of sound for all your entire product line is a terrible idea and fails.
Roomba vacuums went from a few simple tones to indicate stuck/complete/etc to actually using voice. This was STUPID. extra costs of localization + unnecessary and not as intuitive/easy to hear.
Designing with Sound book
Experiment: she used a contact mic on umbrella to hear rain. Didn’t sound like rain at all!
To make big sound from small phone, old technique is to calculate cavity resonance of the phone and play sounds to resonate the case as well.
Watch out for serious privacy issues. People could turn on Live Listen on their Airpods. Then, they’d leave the room with their phone in the room while leaving their airpod buds in. They could hear everything said in the room through the phone transmitting it to the airbuds. Effectively creating unknown surveillance through the phone.
Abelton 10 – simply record a few basic piano chord progressions and time stretch them. Amazing sounding
Being Mortal – by Atul Gawande
Paro Robotic Seal – This therapy doll for alzheimers patients worked stupid well after attempts with more human-like models or animals. Turns out, the unfamiliarity with a seal means the uncanny valley for seal not the same as for people.
Open thinking time:
She talked about fact people walk around now without talking/interacting. Terrible isolation. My idea: how about headphones that bleed into neighbors as you walk past them?
Robert Moses vs James/Jane Jacobs
Does simply sticking to ambient music isolate you?
Fologram combines computer-aided design with the holographic capabilities of Microsoft’s HoloLens headset to help in assembling even complex objects. The hologram can overlay exactly where each piece of the build should go, as well as an outline of the finished product.
Autonomous tractors, harvestors, and even weeding robots – the end of farming as we know it?
While autonomous cars are getting all the press, there is an even more disruptive side to this technology that is likely far easier and will likely come sooner. Fully autonomous farming.
What about a robot that can patrol fields and kill weeds with pinpoint precision. This would use massively less herbicides. Alternative forms could be developed that fertilize or analyze individual plants or patches of a field for particular problems.
But the really big guns come out below. What if you could replace field work completely and do it all from the comfort of your air conditioned office chair at home?
These are almost certainly going to become realities – probably in our lifetime.
The end of scuff-n-run parking lot door dings? Car prowling a thing of the past?
Elon Musk says Tesla is working on a “sentry mode” security feature that could let owners record damage and break-ins. The announcement came in response to a customer’s tweet complaining of a dent to his Model 3 and suggesting a “360 dash cam feature while parked.”
Tesla introduced 360-degree surround camera views for cars with Hardware 2.5 as part of its October software update. The feature lets owners capture dash cam recordings from the car’s front-facing camera, which can be saved to a flash drive that plugs into the vehicle’s USB port. Pressing an icon saves a 10 minute clip, while holding it down pauses recording. The update also taps in to all eight cameras on every Model S, X and 3 to create a surround view of nearby cars
It also raises the probability of unintented side effects. Cars, in our lifetime, will almost certainly become fully autonomous, always wirelessly connected to national networks, and have 360 cameras/sensor packages/etc.
This has really interesting implications for privacy, crime, and social contracts. It’s highly probable that when any kind of crime or incident occurs, there will now be a whole host of cars that will have recorded what happened from every angle. By using this footage, it would be possible to re-trace suspect paths back for possibly hours if not days/weeks.
For example, the Boston bombers could have easily been identified by watching feeds back until they planted the bomb in the trash bin. Then it might be trivial to follow them back, block by block, looking at the footage of every car they passed until they arrive at their home. You might even be able to follow them back for the weeks leading up to the crime – identifyng every store and person they met with. A complete, airtight case might be created – all without a detective leaving his office seat. Spousal cheaters would be turned in by their car. Lawyers might subpeona cloud services for video proof of a suspect/client’s whereabouts during events. Don’t even get me started about surveillance by fully autonomous cars that can follow you wherever you go and trade off every few blocks with other cars so you don’t even know they’re following you.
It also means everything you do next to a street will likely be recorded from many sources – including ones that pass you by and then are gone. All of which likely is immediately uploaded to the cloud and has only the security of those systems to prevent anyone from using that information for whatever purposes they choose. With data breaches becoming a regular occurrence, it’s something that should make us all give pause.