Let's be Fwends is sent via MailChimp, an email markting platform. By clicking 'join now!' to submit this form, you acknowledge that the information you provide will be transferred to MailChimp for processing in accordance with their Privacy Policy and Terms.

Actually, being gullible might be the most human characteristic. So maybe the machines aren't that different from us after all?

View this email in your browser

LET'S BE FWENDS ISSUE #20:

FOOL ME ONCE, SHAME ON ME. FOOL ME TWICE, SHAME ON YOU.

Source: unknown/multiple

Fooling the machines appears to be pretty easy

So much depends on machine learning and smart algorithms nowadays. From the video sensors in self-driving cars to the speech recognition software that powers Siri, Google Home or Alexa.
No wonder people are probing them to find weaknesses and use them as attack vectors.

Here are just some current developments in that area:

Researchers at Berkeley can take any audio input and manipulate it in such a way that it stays 99,9% the same, but will be decoded by a certain speech recognition software in any way they like. Meaning they can inject voice commandos into any audio signal, undetectable by the victim of the attack.
Gives a whole new dimension to that XKCD cartoon.

Robohub published a summary post of physical objects that can be used to attack image recognition algorithms. They show how simple stickers with image data designed to confuse the algorithm can make the software blind to certain objects. Researchers were able to make a STOP-sign invisible to the algorithm just by applied two such stickers to it.

And researchers at the University of Michigan were able to trick a deep neural network into confusing a STOP-sign with a Speed Limit 45 sign.
Making something disappear from a Robo can be useful in the post-apocalyptic Terminator-infested wasteland that was L.A., but in 2018, it’s not really reassuring that you can make STOP-signs into Speed Limit 45 signs.

Engineers at Google have shown how to distract an image recognition algorithm enough to make it miss the obvious object on an image. They created a sticker that will make the AI identify everything as a toaster, regardless of what else might be in the image.
This one actually might be useful, for example to foil mass surveillance and facial recognition.

Scientists at Kyushu University can fool common image recognition algorithms over 70% of the time, just by placing one special pixel on an image.
If you look at the results, you get the impression the AI just threw its hands in the air and said „Beats me. You tell me what that is!“

Where does that leave us? The arms-race between attackers and defenders has just begun, and interestingly, the attackers seem to have the upper hand at the moment. Of course, the algorithms will be protected against these attacks. But this attack vector is here to stay, just like computer viruses.

But hey, humans can be tricked by pixels, too

It’s not just the bots. Humans also aren’t immune to getting it wrong when it comes to pixels.
Check out this ad campaign.

Devious.

Your Data and Advertising

Here is a super-detailed blogpost about what the new General Data Protection Regulation that will come into effect on 25 May 2018.

Scan the examples and wireframes they provide, and you quickly see that this will be a major departure from how the online ad tech industry currently works And being totally compliant seems unattainable to most ad networks. The directive will be in full effect in little more than 4 months, and there appears to be little movement on the side of ad tech.

Don't know about you, but I'll grab some popcorn.

On Good Design

Image Source

Do you know Dieter Rams? Even if the name isn’t familiar to you, I’m sure his work is. He was chief design officer of german brand Braun, and created countless classics (watch aficionados please note that the famous Braun timepieces were designed by Dietrich Lubs, although under the heavy influence of Rams).

Besides his relentless work as an industrial designer, he reflected on what design is, and what role it has in society. And more specifically, he asked: Is my design any good?
As an answer to this question, he devised his now famous 10 Principles for Good Design:

Good design …
is innovative
makes a product useful
is aesthetic
makes a product understandable
is unobtrusive
is honest
is long-lasting
is thorough down to the last detail
is environmentally friendly
is as little design as possible

https://www.vitsoe.com/rw/about/good-design

A digital society is a designed society. Rams principles needed an update. With a broader application of design, the question about good design broadens as well. And Suzanne LeBarre has new answers to it:

Good design ...
is transparent
considers broad consequences
is slow
is honest
is political
is mindful of systems
is good writing
is multifaceted
takes risks
is for people — and machines

https://www.fastcodesign.com/90154519/10-new-principles-of-good-design

And on a similar note ...

Here are 13 ways to screw up a client presentation.

Can you measure good design? And if the answer is yes: How?

Here's an interesting post on wether you can use Net Promoter score to measure ... well, frankly, anything.

You can read this as a torching of Net Promoter Score, but it really is a stern reminder to be always careful when transforming emotions into numbers and feelings into aggregates. It seldom works.

Beautiful Pixel-Illustrations of vintage computers

Artist ciara. makes beautiful 8bit-ish illustrations of vintage computers and other objects of interest.
Check out that GameBoy or that Apple.

Horses. Really smart animals.

That's it from this edition of Let's be Fwends. If you made it down here and still think that computers are pretty amazing then please high-five yourself you inspiringly optimistic and positive being! 👊

Enjoy what you see? Please recommend this newsletter to someone who might enjoy it equally. Thanks!

Send a tweet to your friends over at twitter.

View this issue of Let's be Fwends in your browser

Fool me once, shame on you. Fool me twice, shame on me.