PEPPERSNORT.COM
  • Home
  • Subscribe
  • Pepper Blog
  • Meditations
    • Guide
    • mBlog
    • The Others
    • Listening
  • About
  • Covid-19 Submit Success

The Pepper Blog

Picture
​All content and blogs copyright 2019 peppersnort.com and writers

RSS Feed

Archives

March 2020
February 2020
March 2019
January 2019
December 2018
June 2017
May 2017

Categories

All
Philosophy

Don't worry, you've been training AI for years.

1/16/2019

1 Comment

 
Picture
There has been a brief trend across Facebook with people sharing two images of themselves  - a current image, and an image from 10 years ago.

​Some people got wise to this one fairly quickly and highlighted the risk that you might just be helping train facial recognition software.  
Imagine that you wanted to train a facial recognition algorithm on age-related characteristics and, more specifically, on age progression (e.g., how people are likely to look as they get older). Ideally, you'd want a broad and rigorous dataset with lots of people's pictures. It would help if you knew they were taken a fixed number of years apart—say, 10 years.
- Katie O'Neill, "FACEBOOK'S '10 YEAR CHALLENGE' IS JUST A HARMLESS MEME—RIGHT?", Wired.com 15/01/2019
Well, here is the good news.  You've already been training artificial intelligence (AI) for up to ten years.  One of the earliest instances was the "captcha" process which was designed (in part), to hamper technical attacks by bots, hackers and other bad actors.  Over the years, a variety of these captcha processes and variants developed, albeit for increasingly different purposes.

CAPTCHA = 
Completely Automated Public Turing test to tell Computers and Humans Apart

​Some examples over the past ten years:
Picture
Picture
Picture
Picture
Picture
Picture
Around 2009, Google got involved and started using the captcha process.  Users encountering the captcha process were unsuspectingly training computers to recognise unclear text scanned in from newspapers and books.  A computers ability to read your handwritten text also has you to thank for the training you've provided.
While used mostly for security reasons, CAPTCHAs also serve as a benchmark task for artificial intelligence technologies. According to an article by Ahn, Blum and Langford, any program that passes the tests generated by a CAPTCHA can be used to solve a hard unsolved AI problem.
- https://en.wikipedia.org/wiki/CAPTCHA
More recent captcha deployments have been used to train driver-less car.  You may have been shown images of cars, traffic lights, shops and street signs to differentiate and select from.  Some of these images might be a single image divided into 9 - 16 panels, or 9 separate images.  And while the security rationale still exists somewhat, the main purpose is to get you to spend around 10 seconds to train a computer - one that will likely take your job before the next 10 years.

Your selections are then compiled, compared with other responses that enable computer training.  AI software programs can be tested for accuracy, consistency and learning ability against the human response set.  

The process is continuously advancing and now you are being unwittingly invited through these meme games to invest even more time training the AI computers.  The ten year challenge meme invites users to post a photo of themselves 10 years apart.  This helps provide a clear point of reference for the technology to analyse and re-calibrate facial detection.  Even if you don't participate, or if you post bogus pictures, does not matter.  Providing enough people participate randomly (~around 3,000+), the data set gathered will be sufficient to use for computer analysis and testing purposes. 

In fact, your bogus pictures are incredibly useful as well.  Computers need to learn irrelevant data sets as much as they need correct results.  
​Here is a summary of what to expect next in the world of AI training (some of which is already happening):
  • expect further trends/memes (e.g. post pictures of your face at different times of the day - apparently a morning face is difficult for some AI systems to analyse correctly);
  • side by side training (whereby human user and AI perform tasks in parallel and are tested/compared);
  • invisible captcha (the detection test by your action/inaction to determine if you are human or robot is already being performed without you even knowing);
  • your search engine requests linked with credit card transactions (e.g. you searched for x over y period resulting in $z transaction) so that people like you can be targeted with ads more effectively in the future;
  • advertisement choices (by providing you with advertisement settings and choices, you are training AI to target you with more effective ads and relate your ad choices with your lifestyle profile);
  • voice assistant software (e.g. Siri, Alexa)
Picture
Every month, AI systems are improving rapidly.  They now perform millions of tasks that humans are either ineffective at performing or are too slow.  Ultimately, AI systems will run your life for you, and eventually (if they have not already) will make your job redundant.

Thanks for your help to train computers!



More reads and references:


​http://mentalfloss.com/article/81927/surprisingly-devious-history-captcha
https://gurus.pyimagesearch.com/lesson-sample-what-is-face-recognition/
https://www.wired.com/story/facebook-10-year-meme-challenge
https://www.wired.co.uk/article/captcha-automation-broken-history-fix
​https://hbr.org/2019/01/the-future-of-ai-will-be-about-less-data-not-more
​https://developers.google.com/recaptcha/docs/invisible
https://petapixel.com/2015/03/25/how-humans-are-teaching-computers-to-see-and-understand-photos/
​
1 Comment
Jarred
1/25/2019 02:10:24 pm

Merely
doing
this
as
an
example
of
teaching
AI

Reply



Leave a Reply.

    RSS Feed

Powered by Create your own unique website with customizable templates.
  • Home
  • Subscribe
  • Pepper Blog
  • Meditations
    • Guide
    • mBlog
    • The Others
    • Listening
  • About
  • Covid-19 Submit Success