Over-hype in Ai: a thread. 1/8

Let's start with a story. Remember this one from last year? An algorithm that can tell if you're gay or straight from a single photo. It's a prospect that is both awesome and terrifying. 2/8

https://www.telegraph.co.uk/technology/2017/09/08/ai-can-tell-people-gay-straight-one-photo-face/

Especially when you realise that these things are getting it right 81% of the time.

Terrifying, right?

3/8

But you see, with @joiningdots's help, I've got my own algorithm that can beat this one. And it doesn't require any artificial intelligence or messy facial recognition.

4/8

All you do, is just label everyone as straight. And then because 94% of the adult male population identify as heterosexual, you beat this other algorithm by an astonishing 13%!

(ref: https://www.ons.gov.uk/peoplepopulationandcommunity/culturalidentity/sexuality/bulletins/sexualidentityuk/2017)

5/8

If data is the new gold, we're living in the wild west. We're surrounded by extraordinary claims, often with little proof that the algorithms can or will do what they say.

Some work but don't work well. Others are nothing more than snake oil.

6/8

In the next few days, I'll show you a few examples of the worst offenders.

But in the meantime, if you want to separate the fact from fiction, find where we really lie between hype and hope, then Hello World is 99p as part of Amazon's Black Friday.

https://www.amazon.co.uk/dp/B0785PY2Q4/ref=dp-kindle-redirect?_encoding=UTF8&btkr=1 7/8

It's the best* possible beginners guide to how things work in this bewildering new world.

*at least, the best one by me. I can honestly say I've never written a better one. And if that's not an over-exaggerated claim I don't know what is. /end