The Utah government paid $20 million for an AI software that scans social media posts and identifies criminal activity in real-time.

Due to some problems they had to start an investigation: is the AI racially biased?

What do you think they found?

I love this story, so a 🧵 /1
Before you read the rest of this thread, try to guess what the investigation found in response to the question: Is the AI algorithm racially biased?

I bet the answer will surprise you.

But let me tell the whole story first. /2
Banjo, was a software company that made Live Time, a software with the capability to "detect active shooter incidents, child abduction cases, and traffic accidents from video footage or social media activity".

Impressive, no? /3
The Utah government awarded them a $20.7 million contract.

But, news broke that the founder of Banjo was a white supremacist with ties to the Ku Klux Klan.

As a result, there was concern that the AI algorithm itself had white supremacist biases. /4
This was around the time of the George Floyd case…

Worried of the negative PR, the government stopped the use of Live Time, and called for an audit. /5
The audit report came out last week. Are you ready for the big reveal?

The audit report resulted in this memorable headline:

Government audit of AI with ties to white supremacy finds no AI…

The Live Time software claims to read Twitter feeds and analyzes the text and videos there.

The audit found that the software doesn't actually read any data from Twitter, and the algorithm does not "meet the industry definition of Artificial Intelligence" /7
The moral of the story is:

if you're buying AI software, ask the hard questions to find out if there is really AI in the software, and is the AI really adding value, and then ...
you're still not ready to buy the software

You then have to check if the software is biased. /end
Does anyone have a link to an article or Twitter thread giving instructions on how to evaluate the claims of an AI software vendor before buying? I seem to remember retweeting a thread like this, but I can no longer find it.

• • •

Missing some Tweet in this thread? You can try to force a refresh

Keep Current with Navin Kabra

Navin Kabra Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!


Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @NGKabra

19 Mar
1. "There is too much variability in our hiring process," says the business leader. "Depending on which interviewer they get, people with very different skills get hired. Your tool will help us standardize."

Supposedly the customer is always right, but this is where I push back
2. I feel like wisecracking: if our software outperforms your employees, you need better employees

But I know that's not true. They are usually good employees who need to be given better training
3. So I dig into what is happening in the interviews.

Depending on availability, a candidate gets randomly assigned to one of 3 different interviewers. And each of those interviewers asks questions based on their experience, background, and what they're currently obsessed with.
Read 22 tweets
9 Mar
Did you know that all the nuclear bombs we exploded in the '40s-50s-60s have permanently contaminated all the steel the world has produced since then? All steel in the world is divided into two categories: pre-1945 non-contaminated steel, and post-1945 contaminated steel.

Nuclear bombs (including the tests) create a lot of radioisotopes that are not found in nature. For example Cobalt-60 (a radioactive version of Cobalt-59). And since the 1945 Trinity test, these have all dispersed in the atmosphere.

What does this have to do with steel?
The process of manufacturing steel uses atmospheric air (or atmospheric oxygen). Some of the radioisotopes like Cobalt-60 get pulled in too, and as a result, any steel produced after 1945 has embedded in it some of these radioisotopes, sitting there emitting gamma rays.
Read 10 tweets
22 Dec 20
1/ I've now taught my "Defense Against the Dark Arts" course (teaching teenagers about misinformation, fake news, misleading ads, etc) to 4 batches, and it has been a very rewarding experience for me

A quick thread on my learnings and suggestions for you…
2/ My biggest learning has been how quickly children figure out things with the help of the internet as long as they are just pointed in the correct direction.

They don't need teachers. They need people who can motivate them and point them in the right direction.
3/ Consider the exercise in this photo. I gave them no other coaching, but within minutes most of them had figured out why each of these ads was misleading and they were able to articulate and discuss subtleties like "The product is a good product, but the claim is exaggerated." Image
Read 13 tweets
24 Nov 20
If, like me, you're confused about the Oxford/AstraZeneca (Pune-walla!) vaccine, which is 70% effective, but also 90% effective, then here is an explanation. 🧵
The Oxford vaccine requires two shots one month apart. The original design called for two equal shots. However, someone made a mistake and a small fraction of the volunteers got only half a dose in the first shot, and the full dose in the second shot.
Later, analysis of the data showed—to everybody's surprise—that the ½+1 dose regimen was 90% effective and the standard 1+1 dose regimen was only 66% effective.

The average of these two regimens across all volunteers is 70%, and that is what has to be reported 🧵
Read 13 tweets
29 Sep 20
As a child, I always wondered why Indian words like राम always got written as "Rama" in English. To me "Ram" seemed like a much better fit for the pronunciation. I thought it was because of British arbitrariness.

Only recently I learnt the real reason for this. /1
Consider गम. How would you pronounce it? Like the English word "gum" right? As a Hindi/Marathi speaker, I can't imagine any other pronunciation.

But, for a Sanskrit speaker, things are different. गम् is pronounced "gum". In गम the full म is pronounced, so it becomes "gum-uh" /2
The extra "-uh" sound which is the difference between गम and गम् is called a schwa

And modern Indian languages like Hindi and Marathi all have an (unwritten) rule that the schwa at the end of a word isn't to be pronounced

This is called schwa deletion… /3
Read 13 tweets
27 Sep 20
Ok, let's do this. What are good YouTube channels that you watched (or still watch) because they're fun to watch, but you also ended up learning from them?

Only recommendations from people born this century are allowed

@gauravjoshi2000 @arshkabra @KabraRuhi @ShahAvanti02
Include a link to the YouTube channel, what it's about, and why you liked it. And tag others who could add to the list. Younger kids deserve better recommendations than the ones from the previous century.
YouTube for youngsters: Epic Rap Battles of History. This did more for @ArshKabra's history knowledge than ICSE History ever did.
Read 40 tweets

Did Thread Reader help you today?

Support us! We are indie developers!

This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Too expensive? Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal Become our Patreon

Thank you for your support!

Follow Us on Twitter!