Thorough examination of bias and variance in the linear regression by Arnaud Capitaine via @TDataScience

The connection between probability definition and machine learning explanation.

Great guide and clear explanation.

Achieving Data Literacy: Businesses Must First Learn New ABCs by @ScottZoldi via @datanami

Do you speak data? That’s the essential question Gartner poses to data and analytics leaders in promoting data literacy.

Definitely a reminder of some basic skills that we all need if we are going to use data for anything meaningful (which we absolutely should).

Why Statistics Don’t Capture The Full Extent Of The Systemic Bias In Policing by Laura Bronner via @FiveThirtyEight

Because of a statistical quirk called “collider bias,” the criminal justice system may be even more racially biased than studies suggest. Here's how collider bias works, including charts that clearly show the problem.

This was interesting and the same problems I'm sure are repeated in other areas too. Another bias to try to remove.

WEBINAR: State of AI Bias - 16 January 2020

Sponsored News from Data Science Central

January 16, 2020 (Thursday)   10:00 AM - 10:45 AM ET     While many people believe AI can help solve complex problems plaguing modern societies, can we trust that the AI solutions directing our work and livelihood are rooted in reliable, unbiased data? Do organizations have the proper systems in place to prevent, or quickly address, issues resulting from AI bias? DataRobot surveyed more than 350 U.S. and U.K.-based CIOs, CTOs, VPs, and IT managers involved in AI and machine learning purchasing decisions to learn how AI is being used by businesses today. In this webinar Colin Priest, VP of AI Strategy at DataRobot, will share insight on: How AI is being used by businesses today Current perceptions of AI bias How to prevent AI bias efforts in the future Best, DataRobot team     Save my seat →

The Problem with “Biased Data” by Harini Suresh via @Medium

Poorly defined terminology could actually play a role in biased data, says Harini Suresh. “The right terminology forms a mental framework, making it that much easier to identify problems, communicate, and make progress. The absence of such a framework, on the other hand, can be actively harmful, encouraging one-size-fits-all fixes for ‘bias,’ or making it difficult to see the commonalities and ways forward in existing work.”

I like this great article by Harini Suresh. I have noticed that you need to have an agreed set of definitions for all the data fields, the calculations, the methodologies, and even the data sources because that there are so many synonyms and opposing definitions for all of those that you need to measure like with like in the same way if you want to try and avail bias - if you do not you have already lost the battle.

Pay attention to the man behind the curtain by @quaesita via @topbots

Google’s chief decision scientist, Cassie Kozyrkov, talks about AI bias and how to handle it.

Great observations and well worth reading and thinking about in relation to your own work and the quality of what you are doing.

Strong data quality key to success with machine learning, AI or blockchain by Tendü Yoğurtçu via @infomgmt

Enterprises must be skeptical of data as it essentially determines how the AI will work and bias in the data may be inherent because of past customers, business practices and sales.

The past bias could be inherent in the data due to the design of legacy systems, the team typing it in, the customers, the way it was governed by the business or a combination of them.  Historic data should therefore always be treated with great suspicion until you have completed an exercise to check the systems, the data meanings, standards and governance. Please don't make huge business decisions based on data you can't prove it clean and unbiased.

Can We Make Artificial Intelligence Accountable? by @ctowersclark via @forbes

The ability to open the black box is the holy grail of AI—particularly for industries like law, healthcare, and finance that handle sensitive customer data. IBM may have an answer.

I love the sound of this bias detection software as it's one of those things that you have to watch out for but find it hard to find in your own model - I usually recommend someone else not connected to your area check for bias as they are fresh eyes but if you can use code it will a) improve the detection and b) give some kind of audit trail to show that you don't have bias and did everything possible to ensure there was none.

The bias problem with artificial intelligence, and how to solve it by Sanjay Srivastava via @infomgmt

AI bias may come from incomplete datasets or incorrect values. Bias may also emerge through interactions overtime, skewing the machine’s learning. Moreover, a sudden business change, such as a new law or business rule, or ineffective training algorithms can also cause bias.

I agree - you need good quality and representative training data if you want to get good results from any AI and ML you want to use. My advice would be:

1.  Take your time - rushing always leads to mistakes so be realistic with plans.
2.  Be careful with the methodology you use to create and split your data into Training and Data.
3.  Try to use separate teams to test the same piece of code - the hope being that it will help to avoid the bias. Think of it as a human version of a small parallel ML solution.
4.  Check, check and check again.

Your Data Are Probably Biased And That's Becoming A Massive Problem by @Digitaltonto via @Inc

Nobody sets out to be biased, but it's harder to avoid than you would think.

I really liked this and he makes some very good points.  I believe if you are aware of bias and the ways it can happen you are part way to get around it as you can look out for it and adapt your ways of working to avoid it.  Peer reviews are also great as the more eyes on a problem the less chance you are all going to have the same bias - especially if you are a from different backgrounds and skillsets.

4 Reasons Your Machine Learning Model is Wrong (and How to Fix It) by Bilal Mahmood via @kdnuggets

This post presents some common scenarios where a seemingly good machine learning model may still be wrong, along with a discussion of how how to evaluate these issues by assessing metrics of bias vs. variance and precision vs. recall.

Incredibly useful to remind yourself of all the things you know but have forgotten in your frustration to fix it.

If Correlation Doesn’t Imply Causation, Then What Does? by @akelleh via Medium

Adam Kelleher’s interesting post looks at when and why you might want to use causality.

He has a second post here which discusses all about understanding bias.

Please read these at least twice as it's worth understanding these articles and the points within them well.