‘Where there is a decision – there is perceivable bias (no matter how small it is)’
This bias – is going to shape our foundations of future, the question really is not whether it’s avoidable or not but what constitutes a good vs bad or incoherent bias.
Professor Patrick Henry Winston (at MIT) mentioned the fact about machines being able to tell their stories one day.
That, in itself, if done accurately is a big accomplishment.
Let’s understand why – in the age of artificial intelligence or so they call it, there is a black box which has the result, the input and a bunch of neural nodes, every time making sure that the result matches the criteria which makes the result fitting. The real question is, no one knows how the nodes worked among themselves to process the input so that the result matches the criteria. Every time that happens – we have just got the nodes, the expected criteria and the input – what we hope every time which comes as a result – is always aligned and befitting to returns. Well, yes true – but not always and not forever. Over a period of time, this result would change and gradually, it will go off the criteria of success to failure, the only effort is to make sure that inflexion point doesn’t exceed the expected offset of returns vs expenditures. In order to make sure that happens, someone should be able to write the critical decision points leading to the result and make sure that those decision points are not bypassed by any means in near future while running neural processing on the data involved. That’s exactly where the story telling comes into picture.
Now, where does bias come into play here. Let’s go back for a second to the drawing board here. Every decision is an individual’s interpretation of what is right and wrong with some mathematics to it & his confidence on a broad set of experiences in sub-conscious. Whenever a decision is taken in a situation where a group is involved, human tendencies make it easy for the group to interconnect and start influencing the decision within the group one way or another. Imagine a classroom where a group of students, maybe some chosen few regard someone’s abilities to be superior as compared to others. Another example is a party of few people where almost always some are backing each other up. Group influence and bias created because of a few people supporting each other is very common and sometime with improper governance leads to decisions which, seem to be, for general benefit but are actually influenced to serve a few and can lead to crackled decisioning in turn inducing bad bias over the long run.
How to solve this?
Ever heard of anti-patterns, well, let’s take an example. In the above situation – wherein in a classroom with 10 presentations, 2 got applauds by 30% of the students and other 8 were just liked. A survey done later found that one presentation out of 2 was good, but then out of the other 8, 2 were good as well but were lacking the applauds. If there was an antipattern which was introduced to judge the presentations on merely the structure and content, rate them and then show them to audience without the presenter details, that would have caused a clean decision to come but in the absence of this technique, another way is to start taking that bias into consideration and ask questions, not so much on why the 2 were good but why the other 8 were not so good. That’s an anti-pattern, starting to reverse question and change ownership so that there is an anonymity and a set of decisioning parameters which keep the overall decision clean.
Another example let’s take a behavioral approach, in a group, there is always situations where conformance plays an important role, if there is a question and a number of people presenting points, which points are taken or supported plays a critical role in group dynamics of agreement. What happens to this, when a few answer a question and support response of few and negate others. Let’s say, there are 3 people who always support each other out of 10 in group and only 4 or 5 speak in that group when a point is debated. It’s very easy to figure out that any agreement would in turn be agreement of the first 3 involved, other 2 being negated every time hence basically 7 others are quietly ignored. This is a bias which develops unknowingly or knowingly, humans deal with these biases by observing the pattern and changing the group to separate or remove those 3 participants to be in the same group or getting feedback from the rest (the quiet ones) on the decisions. This can be more dangerous if machines start to form groups which support each other, the issue becomes more complex when they can go in and out of groups and can change the overall dynamics in a way so that results are always confirming to their whims. In order to make sure that doesn’t happen, again antipatterns have to be introduced, a support which can look like a plausible supportive candidate which later withdraws to make sure that the pattern doesn’t influence any decisions is key to making this whole process streamlined in doing the right thing apart from making sure that the story is well written.
(PS: The thoughts in this article are completely my own and do not represent the thoughts of any organization or group).
