How to keep Human Bias out of AI[Artificial Intelligence]-Ted.com,

The speaker at the Ted.com, talk, seemed to have misunderstood the word, Human Bias and Discrimination, and had gone off tangent from the core subject of the talk.

Check the talk on Ted.com

The speaker has poorly represented AI. If there is bias, then the speaker should have highlighted the type of Bias [there are various bias that existing] she is talking about, instead provides irrelevant illustration and qualifies them as bias.

Understand what AI is

To those of you, who are not aware to what AI is, let me give you all some introduction. AI is a branch in Computer sciences. The Objective of AI is to create systems that can function independently as well as intelligently. Please be aware, this is just the Objective, and it does not necessarily mean that AI will always achieve the objectives.

Symbolic Learning and Machine Learning

AI can be further divided into two branches, Symbolic Learning and Machine Learning. Since the Speaker was talking about Machine learning, We shall restrict this post to give details about Machine Learning.

If you had been watching any Crime thriller movies, one of the most popular scenes in the movie would be the scrutiny of the brutally murdered victim’s photographs.

At least a few minutes of the movie would be spent with the Camera angle being focused on those photographs.

Do you know the reason for this?

The director of the movie wants to convey that one of the ways the Detective will nab the criminal would be through the  careful study of the patterns that the Criminal has used to inflict harm, and commit the horrendous crime on the victim. The pattern would include, knife cuts, punches on the face, any assault on the genital area, etc.

Why Patterns are important?

Study of Patterns would be the key that would unlock the mystery that would lead to the killer?

Patterns would be the vital evidence, both to find the killer as well as to convict him/her. For they offer vital evidences.

Pattern Recognition-Machine Learning

A detective would only be able to use a maximum of 3 Dimensional data in pattern recognition, whereas, a Machine would be more adept in Pattern recognition, using multiple dimension. This is called Pattern Recognition and this exactly what we call as Machine Learning.

Further branches in Machine Learning

Machine learning is further divided into Statistical Learning and Deep Learning. Here we shall restrict our explanation to Deep Learning.

What is Deep Learning? This is a million dollar question. The most important asset in the human body is the brain, the famous quote, “I think, therefore I am”, cannot be more true when it comes to the brain.

The ability of the brain lies in the fact to make informed cognitive decisions. The communication within the brain as well as between the brain and the body, is all carried out by the Neurons.

In Deep learning, we replicate the structure and the function of the brain to make intelligent decisions. There are many different types of Deep learning that Machine use.

Convolution Neural Network

For example, CNN, which stands for Convolution Neural Network. Here the machine has the ability to recognize objects. This is area of Computer Vision by which a computer can recognize and identify people/objects.

Machine learning is based on Data. That would mean, the programmer/user/designer ought feed the machine with data, the more data, the better would be the result.

Coming to the Example as to what the speaker mentioned at the start of the talk, “A Black or Latino person is less likely to pay off their loan on time”.

This is not bias, this is the data available to the Machine to learn and if the data reveals this, then this would be the result that the Machine would infer and advice.

What the speaker fails to mention here is the limitation with which a human would make her inferences, whereas the same Machine can look at the data in multiple dimension, and determine Patterns.

The underlying point to be stressed is, the Patterns, which the speaker failed. Also there are further differentiation to Pattern learning, they are Classification and Prediction.

If the available Data of Loan payment reflects that White People repay the loans on time regularly to a Black or a Latino, then this data falls under the category of a Classified data.

On the other hand, if the data predicts, that a Latino or a Black has often defaulted on loan payment then this would be categorized under Prediction.

Supervised Learning

On the same note, specific Algorithm can be designed with answers for training the Machine on specific requirements. This is called Supervised Learning.

For example, a Diplomat of country is moving in through Airport Check-in, the computer vision can be trained to identify the diplomat and give the person a high priority treatment.

There is also another two branches simultaneously occurring to Supervised learning. They are Un-Supervised Learning and Reinforced Learning.

Un-Supervised Learning

In Un-Supervised Learning, the Machine will make the judgment based on the data provided. For example a user would feed past loan defaulters’ history and expect the Machine to give its decision.

Reinforced Learning

In Reinforced Learning, you feed the Machine the data and expect the machine to make decisions on Trial and Error. This area is used predominantly in Robotics. For example, a Robot’s attempt to pick up a needle from ground.

Irrelevant illustration

With regard to the speaker’s illustration of the use of AI, citing the case of a pregnant woman in Congo, getting her diagnosis on her phone, instead of walking to the clinic which she claims is 17 hours away, raises few concerns,

It promotes Self-medication-which can often have disastrous effects, threatening the life of the mother and the child.

Secondly, let us imagine that diagnosis is available, now how to implement the diagnosis into action? still the Clinic has to be reached for action.

Facts about Congo, before you venture into an Example-advice to the speaker

  1. Congo Unemployment rate is 50%.
  2. Three Million Children suffer from Acute Malnutrition.
  3.  Due to Conflicts, 3 Million people are disabled
  4. 47% of Children suffer from Anemia
  5. 40% of Women suffer from Anemia
  6. Half a Million women suffer from Malnutrition
  7. 60% of Children are less than average in Growth
  8. 7 Million people suffer from hunger induced issues
  9. Congo is the poorest Country in the world.
  10. Poor roads
  11. Poor Infrastructure or missing
  12. Poor Drainage facilities
  13. Mosquitoes rampant-one of the reasons for disease causing
  14. High Mortality rate
  15. High Conflict prone zone
  16. GDP less that $300

Now given the status of the country, the pregnant women would be much happier to receive good food, quality health supplement and good medical care, rather than a iphone with AI, capabilities.

Concluding Remark

A real disappointing Ted.com, talk. We have more to say, but in the interest of Time, Money and Research involved in presenting the Post, we would consider it apt to conclude here.

The Post is meant to serve as a knowledge dispenser

error: Content is protected !!