Responsible use of AI

The Sri Lanka Association for Software and Service Companies (SLASSCOM) successfully hosted ‘SLASSCOM AI ASIA Summit 2019’ for the second consecutive year on 6th of November at the Shangri-La, Colombo.
While more than 400 participants took part to this event, eminent experts in AI from diverse industrial and academic backgrounds spoke about different interesting topics related to AI.
Out of these interesting topics, Dr. Inga Strumke, Manager - AI and Machine Learning at PwC Norway, shared an important AI factor which is vital from both business and software engineering perspective.
It is ‘Responsible use of AI’
  • What is Responsible AI?
Assume you want to translate ‘She is a doctor. He is a nurse.’ from English to Hungarian using a language translator which is using AI.
You will get it as ‘Ő egy orvos. Ő egy nővér.’.
And then again, if you translate it from Hungarian to English using same translator, you will get it as ‘He is a doctor. She is a nurse.’
Even google translator also gives the same answer.
That means, these language translators think most of the doctors are male.
But, in a time period where people in every field value the gender equality, that will be a biased answer.
Hence, AI systems should be designed in a way which respect societal laws and norms, human rights, diversity and democratic values while ensuring automated decisions are explainable and justified that help to build trust and individual privacy.
According to the Dr. Inga Strumke, “if your AI isn’t responsible, it’s not intelligent. Companies must keep this in their mind when they build their AI enabled future”
  • Do not use Blackbox approach for AI
Most of the AI benefited organizations do not aware about the AI algorithms used for their AI systems. They use black box approach to these systems.
It will cause these organizations to face ethical, legal and financial risks, if they continue this black box approach further.
Hence, it is important to have transparency and responsible disclosures around the AI systems while ensuring that end users can understand that and can challenge outcomes of the systems anytime.
  • Design
Issues like biases, unfairness are raised due to lack of data, incomplete data, outdated data used while designing and training the AI systems.
Hence, AI development organizations should use human centered design approach when designing AI systems using more human interventions.

Throughout the development cycle, It is important to engage with diverse set of users and use cases scenarios and need to evaluate all the decisions resulted from resulting system. This would help to develop more Responsible AI at the end.

Decisions of the AI systems should be explainable, which means that the ability to explain the rationale for decisions.
It will cause to build trust in end users for the AI systems.
  • Monitor
AI systems needs close supervision on the accountability for the proper functioning inline with the responsible AI principles and cyber-attacks.
Continuous monitoring and risk assessment of the AI systems is required to make sure systems are functioning secure and safe way throughout their life time.
Hence, if you develop AI in responsible way, AI has power to make this world a better place.

Comments

Post a Comment

Popular posts from this blog

How to add standard and custom Metadata to PDF using iTextSharp

In a previous blog, I explained How to add custom metadata using Syncfusion library. But this package gives so many difficulties when adding metadata to pdf. Such problems are,    1. When once add custom meta data to a pdf, it does not able to change or remove those custom meta data from the pdf.    2. It does not compatible with meta data search engines. It does not show custom meta data properly when those are searched through the search engines. Due to these reasons, Syncfusion is not good library for custom meta data. But, iTextSharp provides flexible framework to deal with custom meta data as well as standard meta data for PDF. Below I have provided the source code to show usage of iTextSharp library. You can download iTextShrp.dll from here                              PdfReader pdf = new PdfReader(input_path); ...
Read more

How to set and get Metadata to image using BitmapMetadata class in C#

When images are considered, those are storing their Meta data using different types of Meta containers such as EXIF, Adobe XMP, IPTC, Photoshop Image resource and Text descriptions. This every container has their specific unique rules to encode (store) and decode (extract) Meta data. As well as these each Meta container is storing specific Meta information. Exif Meta containers are mostly storing location information and GPS (Global Positioning System) records. As well as, Adobe XMP is responsible to store Dublin core Meta data which provides standard Meta tags that can be used to describe digital objects basically and IPTC is storing Meta information about application records. Meta data extractor is mainly considering these Exif, IPTC and adobe XMP Meta containers when it is extracting Meta data from images. As well as, these different Meta data containers are considered; they are structuring Meta tags in different dir...
Read more

How to generate direct line token to start a new conversation between your own client application and bot framework in Java

You can enable the communication between your bot framework and your client application using Direct Line API. When you add your client application as a site to communicate with your bot framework through azure portal, bot framework will generate secret keys. Your client application can use these generated secret keys to authenticate direct line requests issued to your bot framework. Let me explain how we can generate token to start a  conversations between your own client and bot framework. When we send a request to direct line for generating token,  if the request is successful it will send the conversation id and token that is valid to that conversation id. If client application already has a earlier used conversation Id, you can pass that id as an input to this method for generating token. If there is no conversation id, you can pass null value.  Return type of this method is a Map to retrieve generated token and conversation Id which is valid for that to...
Read more