logo
logo
Sign in

Responsible and Ethical AI — Building Explainable Models

avatar
USM BUSINESS SYSTEMS
Responsible and Ethical AI — Building Explainable Models

 Ethical AI, in simple words, is to ensure that your AI models are fair, ethical and unbiased.

 

So how does the bias fit into the model? Suppose you are building an AI model that offers salary suggestions for new hires. As part of building the model, you have taken gender as an indicator of salary. The model is trying to discriminate based on gender. In the past, this bias went through human judgments and various social and economic factors but if you include this part of the new model, it is a recipe for disaster. The whole idea is to build a salary model based on people's experiences and qualifications that are non-partisan.

 

Take another example of an app that provides restaurant recommendations to the user and allows the user to book a table. The AI   application is designed to look at the amount spent on previous transactions and restaurant ratings (along with other features), and the AI   system begins recommending expensive restaurants. While there are good low-cost restaurants nearby, those restaurants may not be one of the top recommendations. Also, the amount of money the user spends represents more revenue per restaurant application. So in short, you are driving the consumer class towards spending more in high-end restaurants without the consumer knowing about it. Is it classified as a bias or smart income-generating scheme?

 

Ethical AI is a great topic of research and debate, as you will see a lot of development (as well as general marketing buzzwords) and governance in this area.

 

USM is a pioneer in providing AI & ML solutions that suit's your business needs and help to grow your customer base. To know more about our application development services visit our website https://www.usmsystems.com

 

So how is your model ethical and how do you validate it?

 

Designing a Model Without Prejudice - Make sure you don't include features that may bias your model. For example, do not include gender when evaluating salary packages. Take the time to verify the data sources and attributes that are used to create the model.

Describe model output - Applications design should be a key design principle in terms of detail. If the user receives the output from the AI   algorithm, it must build in the algorithm why the output is displayed and how relevant it is. It should empower users to understand why certain information is being displayed, and turn on/off any priorities associated with the AI   algorithm for future recommendations/suggestions.

Verify the model - Verify the model with enough test cases. You will grow many offerings (ethical AI services) in the future around this area. Again, offerings/services should be vertical instead of pure-play horizontal AI services (otherwise it ends up as chatbots hype

Accountability - Ultimately, humans need to look at the product that comes from the AI   system and take corrective action for critical tasks. I have not seen machines acquire human intelligence for complex tasks in the future. For example, a doctor should carefully consider the cancer treatment option offered by the AI system, but a fashion website that recommends false products for a consumer is not critical and can be corrected by feedback later.

If we go back to the restaurant app if we design the app and explain the output to the user in keeping with the above guidelines, why are we giving at least 4 levels of recommendation (shown in the application tiles):

 

  • Recommend restaurants based on previous restaurant costs, ratings, and history and customer preferences
  • Recommended for highly rated and low-cost restaurants based on ratings, history and customer preferences
  • Recommends new restaurants based on user history and user preferences
  • System generated recommendations without applying user preferences

The revised app now provides enough evidence to back up the various recommendations and recommendations, and ultimately the choice if the restaurant is left to the user to book a table.

 

The above is an example of a very simple application, but imagine when deploying in AI industries and in government agencies. Developing and monitoring the artificial intelligence system for ethical principles is critical. The creators and validators of the model (agencies / third party systems, etc.) are very important to ensure that the models of AI are fair, ethical, and unbiased.

 

Since we are creators and validators of AI systems, we (humans) have a responsibility to ensure that technology is used for good.

collect
0
avatar
USM BUSINESS SYSTEMS
guide
Zupyak is the world’s largest content marketing community, with over 400 000 members and 3 million articles. Explore and get your content discovered.
Read more