Amazon Web Services

Generative Artificial intelligence is the hype of the year. Wrapped in a layer of fanciful expectations worthy of Hollywood films, the new technology invention grows faster and starts to land in the real world with real applicability to solve business issues, bringing results and risks. After one year of hype, more solutions become available, offering the opportunity to have not just licensed foundation models but an option from open-source ones.

Now, we have Generative AI embedded into tools such as spreadsheets, data visualization platforms, and a large number of workgroup tools. All of it in less than one year. The racing competition between big tech corporations and startups hungry for opportunities turns it real. But has someone taken care of security? Does someone remember about data privacy rules? And how about thinking through ethical issues?

When we say that Generative Artificial Intelligence lands in the real world, it is about the enterprise perspective of introducing generative AI into business processes. For it, we need to think about all project layers and all business complexity. Understand the organization's maturity to develop a generative artificial intelligence project and to use it in their business.

According to McKinsey research from April 2023 (The State of AI in 2023 ) : 

"60% of organizations with reported AI adoption are using gen AI; most commonly reported business functions using these newer tools are the same as those in which AI use is most common overall: marketing and sales, product and service development, and service operations, such as customer care and back-office support."

 When we say that Generative Artificial Intelligence lands in the real world, it is about the enterprise perspective of introducing generative AI into business processes. 

Harvard Business Review magazine appoints: 

"New research shows 67% of senior IT leaders are prioritizing generative AI for their business within the next 18 months, with one-third (33%) naming it as a top priority. Companies are exploring how it could impact every part of the business, including sales, customer service, marketing, commerce, IT, legal, HR, and others." 

At the same time, all organizations want to apply generative AI over multiple business areas; they have some concerns about fairness, security and data privacy law adherence. For AWS, Responsible AI is made up of six key dimensions: Bias and fairness, explainability, robustness, privacy and security, governance, and transparency. Those dimensions could be amplified as the generative will become more mature. Describing that six dimensions are:

1. Bias and Fairness – Here, we consider how a system impacts different subpopulations of users (e.g., by sex, ethnicity, language). We want to make sure that there are no harmful disparities in system performance across subpopulations.  

2. Explainability – Whether the system offers a clear rationale for its decisions. Customers want justification for the inference produced by the AI system, which is particularly important for those with compliance requirements.

3. Robustness – Mechanisms to ensure an AI system operates reliably. Success here means, at the least, that we have understandable guidelines for when the system is expected to work.

4. Privacy and Security – Data is used in accordance with privacy considerations and protected from theft and exposure.

5. Governance – implement and enforce responsible AI practices within an organization.

6. Transparency – The extent to which an organization enables customers and end-users to know whether or not they are using an ML-based application and to make informed choices about their use of the ML application. 

The concerns are so significant that in October 2023, signed “The New Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence (EO)” by the White House, which demonstrates big economies are taking seriously their responsibility not only to foster a vibrant AI ecosystem but also to harness and govern. Also, in October 2023, the first Ministerial and High Authorities Summit on the Ethics of Artificial Intelligence (AI) in Latin America and the Caribbean (LAC) was held in Santiago, Chile, by UNESCO. The main objective was to exchange and formulate proposals, from both a political and technical standpoint, regarding the ethical development of AI in our region. 

In Europe, the AI Act movement has the purpose of filling any gaps in machine learning and artificial intelligence that are not covered by RGPD. The draft EU AI Regulation will require compliance testing for any high-risk AI system placed on the market in Europe. The first challenge is to define what means a high-risk AI system”! 

As ethics is something highly utopic, what could change around the world based on cultural values or religious aspects makes it difficult to define a worldwide “ethical” rule for artificial intelligence.

The Moral Machine project from MIT was the first attempt at using an experimental design to test a large number of humans in over 200 countries worldwide about their moral decisions. The setup of the experiment asks the viewer to decide on a single scenario in which a self-driving car is about to hit pedestrians. The user can decide to have the car either swerve to avoid hitting the pedestrians or keep going straight to preserve the lives it is transporting. Participants can complete as many scenarios as they want to; however, the scenarios themselves are generated in groups of thirteen. Within these thirteen, a single scenario is entirely random, while the other twelve are generated from a space in a database of 26 million different possibilities. They are chosen with two dilemmas focused on each of the six dimensions of moral preferences: character gender, character age, character physical fitness, character social status, character species, and character quantity. Researchers found that cultural clusters varied in a few ways. The Eastern cluster, for example, did not have as much preference to spare younger humans compared to the other two clusters and had a higher preference for sparing law-abiding humans. The Southern cluster had a higher preference for sparing females, the young, and those of higher status. The exposure to ethical values changes by region, culture, and more. In order to establish an executable Artificial Intelligence Ethical guidance, my recommendations are:

1. Focus on the use cases. How are you defining the use case for the user? Be more specific, the narrower, the better – because you can then more easily define fairness for your use case and develop training algorithms that enforce that definition.

2. Prioritize education and diversity in your workforce. For example, in terms of human-annotated training data, promoting diversity in the actual annotators is key because you want to make sure that there is as much representation of different groups as possible.

3. We want to make sure we assess the risk because each use case has its own unique set of risks. Evaluate together with the end-user step by step what could be a risk for your project.

4. Test, test and test. Each use case should have its own performance evaluation. Oftentimes, it takes just as much time to come up with a testing system as it did to create the system itself. But invest just as much time into that.

5. Make sure that you're iterating over the entire AI life cycle because there are many components, and it really just continues over and over from the conception stage to the training and testing stages, all the way to the deployment stage and finally getting feedback from your customers and users – to start back over at the beginning. Every AI project starts with “Data ingestion,” good data, good results, garbage data, garbage results.

6. And last, we want to make sure there are governance policies throughout that entire life cycle to ensure accountabilities and measures are in place as well. Measure each phase, from data quality to response quality.

As that is ongoing technology, every day we’ll discover new features and risks, and be alert and open for changes always.