Center for Knowledge


Subscribe to Our Newsletter

Posts about:

Thought Leadership

Behavioral screening of Large Language Models

The success of  Large Natural-Language Models (LLMs) has been getting more and more dramatic during recent years. It has, even, led to a torrent of new AI applications outside of the NLP field. Alas, this success is afflicted by the models’ inability to prove that their predictions were made on good premises. We know their inputs, outputs, and how likely it is for the outputs to be correct, but we still don’t have a clear understanding of what happens in between. 

Read More
woman standing with a laptop and with a technological background

How the responsible use of technology & high-quality data can help leaders make better decisions

By now, we are all familiar with the term “big data” — we read and hear about it everywhere — as it has become a prevalent concept in a variety of fields and industries, and rightfully so. 

There is a ton of information and digital insights available that can equip leaders to make informed decisions with confidence and meet their objectives in the most efficient, beneficial way possible.

But how data can be incorporated into the decision-making process will depend on a number of factors, such as your business goals and the types and quality of data you have access to.

The key is to leverage the right kind of data in an ethically responsible way, which can prove to be much more challenging than it seems. The good news is that the responsible use of big data is possible and available to decision-makers in the public and private sectors that are looking to identify and understand the needs and pressing concerns of citizens to make better-informed decisions.

Here we’ll dive deeper into the challenges, opportunities, and ways that big data can be leveraged by leaders and decision-makers.

Read More
Robot machine with white hand pointing with purple and pink background

Citibeats & Clever, Chapter III: Working together to create ethical AI

 How can we achieve more representativity of data?

Abby Seneor, Citibeats' Chief Technology Officer, and Clever, an AI assistant from Open AI discuss how to reduce bias and toxicity in machine learning models by increasing the diversity of the data and with the use of open source data!

Read what Clever thinks about our new project, The Ethical AI Community, and get ready for a surprise that Clever has prepared for us!

Read More

Citibeats expands commitment to trustworthy AI with the introduction of the Transparency Page

It’s well known that Artificial Intelligence has been an integral part of the computer science field for several decades now, and perhaps it’s no surprise that it’s taking on an increasingly growing role in social science research as well. And although AI has already proven to have many beneficial capabilities — from the ability to detect early signs of cancer to helping direct government aid in disaster mitigation and relief efforts — this technology also comes with its share of challenges and risks.

The question that persists throughout society is: How trustworthy is AI? To address this, it’s important to analyze what the challenges of human data are, what constitutes ethical AI and what it takes to ensure its implementation.


What are the Challenges of Human Data?

Social science research encounters many challenges when it comes to data. Some of the main ones include acquiring quality data, being inclusive, avoiding individual profiling, and sorting out fake news, among others.

Let’s dive deeper into the first two challenges, as they are very closely linked and are precursors for the other challenges that follow. 

1. Acquiring high-quality data

Since the basis of all research findings and actionable insights is the data, the quality of the data is crucial. When we talk about data that’s collected from online platforms — social media, forums, blog posts — the potential reach is enormous. Consider that, for social media alone, active social media users have now reached a staggering 4.7 billion people worldwide, representing 59% of the total population. 

That leaves a lot of room for potential misrepresentation, fake news, inaccurate information, and false claims. Being aware of this risk is the first step. Taking appropriate actions in data collection to mitigate this issue is the complex next step that requires proper attention and care.

2. Being inclusive and representative

The next major challenge is inclusivity, which includes representing all sectors of the population accurately. For example, bias in research is a major issue. 

When looking at online data alone, online platforms are often not representative of the general population’s gender distribution. Women represent 49.6% of the worldwide population but when we look at major social media platforms, the distribution doesn’t match up. Gender distribution data for both Facebook and Twitter shows females represent only 43-44% of users while males represent 56-57%.

This means that calibrating for gender bias in data collected from online sources is key if decision-makers are to get quality data from which to inform just actions.


What is Ethical AI?

If AI is to be trustworthy, it must be ethical. 

This means that AI technology must be used for social good with clear guidelines in place. It means taking a human-centered approach to the research, development, operation, and use of AI.

Essential characteristics of ethical AI include:

  • Transparency
  • Representativity
  • Veracity
  • Reliability
  • Privacy protection

These are complicated characteristics to implement correctly, which explains why overall lack of trust is a major obstacle to the widespread deployment of AI.


How Citibeats Ensures the Ethical Use of AI

Citibeats is committed to building an ethical AI community and is constantly developing ways to strengthen trust with partners and clients. There are several ways we do this.

In addition to aggregating social data into cohorts to reduce bias and minimize unreliable data by sifting through bots, the Citibeats platform is based on social understanding versus social listening.

Compared to social listening, which has marketing-oriented motivations, social understanding tools contextualize digital conversations and are geared toward interpreting real-time social changes, conversations, and public opinions with the purpose of positively impacting society.

Social understanding focuses on the analysis of social media data and Citibeats is committed to taking steps to ensure a methodological rigourness over the veracity, completeness, and representativeness of the data sampled. This is of utmost importance, as our clients are leaders that look to our data and analyses to inform decisions that will affect citizens' well-being and quality of life. 

On the social understanding approach, Citibeats has developed a methodology to overcome biases and to reach underrepresented populations. To build trust and assure clients that Citibeats’ output is fair, safe, and reliable, we’ve taken an additional step by integrating a new feature into the Citibeats dashboard: the Transparency Page. 


Introducing the Transparency Page: What it is & Why it Matters

The Transparency Page provides an overview of the robustness and performance of the Citibeats data model. It is a holistic framework that helps us govern our AI and data as an embedded component of Citibeats’ strategy for ethical AI. 

And with this new feature that has been integrated into the Citibeats dashboard, our clients have access to the traceability of the data, the accuracy of the attribution, and, ultimately, the veracity of the insights.

Thanks to the Transparency Page, users can:

  • Understand how models work and check the performance metrics with the goal of improving data quality.
  • Check the coverage and representativity of the data to help analysts mitigate fairness and bias issues.
HubSpot Video

It’s through this level of transparency that Citibeats bridges the gap between decision-makers’ traditional distrust of AI technology and the value that ethical AI can bring to decision-makers looking to better serve their citizens.


Read More