New Voices on AI: Part 2 of Our AI Q&A

Last month, I spoke with 2 of Toluna’s AI experts about how AI is transforming the insights industry and how we at Toluna have been embracing the AI wave. Our conversation inspired me to dig even deeper into the world of AI.  

Today, I’m speaking with Marie Hense, Global Head of Quality, and Gene Saykin, VP/Head of Quality Strategy, who were gracious enough to let me pick their brains. 

Q1. Shivaani: Thanks for joining me today! To start off, everyone is talking about AI – it’s a hot topic. In your opinion, what’s the biggest misunderstanding about AI in the market? 

A. Marie: There’s a misconception that AI can do everything, answer everything, and solve everything. And that it can fully replace human intelligence. As exciting as the idea is, it just isn’t true. In our opinion, the power of AI is best unlocked in combination with human intelligence, rather than in replacement of it. You’ve probably heard the term augmented intelligence being echoed across the organization – it’s what we call Toluna’s approach to AI, which is centered on leveraging both artificial and human intelligence to get the best out of both. 

A. Gene: Yes, the human component is pivotal in ensuring we deliver high-quality data and superior client insights. AI on its own has limited understanding and contextual awareness. It needs to be taught and trained to accomplish tasks with the degree of accuracy and the specific results we expect from it. It’s like if a 4-year-old child had soaked up all the information he could find up to a certain point in history… the child might understand what is being said but not why. Even if you equip this child with tools to search the web for new information, he/she would still not always understand what specific outcome you want. Similarly, AI has tremendous ability to learn, but expecting it to be able to do everything on its own is futile. That’s where the human element comes in to fill the gaps, to train and guide it to produce the results we’re aiming to achieve. 

 Q2. Shivaani: How do you see the relationship between human and artificial intelligence evolving?

A. Gene: The relationship is symbiotic. Human intelligence is and will continue to be necessary for the training, development, and guidance of AI models, as well as for the oversight and governance of responsible AI implementation. Using AI ethically is of in ways that prevent harm, promote fairness, and foster trust is of utmost importance. Trustworthy AI requires constant efforts to detect, mitigate, and prevent bias. Transparency about how decisions are made using the Explainable AI (XAI) concept is key.  This allows users and stakeholders to understand how an AI model arrives at its decisions, making it easier to trust and challenge those decisions when necessary. AI systems should be designed with safety mechanisms in place to prevent harm or unintended consequences. Rigorous testing, robust error handling, and fail-safes are essential to making sure that the AI tools we develop are ethical. It’s not just about building systems that function well, but also ensuring they align with human values, fairness, and respect for human rights. It should enhance the human decision-making process, not replace it.  

A. Marie: And while we anticipate that AI will become smarter, clients will always want to know how and where our expertise as researchers and insights professionals comes into play with our AI-based solutions. To go back to what I said earlier, the best AI outcomes are the result of the collaboration between AI and human intelligence. The two aren’t competitors!  

Q3. Shivaani: Putting on your Quality hat, what are the biggest challenges when it comes to AI? 

A. Marie: To start off, ensuring that our integration efforts are conducted responsibly. And that our actions have a significant impact internally, as well as on our clients, while also staying aligned with our company values and philosophies. We must be confident that we’re developing tools we can stand behind – those that deliver high-quality insights, built on high-quality inputs.  Take the topic of synthetic data, for example. Creating synthetic boosts and synthetic personas is only possible if the data that goes into the AI model is of high quality and comes from real, genuine, and engaged respondents. So, ensuring we apply AI where it makes the biggest difference, even if it’s on topics that are less appealing such as fraud pattern detection, is key.

A. Gene: There’s also a lot of intrigue surrounding AI, so clients are always up for testing and experimenting. But AI isn’t perfect, and accepting that is something I’ve seen market researchers struggle with and struggle to address. One example is our automated open-end checks. AI may be good at differentiating foreign language, gibberish, duplication, etc. in most cases. But when it comes to recent developments like Gen Z speak, specific industry terms, acronyms, and even slang, it struggles. That’s where a human is needed to continuously feed it new information while guiding AI in how it processes that information. And speaking of new information, one key issue with AI and the vast amount of data it will potentially generate in the future is that it will become a ‘closed loop system’ – essentially a system where future AI models feed off previous AI models’ outputs. Casually, this is also referred to as the ‘Habsburg problem’. So it’s absolutely crucial that new data is generated continuously, even in the future, to make sure AI can solve up-to-date real–world tasks. For example, if you had an AI trained on data before the term “selfie” was invented, asking it to generate a selfie-style image would lead to some strange results. If you’re interested in what the AI models know today, just start by interacting with them, bounce some of your ideas off it, or use it to walk you through something you‘ve wanted to learn for a while.  

Q4. Shivaani: Finally, AI is a vast field with so much going on – and so quickly! How do you stay updated on the latest developments and ensure we’re staying at the forefront of innovation? 

A. Gene: We have dedicated Innovation and Machine Learning teams who divide and conquer. They’re responsible for becoming subject matter experts on specific developments and trends, and they share their knowledge cross-departmentally. The Quality team is strongly connected with these teams, as there are substantial benefits from AI in all topics – from respondent screening and methodology best practices all the way to how our platforms like Toluna Start support users with executing high-quality research projects.  

A. Marie: Beyond that, there is an abundance of information available online. This can be overwhelming, especially with rapid changes occurring within the field constantly, but there are some great podcasts and blogs that really highlight the key points everybody should know about AI. It’s especially important to look beyond the horizon of the MR industry and to look at examples of how AI is being applied in other industries. For example, we are currently testing biometric verification as part of our suite of identity checks, which is an AI-based verification that works well in industries such as finance and banking but isn’t widely used in online research yet. Learning from other industries and thinking about ways of applying what they do in MR is crucial.