AiThority Interview with Wendy Gonzalez, CEO of Sama

The recent advancements in Generative AI, LLMs, and deep learning (DL)  have triggered a new wave of innovations in the data science industry. It is impossible to think of building an AI ML foundation without labeled “ethical” data. In 2024, we are eagerly anticipating the growth of Gen AI and Machine Learning models in the data annotation solutions market. At its current rate of adoption, the data annotation solutions market will be worth more than 5 billion USD by 2023. The way the industry is thriving since the arrival of GenAIs and LLMs, we shouldn’t be surprised if the industry exceeds the projected numbers in the next 5 years. By adopting new-age data annotation solutions such as Sama, businesses can derive exception data accuracy to build powerful data pipelines. Last year, Sama announced a completely new version of their computer vision platform to reduce ML algorithm failures and associated risks. The new platform, codenamed Platform 2.0, consists of SamaIQ™, SamaAssure™ and SamaHub™. Last week, we had the privilege to sit down with Sama’s CEO Wendy Gonzalez to discuss the latest trends in the data annotation solutions market and how clients use the platform to deploy AI ML models faster by harnessing Sama’s production-ready bias-free ethical AI platform.

Here’s the full interview.

Hi Wendy, welcome to the AiThority.com Interview Series. Please tell us a bit about yourself. How did you start at Sama?

I’ve always looked to solve the problem of using technology to move business forward. As my career progressed, I began to wonder how I could also positively serve communities in the course of my work. Sama spoke to everything that has been a focus in my career from the beginning. I originally joined the company as an SVP and Managing Director, was promoted to President and COO, and then became the CEO after the passing of our founder, Leila Janah. I work every day to uphold her vision of providing meaningful opportunities to join the digital economy in places where those opportunities aren’t always available.

What is Sama? What are your core offerings?

Sama is a leader in data annotation, data curation, and model validation. Through an enterprise-ready ML-powered platform and SamaIQ™, we uncover actionable data insights with both proprietary algorithms and a highly skilled on-staff team of several thousand data experts, Sama counts 25% of the Fortune 50 companies, including GM, Ford, Microsoft, and Google among its customers.

We are also a social enterprise and registered as a public benefit corporation.

Founded on the principle that talent is equally distributed, but opportunity isn’t, the majority of our operations are in Kenya and Uganda, where we hire based on impact to empower people who often don’t have other formal employment options.

Please tell us about your Ideal Customer Profile (ICP) and the problems you solve for your customers. Which is your biggest market in terms of customer lifetime value and revenue generation?

Sama’s ideal customer is an enterprise company building generative AI (GenAI) models, large language models (LLM), or computer vision models.

Our customers represent several industries, and we primarily work with high-tech, retail, automotive, agriculture, and robotics companies.

For any model to perform well, a company needs a lot of data that is properly labeled and validated. Once a model is developed, companies need partners who can identify where and why models are not performing to standard and can then create or source the necessary data to resolve that issue. Sama provides both of these services and more.

Our biggest market is in GenAI. We can help companies that are building models from scratch or ones that instead choose to finetune extant models for unique applications.

How can retail and e-commerce users benefit from Sama’s data annotation capabilities? Could you provide some insights into Sama’s annotation AI training data for the Retail industry?

AI offers retail and e-commerce several key benefits, but three of them are especially predicated on having well-annotated, high-quality data: personalization, visual search, and VR/AR. If a model doesn’t have high-quality labeled data, it can’t find the right kind of item a consumer is looking for in the first place, no matter how vague or specific they are in their search terms.

On top of that, well-annotated data lets these algorithms pivot to recommended items — increasing average cart value and improving the customer’s experience.

For VR/AR, displaying the items correctly is essential and is dependent again on good data annotations. If a virtual try-on of a new lipstick color doesn’t actually put the color on the customer’s lips, they’re less likely to use the feature in the future, making it harder for them to visualize themselves with the product and less likely to buy it.

You recently launched Sama Red Team, what’s its purpose and what does this bring to the industry?

Sama Red Team is one of the first red teaming solutions specifically designed for Generative AI (GenAI) models. The idea of red teaming — deliberately attempting to break a piece of software to expose vulnerabilities — isn’t new. For GenAI models, it takes a slightly different form, though, because there are so many different ways to evaluate a model.

Is the model fair?

Is it accurate?

Does it comply with laws and regulations?

How easy is it to evade its safeguards and make it output potentially harmful or discriminatory content?

Our Red Team solution looks at all of these questions and more.

This is really critical for GenAI because it is very good at sounding confident when it isn’t, and it’s not necessarily able to guard against malicious actors trying to expose private information or obtain information that could threaten public safety. Red teaming allows these vulnerabilities to be safely exposed before they become real problems.

AI ML and Leadership at Sama

How big is your AI team? Could you tell us what kind of use cases you are currently focusing on?

We have over 4,000 associates, who make up the bulk of our workforce and are primarily based in East Africa. While we provide data annotation, model validation, and data curation services across a number of different applications, we are particularly focused on GenAI (of course), retail and e-commerce, automotive, and AgTech.

What are your thoughts about the recent EU AI Act? How would it impact the global AI ML development?

The EU AI Act is the first truly comprehensive approach to AI regulation, and it tries to strike a delicate balance between protecting privacy and enforcing misuse of AI while also fostering innovation. There is still a trust gap between AI companies and users; this Act aims to close that by requiring more transparency in interacting with an AI, clearer explanations on if and how data is being used, and so on.

What is clear across the board, no matter the method or approach is that governments do not want to stifle innovation. There are plenty of regulatory sandboxes being created to give companies the chance to test and experiment with new models.

At the end of the day, AI will be regulated around the world. That is unavoidable; this Act’s passage is just the first step. We are no longer in the wild west of AI, but rather in a new era where AI development is more closely watched by regulatory bodies.

Which LLMs and GenAI tools have you tested and used? What are the benefits and latent challenges associated with working with these new-age AI tools?

At Sama, we’ve worked with a number of the leading LLMs and GenAI models. LLMs and GenAI will have huge impacts thanks to their ability to automate or greatly reduce the time needed for repetitive tasks. They can also play a huge role in creating synthetic data, which will become more necessary as new data protections (like those in the EU AI Act) take effect.

But like any AI model, LLMs and GenAI models are only as good as the data that they’ve been trained on — and with the wrong data, a GenAI model could confidently output completely false information (model hallucination). This misleading information could have real consequences, such as the case of Air Canada having to honor a policy its chatbot completely made up. It makes having a Human in the Loop (HITL) approach, like Sama does, critical, as this combined human-and-machine analytical approach allows us to identify when models are too confident, then provide additional training data to adjust that confidence level accordingly and ultimately help them perform better.

Sama recently announced investments to support women professionals in the tech industry. What inspired this decision?

From the beginning, Sama has been focused on empowering marginalized communities – and that includes women. The issue of women’s representation in tech is a well-known one, but I also want to call out the need for everyone to have a seat at the table when it comes to AI development, which includes people of all genders and from all parts of the world. We’re proud to say that 53% of our workforce identifies as female; 48% of senior managers identify as female; and 50% of the executive team identifies as female. Our decision to further invest in women in the industry and foster the next generation of women tech leaders is an extension of this mission.

As a woman leader in the AI ecosystem, what message would you like to share with young women professionals pursuing STEM and marketing jobs?

Be the best version of your authentic self.  For example, when I first became Sama’s CEO, I thought I might need to upgrade my communication style. While that was very true and I continue to improve my craft in all ways, you also don’t become a new person overnight.  I learned after a while that the best way I could show up was really just like myself, even if those traits aren’t stereotypically a CEO’s traits. What makes you the person you are has value.

Don’t lose sight of that.

Future and Predictions

Is it possible to replace humans with AI jobs? If your answer is “No”, please explain with examples.

No. At Sama, we believe that AI needs a human in the loop, giving it feedback and working in concert with it. This helps improve model performance and reduce the risk of catastrophic failure. It is difficult, if not impossible, to have responsible AI without human-centric or human-validated AI; if we are committed to the former, then we must have the latter.

This does not mean that AI will not transform how a lot of jobs are carried out. We need to be mindful of that as we continue developing AI. But AI is a tool, and we’re still learning what it’s capable of. As we develop models, we need to be mindful of the impacts they will have on jobs and individuals and work to ensure that people aren’t left behind.

AI regulation and privacy concerns – how should CEOs take a leadership role in executing an organizational AI strategy?

Many companies are already following some best practices that were codified into law by the EU’s AI Act, such as AI governance teams and AI impact assessments. However, companies may have so many initiatives going on at any given time that they don’t have visibility on all of the use cases being developed.

Putting more formal systems into place to capture and follow all AI initiatives has to become an immediate priority for CEOs to ensure they are compliant not only with the Act once it comes into effect, but also with current obligations, whether they come from currently applicable laws or their own contracts and policies. This has to come from the top down to ensure every initiative is being monitored for its compliance and development. As a CEO, our job is not just to drive strategy and hold people accountable to execution, but to build the right teams that can put these structures in place and maintain them.

Your take on the future of AI-led marketing and sales roles – how should Marketing leaders prepare for an AI-led organization?

AI is developing so rapidly that it can seem tempting to immediately implement it to feel like you’re keeping up.

Unfortunately, that can be the wrong approach. Instead, companies need to be thoughtful about how they implement AI and find places where AI can augment a person’s work and create efficiencies.

In particular for marketing, AI can’t be the end-all and be-all. Companies shouldn’t plan to just click a button and have AI generate a sales email or a blog post. AI is capable of making mistakes and coming up with completely false information (hallucinations). These situations require a human to review and revise.

Your approach to AI needs to be strategic.

A clear understanding of your organization’s goals and vision is important as this will guide which tools can provide the most ROI for you. In addition, it will inform your decision when it comes to choosing the right partners in model development.

Thank you, Wendy! That was fun and we hope to see you back on AiThority.com again.