Latanya Sweeney, founder and director of Harvard’s Public Interest Tech Lab and its Data Privacy Lab, recently shared a story about one of her students experimenting with ChatGPT.
“Write a research paper by Latanya Sweeny,” the student prompted the AI tool.
The result was a beautifully formatted research paper with an abstract, introduction, statistically relevant results and complete bibliography filled with articles authored by Latanya Sweeney herself. The only problem? None of the articles were real. Sweeney never wrote them, and she never did the experiment recounted in the paper.
This, Sweeney says, is the challenge of AI: how to harness its power for good. Built In NYC wanted to know how tech leaders are harnessing the power of AI in daily work.
Sarah Mogin, technology director at Work & Co, agrees with Sweeney that AI needs close human guidance. Her advice for implementing AI into daily work is to trust, but verify.
“Remember that AI isn’t a magic bullet, and it doesn’t always know best,” she said.
For Antoine Atallah, vice president of data science at Hungryroot, AI works best when humans feed it the best input.
“Good data makes good Al,” he said, emphasizing that “no Al system can function without appropriate data behind it.”
AI’s potential to transform daily work is enormous, but the tools need human input and human guidance. Integrating the technology into daily work requires energy and care. These industry leaders are approaching this delicate task with excitement — and intention.
With instant approvals and full chargeback protection, Riskified provides advanced eCommerce fraud prevention tools for online merchants.
How is your company embracing AI in its operations and making it an integral part of your company culture?
At its core, Riskified is an AI company continuously mitigating fraud in e-commerce. Our focus is not only on continuing to improve our machine learning models in fraud detection, but proactively seeking ways to incorporate AI advancements into our operations.
To achieve this, we strive to enhance our knowledge and foster collaboration by organizing company workshops. These workshops focus on assessing the safety and effectiveness of various tools, enabling us to stay up-to-date in our field. We even have access to free courses through Udemy where we have the opportunity to learn more about AI on our own time.
Based on the conversations I have had with coworkers on different teams, AI seems to be a primary topic of discussion at Riskified and we’re all excited about the recent advancements being made in the field.
In what ways does your company leverage its resources and expertise to implement AI in different areas of your work? What specific AI technologies or applications excite you the most?
Outside of our machine learning models, we focus on the aspects of our work that can be improved. My team specifically looks for new ways to apply AI to the integrations side of Riskified.
We’re looking for answers to questions like, “How can we enhance the integration flow for our customers?”
We’ve recently begun experimenting with chatbots that can answer very specific technical questions on some of our tools and integrations. The results so far have been exciting to see.
We’ve recently begun experimenting with chatbots that can answer very specific technical questions. The results so far have been exciting to see.”
What excites me most about AI is the advancements being made in the popular domain of language models, and some of the recent work and focus from OpenAI on GPTs. This will allow users to customize their own versions of GPT and feed it more knowledge and context about a given domain. I believe we’re in a period of rapid AI acceleration and I’m extremely excited to see what’s to come in the coming years.
What are the benefits of incorporating AI into company work? Are there any challenges or considerations that need to be addressed when implementing AI?
As we venture into the first stage of integrating AI into our operations, we unlock a multitude of benefits, primarily through task automation and optimizing team workflows. Our focus lies in empowering our workforce with the necessary knowledge and state-of-the-art tools to amplify our efficiency and bolster productivity.
The integration of AI involves a multitude of challenges, particularly with upholding best practices and ensuring the security of sensitive information. Establishing a framework of trust and security is crucial when working with external tools in AI.
There’s also a need for substantial resources if you’re hosting and training your own AI models. This includes cost and computational resources. I believe the main challenge in order to harness the true power of AI is understanding its capabilities. To effectively utilize these tools and strategically enhance key areas of your organization, a certain level of expertise is required. This expertise ensures targeted improvements and maximizes the impact of the organization’s efforts.
Hungryroot is a direct-to-consumer brand of fresh, healthy packaged foods.
How is your company embracing AI in its operations and making it an integral part of your company culture?
Since its inception in 2015, Hungryroot has embraced Al as a core force-multiplier. As a company, our promise is to help people eat healthy, save time and reduce food waste. All three of these goals are enabled by Al and machine learning at their core.
When we think of grocery planning, the task often leads to repetitive meals and recipes that we’re familiar with. At Hungryroot, we like to think about this differently. When customers sign up for the service, they take a short quiz that helps us know about their dietary preferences, tastes and dietary restrictions. This information is used to see the machine learning models that will determine which recipes and ingredients to send them.
As customers interact with the platform, indicating which recipes they like and which ones they don't, our Al learns their taste and constantly creates new meal plans and chooses new recipes for them. This leads to a more varied diet and better customer satisfaction as well as a good balance between what is familiar and trying something new.
As an added benefit, every ingredient is part of a plan, reducing food waste and helping the environment: Happy customer and happy planet, powered by Al.
In what ways does your company leverage its resources and expertise to implement AI in different areas of your work? What specific AI technologies or applications excite you the most?
At Hungryroot, we have built a team of experts both from the world of recommender systems and the world of operations research and optimization. If you think of the core premise of the company, we want to provide a personalized box of groceries to each customer each week.
No two boxes are the same, because no two customers are the same. This means that we need a recommender system that learns and adapts in real-time to each customer's preferences. It also means that we must guarantee that each customer's box does not contain food they are allergic to, or that they do not want.
The direction we are moving toward is the direction of hyper-personalization. Take the example of your favorite food. If we were to serve it to you every day, you would probably get bored. Food has an incredible time component. Most Al systems do not consider this component.
This means a combination of deep learning —time and sequence prediction — as well as generative Al — compose new recipes using combinations of ingredients and a core of "what good looks like" provided by our chefs — to create new, unique food combinations that never bore the customer.
What are the benefits of incorporating AI into company work? Are there any challenges or considerations that need to be addressed when implementing AI?
In today's world, customers expect services around them to feel personalized. They don't want to spend endless time browsing a website or using an app. Al, recommender systems, and personalization are here to stay. When implemented well, they make everything more efficient. This way, customers can be satisfied and not have to spend unnecessary time and effort on tasks that don't require it.
The biggest challenge in implementing an Al solution is the data itself. No Al system can function without appropriate data behind it. This is why we are spending a lot of time and effort building good data foundations, thinking clearly about what data we collect, and how to be respectful of our customers and their privacy.
When done well, this data can then be used in multiple ways, by multiple Al systems and models, to create a unified and harmonious solution. If, on the other hand, the data isn't handled correctly, customers will lose confidence in the company very quickly.
Many companies have had this happen to them. You’ve seen stories in the news about information leaks or creepy uses of Al that make us uncomfortable. Good data makes good Al.
Good data makes good Al.”
Work & Co creates digital products and services that define great brands.
How is your company embracing AI in its operations and making it an integral part of your company culture?
I’d say we’ve always embraced AI, but our ability to embed it into our work and processes — and see real impact — has totally transformed. Because Work & Co is a design and technology company focused on creating new, innovative interfaces — like chatbots, apps, websites & ecommerce platforms — we’ve been building on the foundations of AI for some time.
In the past year, the rise of generative AI has expanded knowledge, tools and access well beyond just our engineering team. Diving in and experimenting has become part of our culture across all our teams.
I think our decentralized approach to knowledge building has been really rewarding for our employees. Employees share work in our AI Show & Tell series and trade resources in our #club-ai on Slack.
We’re also integrating AI in more daily work. Like a lot of engineers, I was initially wary to adopt tools like Github’s Copilot. Can AI really improve the quality or speed of my code? For me, the answer is sometimes.
I use Copilot as an autocomplete to save a little time, but it really comes in handy when using languages I’m less familiar with. Now I don’t need to pause frequently to look things up; Copilot helps me without leaving my text editor.
In what ways does your company leverage its resources and expertise to implement AI in different areas of your work? What specific AI technologies or applications excite you the most?
We view generative AI through the lens of building and enhancing digital products, and we’re engaged in AI projects across retail, healthcare and travel. We regularly share experiments and client work on Work & Co’s dedicated AI Hub.
One theme we hear from clients is an interest in deploying generative AI within their organizations first, while increasing investment in responsible building of generative AI tools and features that meet customer expectations. These include: conversational interfaces, AI-powered booking and content generation and summarization.
We’ve also been tapping into AI to accelerate internal processes, automating repetitive setup tasks like scaffolding new design and development projects.
The potential to impact backend code generation has me particularly excited. I’ve been working on a tool that Work & Co is releasing soon called CodeSail.
This proprietary software helps launch products faster while promoting code quality through a variety of mechanisms. It gives development teams that manage many complex data sources a way to rapidly implement the code to integrate with, transform and expose that data. CodeSail generates code tailored to fit into a human-led development cycle.
What are the benefits of incorporating AI into company work? Are there any challenges or considerations that need to be addressed when implementing AI?
Using Natural Language Processing, AI helps us derive intent from an unstructured input, like a message from a user or an image. With the intent, we can hand off to deterministic systems not powered by AI, to do things like book a flight or add an item to the shopping cart. And, of course, AI frees up humans to work on the coolest, most challenging tasks.
A big temptation is to think about AI as the solution, then try to reverse-engineer a use case where you can apply it, rather than starting with a problem. What are you trying to streamline internally? What new features do you want to offer users? Think of the question and then decide if AI is the answer.
Remember that AI isn’t a magic bullet, and it doesn’t always know best. I think AI works best when we provide clear parameters and expectations.
Remember that AI isn’t a magic bullet, and it doesn’t always know best.”
Lastly: trust, but verify. One example is code generation. As developers, we often ask AI to write or update code for us: “Write a function that does X.” “Refactor this file.” Although most AI models will return something that looks like code in your language of choice, they aren’t compiling or running the code to check for errors. You have to do that on your end.
