Episode 78: Maximizing AI Adoption and Extensibility with Dona Sarkar
We’ve seen it happen: Artificial Intelligence (AI) has been a game-changer for the modern workplace, redefining how we approach problem-solving and innovation.
In a recent episode of the #ShiftHappens podcast, Dona Sarkar, Chief Troublemaker for Copilot and AI Extensibility at Microsoft, shared her invaluable insights on further redefining the AI experience we know with extensibility. She also dives into the challenges and opportunities of AI adoption, the importance of data hygiene, and the future of AI as an agent that can plan, recommend, and execute tasks alongside humans.
With her extensive experience at Microsoft and AI, Dona explores the exciting possibilities and challenges that lie ahead as we embrace the era of AI extensibility.
The Anthem for Our Transformative Times
The episode kicked off with a captivating discussion about the song “Lose Yourself” by Eminem, which Dona chose as a representation of the current state of technology. She drew parallels between the song’s lyrics and the intense pressure, challenges, and opportunities that companies and individuals face in this transformative period.
“I think that song is this generation’s rocky anthem because it is about seizing opportunities; your one shot, one opportunity. And that’s kind of where we’re at. The people who get a head start and grab the opportunities, whether they’re ready or not, are the ones who are going to lead.”
Navigating the Challenges of AI Adoption
As AI permeates various industries and domains, organizations grapple with the complexities of integrating this transformative technology into their existing processes and workflows. Dona Sarkar highlighted three key challenges that consistently surface during AI adoption:
- Data Security: Data security and privacy concerns are at the forefront of AI adoption. Organizations are often unclear about the data sources being used by AI models, raising questions about potential risks and vulnerabilities. Addressing this challenge requires transparency, clear communication from AI providers, and robust security measures to ensure data integrity.
- Understanding Generative AI: While predictive AI models have been in use for some time, generative AI models, which generate new content or outputs based on input data, pose a unique set of challenges. Organizations struggle with the non-deterministic nature of generative AI, where the output can vary with each interaction, making it difficult to explain or reproduce results consistently.
- Change Management and Skill Development: Adopting AI requires a significant mindset shift and the developing of new skills within organizations. Employees must learn to prompt AI models effectively, interpret outputs, and integrate AI into their existing workflows. This change management process can be daunting, as it requires continuous learning and adaptation to keep pace with the rapid advancements in AI technology.
Sarkar emphasizes the importance of providing clear guidance and using case-specific recommendations to organizations, helping them navigate the complexities of AI adoption and ensuring a smooth transition.
The Importance of Data Hygiene
One of the key takeaways from Sarkar’s insights is the critical role of data hygiene in successful AI implementation.
Sarkar emphasizes that organizations must prioritize data hygiene and ensure their data is clean, accurate, up-to-date, and free from conflicts and anomalies. This process may involve manual data cleanup projects, which can be intimidating but are essential for obtaining reliable and meaningful results from AI models.
However, as AI evolves, Dona emphasizes the importance of responsible AI practices and human accountability. If AI generates inadequate data or outputs, it falls on the human operators to identify, correct, or discard them. She states, “If an AI-created data isn’t good data, it is on the human who prompted it to acknowledge that it is bad data and take an action to delete or correct it. In this sense, individuals will assume the role of “managers” overseeing teams of AI assistants, requiring heightened awareness and diligence.
The Future of AI as an Agent
Looking ahead, Sarkar envisions a future where AI transitions from being prompted by humans to becoming an agent that can initiate plans, make recommendations, and execute tasks in collaboration with humans. This agentic model of AI would possess capabilities such as context awareness, memory retention, and the ability to connect to various data sources and plugins.
“It’ll be like a true personal assistant,” Sarkar explains. This vision of AI as an agent holds significant implications for software development and product design. Sarkar envisions AI filling in gaps where developers may struggle, such as user experience design or data organization while assisting with tasks like test case development and execution.
By leveraging AI’s capabilities in these areas, developers can focus on their strengths while relying on AI to augment their weaknesses.
Advice for Building Copilots and Implementing AI
For organizations considering building their own copilots or implementing AI solutions, Sarkar offers valuable advice:
- Learn to use existing copilots: Before attempting to build your own Copilot, it’s crucial to gain hands-on experience with existing AI solutions. By learning to use tools like Microsoft 365 Copilot, Word Copilot, or Dynamics 365 Copilot, you’ll understand their strengths, limitations, and the specific use cases in which they excel or fall short.
- Extend with your own data: Extend existing Copilots with your organization’s data to identify their limitations and determine whether a custom solution is necessary. This process will help you understand the data requirements and potential challenges in building a tailored AI solution.
- Start simple: If you decide to build your own Copilot, start with a simple approach to gain experience with AI development without diving into the complexities of responsible AI practices and security concerns.
- Embrace the AI developer role: If you aim to become an AI developer, be prepared to understand the entire stack, from choosing AI models and data sources to addressing responsible AI practices and security considerations. This role requires a deep understanding of the end-to-end AI development process.
The Road Ahead
As the episode came to a close, Dona’s insights left listeners with a profound appreciation for the transformative potential of AI extensibility while also highlighting the critical importance of responsible implementation, data hygiene, and continuous learning. The journey towards a future where AI acts as a true personal assistant is underway, and Dona’s guidance serves as a valuable roadmap for individuals and organizations alike.
Relevant Episodes
- Episode 88
- Episode 87
- Episode 86
How GenAI is Reshaping Business in Australia and New Zealand
- Episode 85
- Episode 84
- Episode 83