top of page

What are your 2024 GenAI Adoption Plans?

Nov 14, 2023

3 min read

0

1

0


2023 started with the announcement about ChatGPT! By the 3rd quarter, enterprises had adopted and scaled their solutions using GenAI in creative ways. Two examples illustrate the scale of operations that support a simple text query through a GenAI-based interface.


Uber: Links historical customer preferences to make recommendations in real time using 5.3K models in productions, 10M online predictions per second to provide destination suggestions, home feed ranking, and document scanning among a few use cases.

Pinterest: A search for certain images, defined using text, responds with recommendations. This recommendation is backed by a complex system of models that capture, process, and build responses for 428 million active users across billions of pins (images) in the repository.


Today, GenAI's impact on our lives continues to mesmerize many of us. I have listened to thought leaders in data science conferences, cybersecurity, fintech, management consultants, and the discourses on LinkedIn. I, like many of you, am mentally exhausted from hearing the mixed emotions circulating. As the year ends, we should reflect, filter, and refine our assessment of GenAI.


In 2024, how should we prepare ourselves to handle GenAI strategy?


(Source: Tecton MLOPS Conference 2023: Uber case study)


Start with the facts. Start small. Grow your scope iteratively! Some of us may have ignored AI in the past, but ignoring GenAI would impact your industry standing very quickly!

FACTS


What do we know about GenAI tools and applications?

  • Large Language Models (LLMs) are at the heart of GenAI, enabling these systems to understand and generate human-like text.

  • These models learn from an extensive array of internet data, enhancing their ability to respond in ways that align with human expectations.

  • Major tech companies have invested significantly in developing AI that can understand and translate various types of data, much like the human brain does.

  • AI is the umbrella of techniques, tools, and approaches of which Machine learning and GenAI are 2 examples. GenAI helps to democratize access to AI/ML models.

  • LLM models today have some weaknesses, for instance, LLMs can sometimes generate inaccurate or nonsensical information, a challenge known in AI as ‘hallucinations’. (There are others I am not addressing here).

  • These AI applications can quickly search, find, and combine information from many sources, which is incredibly useful for businesses. LLMS are trained on large amounts of data and are used to process large data sources with ease.

  • It is not a tangible asset that you purchase and start using. There is work that needs to be done to train them with your business context.

  • For more accuracy and customizations, a period of training, recalibrating, and guidance will need to be provided to the LLM algorithms.

  • Enabling GenAI is a journey that blends enterprise knowledge with specialized skills to build an enterprise-relevant application.

  • It needs data to work! Unstructured documents are preferred, structured documents need some additional business guidance.

  • As organizations accelerate their efforts to join the GenAI wave, there are some considerations to keep in mind.

“To be an AI first company you have to be a data first company”
  1. Data Security

  2. Data management: Storage, access, usage, and transformations

  3. AI adoption: Progressing from initial POC to full implementation

  4. Design for modular, reusable pipelines, and efficient analytic workflows

  5. Develop data and analytic skills, promote awareness, and encourage training, scrutiny, and innovation

  6. Foster a culture of adoption ingrained across all levels and departments

  7. Establish an engagement model that emphasizes collaboration, sharing, discovery-driven learning, and a focus on business results


Data has been compared to the new token for economic activity. Modern business giants have built their successes by monetizing data in the form of insights and data products. Securing these data assets has a professional and social responsibility dimension that is being proposed by the NIST AI standards playbook [2]. Misuse or improper use of data can also lead to ethical violations that need to be addressed through data management activities. Data used to train AI models or applications should be guided by unbiased business views.


Key considerations include ensuring data security, managing the quality and organization of data, and using advanced tools to monitor large volumes of data.


Crawl, Walk, Run: Start with basic data analytics, advance to machine learning, and then fully embrace AI for optimal results. The capabilities of LLM in GenAI applications are built upon machine learning models, which are grounded in statistical analysis. If the organization has not gained the analytic maturity to have infrastructure and processes to support such analytic activity, it is likely to face frustration in creating strategic initiatives based on this new technology. A few key needs are,

  • Good metadata and business documentation of data flows/ lineage

  • Data accessibility for analysis and performance improvements of algorithms

  • Dynamic and near real-time monitoring of data flow quality

  • Tracking, correcting, and updating processes.



#genai #datafirst #datamanagement #datastrategy #aiadoption.


Reference:

Wolfgang Goerlich, Advisory CISO at Cisco Secure Finance and risk cybersecurity Summit 2023

NIST RMF, https://csrc.nist.gov/projects/risk-management/about-rmf