Democratizing AI: Benefits, Challenges and Governance

Democratization of AI is a hot topic as AI implementation broadly takes effect. But what does its adoption compel regarding governance – best practices around scope of access to AI applications, data security, accuracy, source traceability and intellectual property protection – which will be critical to achieving economic benefits and responsible use?

Tecnovy, a training and IT consulting company, frames up the situation: “Artificial intelligence (AI) technologies becoming accessible to everyone are at the forefront of technological transformation in today’s world. This process offers wider audiences opportunities for innovation and creativity, enhances efficiency in business processes and makes advanced technologies more accessible. However, alongside these positive developments, it also brings significant barriers such as data security, ethical issues and regulatory challenges.”¹

AI Exploration and Implementation Are Mushrooming

• 40% of companies globally have reported they are using AI within their company
• Another 42% are exploring the use of AI technologies in their business
• Over 50% of U.S. companies with more than 5,000 employees currently use AI. This number grows to 60% for companies with more than 10,000 employees.

 Exploding Topics 

AI Democratization Defined

AI democratization is commonly defined as providing AI technology access to non-technical users. The definition can also include the provision of “user-friendly resources such as pre-built algorithms, intuitive interfaces and high-performance cloud computing platforms.”² The goal is to enable in-house developers who lack AI development expertise to create machine learning (ML) applications and other AI-driven software, and to embed AI functionality into business-processes applications.

There is a caveat, however.

There must be guardrails or rules of the road, if you will, set and enforced by the enterprise. Most IT experts would agree that not every employee automatically gets or is granted on-demand access to AI. Enterprise AI implementation requires policies and protocols that stipulate who can access data and other assets based on their role. This process needs to detail the “reading in” and “reading out” of employees as they join or leave the organization.

AI Governance

ChatGPT was launched in late 2022. By January 2023, ChatGPT had become what was then the fastest-growing consumer software application in history, gaining over 100 million users in two months.³ The rapid adoption of ChatGPT spurred the release of numerous competing products such as Copilot, Gemini, Grok and others.

The breakneck rate at which these products have been officially and unofficially implemented has outpaced many organizations’ abilities to develop governance policies. An ISACA (Information Systems Audit and Control Association) Generative AI 2023 Pulse Poll found that “many employees at respondents’ organizations are using generative AI, even without policies in place for its use. Only 28% of organizations say their companies expressly permit the use of generative AI, only 10% say a formal comprehensive policy is in place, and more than one in four say no policy exists and there is no plan for one. Despite this, over 40% say employees are using it regardless – and the percentage is likely much higher given that an additional 35% aren’t sure.”⁴

IBM defines AI governance as “the processes, standards and guardrails that help ensure AI systems and tools are safe and ethical. AI governance frameworks direct AI research, development and application to help ensure safety, fairness and respect for human rights. Governance also aims to establish the necessary oversight to align AI behaviors with ethical standards and societal expectations so as to safeguard against potential adverse impacts.”⁵

 As AI implementation continues to expand, it’s imperative that organizations develop, communicate and enforce rules of the road to ensure the appropriate use of AI. This will enable:

• Gaining the efficiencies, innovation and other benefits of the evolving technology
• Mitigating negative consequences such as intellectual property infringement, inaccuracies due to the use of flawed data or models and/or illegal or unethical use of AI

One way your organization might begin this process is by reviewing policies that were defined for the democratization of data, using them as guides for AI governance policies.

The National Institute of Standards and Technology (NIST) has developed an Artificial Intelligence Risk Management Framework as well as other AI-related documents to help guide your process. ISO/IEC 42001 has also developed an AI management system standard, providing organizations with a framework to establish, implement, maintain, and improve AI management systems, focusing on responsible and ethical AI development and use.

IT consulting and services provider Infosys offers advice on establishing control for data, AI models and usage and consumption. In a blog by Sunil Senan, SVP of Infosys, their three steps to building data and AI governance are:

• Outline an organization-wide strategy
• Establish an operating model
• Operationalize the data and AI governance framework

We encourage you to follow the link provided to get a full understanding of the approach, which is based on principles of responsible AI use: trust, ethics, privacy, compliance and security.

AI Democratization Potential Benefits and Challenges

Like all technologies, AI presents organizations and end users with a mix of benefits and challenges. When discussing AI democratization, key benefits and challenges include:

Benefits

• Increased employee productivity
• Enhanced idea generation, content creation and innovation
• Better and faster data analysis
• More informed business forecasting
• Improved risk management

Challenges

• Evaluating the pros and cons of implementing AI based on specific use cases
• Preventing access to AI models, data and tools by bad actors
• Preventing development of biased models and/or malicious model modification
• Minimizing the potential for intellectual property infringement or inaccuracies
• Minimizing the potential for leveraging AI as a cyberattack surface to access, hijack or damage internal networks, information or infrastructure
• Preventing shadow IT – i.e., the use of any software, hardware or IT resource used on an enterprise network without the IT department’s approval knowledge or oversight
• Ensuring compliance with regulatory requirements such as HIPAA, the U.S. Privacy Act of 1974, GDPR, California Consumer Privacy Act (CCPA) and other state consumer and data privacy acts as well as additional industry-specific regulations regarding the protection of sensitive data
  
  

Critical Factors for Enterprise AI Implementation Success

As high-density computing goes, most current applications don’t get much more demanding than AI. For instance, on average, a ChatGPT query uses nearly 10 times as much electricity to process as a Google search.⁶ According to Lucas Beran, Research Director at Dell’Oro Group, “The average rack power density today is around 15 kW/rack, but AI workloads will dictate 6 - 120 kW/rack to support accelerated servers in close proximity."⁷ And Goldman Sachs Research forecasts AI will drive a 165% increase in data center power demand by 2030.⁸

This will require substantially more power and cooling capacity in data centers. Beyond an environment supporting very high-density computing, successful implementation of AI applications will depend on:

• Training and upskilling for business users and developers not yet familiar with AI basics
• Rapid, secure interconnection between users and cloud-, premises- and partner-based applications and data stores
• Physically secure and cyber-secure facilities and IT equipment
• AI usage monitoring to ensure compliance with organizations’ AI governance policies

Increasingly, organizations are finding it impractical, inefficient and very costly to meet these requirements – particularly the escalating infrastructure demands – in on-premises deployments. Furthermore, the costs for implementing AI in the cloud can be higher than desired, especially after models have been trained and during inferencing, the process of running live data through a trained AI model to make a prediction or solve a task. All of this contributes to a growing trend to move AI workloads to colocation data centers.

Colocation Becoming a Preferred Choice for AI Hosting

Among organizations’ reasons for hosting their IT hardware, software and critical overall operations in colocation facilities are the need for:

• Robust, secure, redundant infrastructure expressly designed to support high-density computing
• Resilient, low-latency connectivity environments (from cloud to edge) including inter-site connectivity that enables virtually unlimited dynamic scalability and the ability to deploy distributed workloads
• Ability to rapidly access and transport massive volumes of data
• 24/7 onsite availability of expert staff to help with planning, implementation, monitoring and remote hands services
• Reduced total cost of operations

The rapid evolution of AI and its application to nearly every aspect of business and life means that AI governance policies will need to keep pace to ensure the integrity of new applications and capabilities. It also means the infrastructure requirements to support this evolution will need to keep pace, which is an additional reason to host your AI, IT assets and applications in a high-performance colocation environment.

Know More

Contact us to find out how CoreSite’s AI-ready data centers can streamline your operations, reduce total cost of operations and drive a competitive edge in your AI strategies.

Learn what 300+ IT decision makers consider important to customers, partners and their enterprise:

Download the 2024 State of the Data Center Report here.

References 

1. Democratizing AI: Multiple Meanings, Goals and Methods (source)
2. AI democratization; TechTarget (source)
3. ChatGPT reaches 100 million users two months after launch (source)
4. AI Policies are Low, Use is High, and Adversaries are Taking Advantage, Says New AI Study (source)
5. What is AI Governance? (source)
6. AI and the Data Center: Driving Greater Power Density (source)
7. Ibid
8. AI to drive 165% increase in data center power demand by 2030 (source)