From security concerns to talent constraints, implementing generative AI in the company can be a challenging and troublesome process. This underscores the importance of expert guidance and access to knowledge as even more valuable assets.
According to the "Enterprise Generative AI Adoption: C-Level Key Considerations, Challenges, and Strategies for Unleashing AI at Scale" report, companies are already taking grasp of the power of generative AI. The report states that 81% of respondents said that unleashing AI and ML use cases is one of their top 3 priorities. Also, 78% of respondents are planning to use the ChatGPT or other LLM-powered solution as a part of their AI transformation.
The adoption speed is not surprising considering the expectations. The report states that 57% of respondents expect a double-digit increase in revenue from AI/ML initiatives. Also, an additional 37% of respondents expect a single-digit increase.
However, implementing gen AI is not a walk in the park, and on top of that there are some crucial challenges to overcome during the adoption process.
Security - the most popular Large Language Model-based service, ChatGPT has already shown that the data leak is a real threat, and for big companies (like Samsung) the leak can be devastating.
Business Continuity - while emerging technologies can be highly disruptive, offering a clear advantage for startups, many well-established firms aim for the opposite. They prioritize maintaining smooth operations over causing upheaval in their existing business models.
Tech Readiness - Deploying advanced tech like generative AI demands a certain level of technological maturity within a company. Organizations lacking integrated systems or a comprehensive platform for data collection and management will find it challenging to fully harness the capabilities of generative AI.
Talent Constraints - The field of AI presents a unique set of challenges for businesses, primarily due to a limited global talent pool and the complex nature of the technology itself.
FOMO - last but not least; with constant updates about transformative tech and their potential benefits, businesses increasingly feel the pressure to adopt new solutions to stay competitive.
How to implement Gen AI in the enterprise
Considering the challenges above, it is not surprising that companies are working on implementing the new technology into their workflows. Yet it is far from easy.
“Modern AI architectures may be incompatible with existing infrastructure. This could cause substantial reworking of current systems. Additionally, siloed enterprise data makes curating comprehensive training data difficult. Generative AI must be woven into complex organizational workflows and processes thoughtfully, so as not to disrupt operations” says Roja Boina, Software engineer and sr advisor at Evernorth and a speaker during the upcoming Data Science Salon Miami conference. “Monitoring and updating AI models differs from traditional software, so upskilling IT teams is imperative. Compliance and legal risks around data, accountability, and bias cannot be ignored either” she adds.
The challenges with the infrastructure and the technical side of the project are also highlighted by Sahab Aslam, Head of Enterprise Data Science and AI at Myriad Genetics and a speaker during the upcoming Data Science Salon Miami conference.
“Integrating generative AI with existing systems can be tricky due to differences in data quality, technical complexities, poor infrastructure, and the need for domain knowledge and an experimenting mindset” she says. “Ensuring security, adapting to change, addressing interpretability concerns, and considering healthcare organizations with extra scrutiny for regulation, compliance, and ethical implications are vital for successful integration. Navigating these challenges requires a well-balanced blend of technical expertise, organizational adaptability, and a strategic approach.”
Human in the loop
Human oversight and expert knowledge are one of the ways to manage implementation and reduce the associated risks with the process.
“Deploying generative AI ethically and effectively requires ongoing human governance. Humans must continuously evaluate AI models to ensure quality and fairness. They need to manage data collection and labeling to minimize biases. Human review of outputs prevents the release of faulty or dangerous content. To integrate AI seamlessly into business workflows, human experts should work alongside machines” comments Roja Boina. “When novel situations confound algorithms, human guidance can resolve exceptions and refine the system. Clear communication about AI capabilities establishes realistic expectations with stakeholders. Most importantly, human judgment provides the moral compass to steer the entire AI lifecycle toward ethical outcomes. With the right framework enabling collaboration, human oversight, and AI automation can be combined for optimal performance” she adds.
“AI-generated outputs should be reviewed for accuracy and appropriateness. There is a need for a mechanism for humans to provide feedback to improve AI results. Also, company needs to ensure that AI is not producing harmful or biased outputs” comments Yifei Wang, Senior Machine Learning Engineer at Meta and a speaker on the upcoming Data Science Salon Miami conference.
“The role of humans should be the rigorous testing and oversight of model development and training to ensure biases are neither incorporated nor reinforced in GenAI outputs” adds Kevin Cochrane, CEO at VULTR and a speaker on the upcoming Data Science Salon Miami conference.
Managing the data and resources
One of the key concerns regarding the implementation process is access to talents and skills.
“Easier access to AI tools doesn't replace the need for AI experts; their expertise remains vital for managing complexities and ensuring accurate implementation. Rapid validation is still key, even with lower entry barriers, as it helps identify and rectify issues swiftly to improve model performance” comments Sahab Aslam “Involving all stakeholders and users of AI solutions is critical for effective change management, aligning expectations, and achieving widespread adoption and success” she adds.
“We need to hire or train experts in AI and machine learning and invest in hardware and software tailored for AI processing. Also, AI landscape evolves rapidly, so regular training and workshops are crucial” comments Yifei Wang.
Apart from access to data and talented individuals, compliance also presents an issue to solve.
“Privacy-by-design is key. In the current model, specific customer PII (Personal Identifiable Information ) is used to drive segmentation and targeting. In a GenAI model, all customer data is used only to build a general model where real-time customer observations can be used to infer the next-best experience” says Kevin Cochrane. “This means that individual customers PII no longer has to be explicitly referenced and used by individual marketers to make pre-defined assumptions and hard-code rules to personalize content or specific experiences.”
The incoming future
Despite the challenges in the implementation process, the experts are sure that the generative AI will be a game-changer for the business and society alike.
“In the next decade, generative AI will fundamentally reshape business models. Marketing content, product designs and other materials will be automated at scale by AI systems with minimal human input. Innovation will accelerate as design and engineering will become democratized. Hyper-personalization powered by customer data will create value. AI will also augment or replace humans in repetitive business processes, allowing work redesign. For decision-making, AI predictions synthesized from vast data will complement human judgment” comments Roja Boina. “Additionally, AI scenario simulation will enable risk assessment of strategies. While generative AI does routine, mundane work, humans will take on more creative roles curating and refining AI-generated content. Realizing this productivity and innovation potential requires responsible governance around data, models, and applications. With thoughtful implementation, generative AI can take businesses to the next level. But risks must be navigated” she adds.
Yet, despite the dynamic development of the field, there are also encouragements to exercise caution and approach the challenges of implementing Generative AI-based solutions wisely.
“It is okay to dream big in AI, but not okay to start big, the key is to find a business use case in marketing, acquisition, retention, and or experience for the customers. Choosing the wise use case and beginning the journey with data strategy, data engineering, data quality and the right skills in data science will enable to stand the AI/ML/Data Science program” says Shakeel Hye, Director of Data Engineering at TracFone Wireless and a speaker at the upcoming Data Science Salon Conference in Miami. “The very basic fact that we can easily forget is at the end of the day what problem are we solving?”
Summary
To keep the uncertainty at bay and maximize the outcomes, the implementation process of the generative AI needs to be done wisely. The best way to gather knowledge and spot the best practices is to attend conferences and knowledge events. The upcoming Data Science Salon Miami is the best and most convenient way to meet with industry experts, exchange experiences and stay on the cutting edge of the technology.
