Managing visual content on social media, especially user-generated content, always present a formidable challenge for businesses. With a constant influx of images and videos flowing through these channels, brands invest significant time and effort in filtering out inappropriate content that could damage their image. Complicating the process further are bots, from popular internet brands, actively scraping images or profiles.  Google’s Generative AI capabilities offer a solution. But how? And what business challenges necessitate this type of implementation?

User-generated content (UGC) can be a valuable asset for businesses/retailers. It helps to build brand awareness, generate leads, and drive sales. Nevertheless, it can also be a source of risk. Inappropriate or fake products in the space of UGC can damage a retailer’s reputation and lead to loss of sales.

For mitigating these risks, retailers need to have a strategy in place for moderating UGC on social media channels. This strategy should include:

• Clear guidelines for what is and is not acceptable UGC should be communicated to the users in a clear and concise manner.
• A process for reviewing and approving UGC should be efficient and effective, enabling retailers to act quickly.
• A team of moderators, trained to review and approve UGC, should be knowledgeable about the retailer’s brand and values, enabling them to make decisions quickly and accurately.

Enable a business service using Google Generative AI models for UGC on social media channels

In addition to the traditional moderation techniques, retailers can also use Google Generative AI to help them moderate UGC. Even better, Capgemini has formed a key technology partnership with Google Cloud Platform that launched last year.  This opened a multitude of new Generative AI technical capabilities, ranging from “ready-to-use” solutions to fully customizable options, all made possible through Google Cloud. Additionally, the Global Capgemini Google Generative AI Center of Excellence (CoE) is fully operational, enhancing customer use cases and applying best practices to leverage the Generative AI services available within Google Cloud’s Generative AI model garden. Find out more here.

But how generative AI models on the Google Cloud Platform can help retailers? Here’s a concrete use-case to illustrate.

Automatically identifying and only approving appropriate UGC helps retailers save time and resources. Furthermore, it ensures that their social media communication channels remain free of harmful content.

Utilizing Google Cloud Platform model capabilities such as Gemini-pro vision, Vision QnA, and specialized models like content moderation helps mitigate the risks associated with UGC, minimizes workload, and reduces costs. Additionally, it enables retailers to leverage UGC to their advantage, perhaps by treating the images as needed to transform them and generate a positive impact instead.

Here are some specific examples of how Google’s generative AI models can be used to moderate UGC on social media channels:

• A retailer could use Google’s Generative AI models to automatically identify and remove comments that contain profanity, hate speech or other offensive language.
• A retailer could use Google’s Generative AI models to score pictures or videos that can damage brand images.
• A retailer could use Google’s Generative AI models to automatically identify pictures or images that do not fit a set of business rules in different markets, regions or countries. This minimizes costs and time-effort required for moderators to carry out this process.

What retailers are looking for in Generative AI?

Generative AI has emerged as a transformative force in the retail industry, offering new avenues to enhance analytics capabilities and drive customer engagement along with the use of the rich internal data stored by retailers.

In the early days, before the widespread use of databases to store data became the norm, retailers used to rely on manual methods to gather insights into customer behaviour and product trends. These methods included conducting customer surveys, analysing store receipts, and leveraging a loyal customer base to understand grocery preferences and shopping habits. Additionally, retailers would incorporate information on population trends and factors such as driving distances from stores to further inform their strategies.

However, with the advent of Point of Sale (POS) systems and digital data storage mechanisms, the landscape of retail data analysis underwent a significant transformation. Retailers found themselves equipped with more sophisticated tools and platforms that allowed for the collection and analysis of a multitude of attributes related to customer behaviour and product performance. Data collection expanded across various operational areas, driven not only by a desire to understand consumer preferences but also by legal requirements mandating the storage of certain information over extended periods.

This evolution has culminated in retailers gaining access to a wealth of internal data that encompasses a broad spectrum of customer interactions and transactional details. Moreover, retailers have increasingly recognized the immense potential of harnessing both internal and external data sources to derive actionable insights. These insights can drive decisions related to inventory optimization, marketing strategies, pricing models, and overall operational efficiency.

As the retail industry continues to evolve, the volume and complexity of stored data continue to grow exponentially. Retailers are continually exploring innovative ways to leverage this vast repository of information to gain a competitive edge in an increasingly dynamic market landscape. The ability to effectively analyse and interpret data has become a cornerstone of success in modern retail, enabling businesses to anticipate trends, personalize experiences, and adapt strategies in real-time.

Then arrived Gen AI. While the initial applications in retail centred around chatbots for customer assistance, the scope has expanded significantly to encompass a diverse range of practical applications. Retailers worldwide are increasingly leveraging generative AI to elevate customer personalization and satisfaction levels with the help of the plethora of platforms that exist today. They can get insights on customer sentiments and product performance reviews, all without the need for explicit campaign efforts by leveraging internal/external reviews and from posts on social media.

Gen AI could also help companies streamline operations and improve overall efficiency. For instance, many retail and consumer products companies have recognized the potential of AI in detecting defective materials or storage issues, thereby enhancing the product quality and reducing waste. Furthermore, AI-powered fraud detection systems have become instrumental in safeguarding retailers against fraudulent activities, thereby bolstering trust and credibility among consumers.

The integration of generative AI into retail operations represents a paradigm shift in how businesses interact with and understand their customers. Through its use, retailers can unlock deeper insights into consumer behavior, preferences, and purchasing patterns. Armed with this knowledge, retailers can tailor their offerings and marketing strategies to better meet the evolving needs and expectations of their target audience.

Implementation of responsible AI is a priority for each project

Capgemini internally promotes mandatory group guidelines for Generative AI projects because the usage and regulatory landscape of Generative AI methods and toolsets vary significantly across our geographies, industries, use cases, processes, context, and capabilities for implementation. Therefore, our teams continuously conduct a proper assessment of our existing policies.

As Generative AI technology is evolving rapidly, guidelines and policies are regularly updated and communicated. Therefore, all actions taken regarding Generative AI at Capgemini using Google Cloud adhere to responsible AI principles, ensuring the success of our customer use cases.

Let’s begin exploring and adding value to your organization’s use cases. We are thrilled to hear your thoughts on implementing Generative AI projects across industries and scaling them using the Google Cloud Platform. I’m confident that the capabilities provided by Capgemini Insights & Data Nordics will pave the way to success in your use cases. Feel free to reach out to us; together, we can implement and architect Generative AI projects for your data teams.

Generative AI is the pinnacle of data science. It will boost profits, reduce costs, and help you expand into new markets. To take full advantage of generative AI’s capabilities, train your models on all your data.

The world is being transformed by AI-assisted medicine, education, scientific research, law, and more. Today, researchers at the University of Toronto use generative AI to model proteins that don’t exist in nature; pharmaceutical giant Bayer now uses generative AI to accelerate the process of drug discovery; and education provider Khan Academy has developed an AI chatbot/tutor, Khanmigo, to personalize learning. And with each passing day, the list of AI use cases across all industries only continues to grow.

According to the Capgemini Research Institute, nearly all (96 percent) of executives cite generative AI as a hot topic of discussion in their respective boardrooms. Generative AI is not just used as an aid to surface information the way a search engine does; with generative AI, organizations can combine their proprietary data with foundation models that have been pre-trained on a broad base of public data to create a sustainable competitive advantage.

Generative AI then becomes the most knowledgeable entity within your organization.

However, as with all analytics, generative AI is only as good as its data. To fully leverage AI, an organization needs a solid data foundation and organizational norms that facilitate responsible and effective use of data.

Data readiness for generative AI depends on two key elements:

1. The ability to move and integrate data from databases, applications, and other sources in an automated, reliable, cost-effective, and secure manner
2. Knowing, protecting, and accessing data through data governance

Automated data pipeline platforms, like Fivetran, allow enterprises to capture all of their data, irrespective of the source platform. These automated tools reduce the friction and overhead required to maintain the flow of data to continuously train generative AI applications.

OPERATIONALIZING GENERATIVE AI

To operationalize generative AI effectively, organizations must establish a solid foundation of automated, reliable, and well-governed data operations. Generative AI requires a modern and scalable data infrastructure that can continuously integrate and centralize data from a variety of sources, including both structured and semi-structured data.

However, as businesses start to operationalize generative AI, they may encounter a number of challenges.

• Data quality and preparation: Generative AI models are only as good as the data they are trained on. It is important to ensure that the data is high-quality, clean, and well-organized. This includes identifying any potential biases in the data that may distort the outputs of any model trained on it.
• Security and governance: Security and governance in the context of generative AI concern masking sensitive information, controlling data residency, controlling and monitoring access, and being able to track the provenance and lineage of data models.
• User experience: It is important to design user interfaces for your model that make it easy for people to interact with your models.
• Scalability: It is important to choose a generative AI platform that can scale to meet your needs at a reasonable cost.

Generative AI models are trained on massive datasets of text, code, images, or other media. Foundation models, which are off-the-shelf generative AI models that are pre-trained on large volumes of (usually public) data, may be specialized by industry or use case. Choosing the right foundation model can have a significant impact on performance and capabilities. For example, a foundation model that specializes in code generation will do so in a more comprehensive and informative way than a model that is trained on a general dataset of text. Other specialties of foundation models may include sentiment analysis, geospatial analysis, image generation, audio generation, and so on.

While you can easily make use of pre-trained, publicly available AI models, your data is a unique asset that differentiates your organization from the competition. To make the most of it, you must additionally supply foundation models with your business’s unique context.

With access to your organization’s accumulated data, a properly tuned generative AI model can become the most knowledgeable member of your organization, assisting with analytics, customer assistance, sales and marketing, software engineering, and even product ideation.

The Fivetran product team leverages generative AI and natural language processing technologies to develop Fivetran Lite Connectors in a fraction of the time of Fivetran’s standard connectors, while ensuring the same high quality, data integrity, and security customers expect from Fivetran.

In addition, several notable organizations have already found practical ways to use generative AI. Global commercial real estate and investment management company JLL recently rolled out a proprietary large language model that employees access through a natural language interface, quickly answering questions about topics such as an office building’s leasing terms. Similarly, the motor club in the US, AAA, now uses generative AI to help agents quickly answer questions from customers. Of the 100 tech companies profiled in the Forbes Cloud 100, more than half use generative AI.

According to Carrie Tharp, VP Strategic Industries, Google Cloud, “Generative AI opens up a new avenue, allowing people to think differently about how business works. Whereas AI and ML were more about productivity and efficiency – doing things smarter and faster than before – now it’sabout ‘I can do it completely differently than before.’”

Until enterprises get the data right, the nirvana of asking generative AI app-specific and contextual organizational questions in a “Siri-like” way will remain elusive. Get the data right, and it opens up possibilities for all analytics workloads, including generative AI and LLMs.

To make full use of an ever-expanding roster of powerful foundation models, you must first ensure the integrity, accessibility and governance of your own data. Your journey into generative AI and the innovation and change it can bring will be fueled by high-quality, usable, trusted data built on automated, self-healing pipelines.

“GENERATIVE AI APPLICATIONS ARE ONLY AS GOOD AS THE DATA THAT POWERS THEM.”

INNOVATION TAKEAWAYS

OPERATIONALIZE GENERATIVE AI

Operationalization begins with centralizing data and modernizing the data stack to include all available data.

AUTOMATED DATA ACCESS

By automating data pipelines, enterprises can focus on improving data models and algorithms to accelerate the efficacy and ROI of investing in a generative AI application.

CREATE AN UNFAIR ADVANTAGE

Generative AI trained on your data will provide insights and guidance driven by your data, creating a unique competitive advantage that cannot be replicated.

Interesting read? Capgemini’s Innovation publication, Data-powered Innovation Review | Wave 7 features 16 such fascinating articles, crafted by leading experts from Capgemini, and partners like Aible, the Green Software Foundation, and Fivetran. Discover groundbreaking advancements in data-powered innovation, explore the broader applications of AI beyond language models, and learn how data and AI can contribute to creating a more sustainable planet and society.  Find all previous Waves here.

Could NVIDIA’s AI and GPU dominance be at risk?

Have you heard about #LPUs, or Language Processing Units yet? This new kid on the block is 10x faster, 90% less latency, minimal energy vs. Nvidia GPUs. What does this mean for #ai‘s #genAI future?

I explore this massive shift in my latest article. Discover how Groq could redefine AI hardware efficiency and challenge the current giant.

Traditional BI and AutoML platforms enable self-service access to mountains of high-fidelity data, but they fail to deliver actionable insights to drive better business outcomes.

Enter autonomous analytics, such as Aible, which can surface anomalous KPIs and trends and the key drivers as actionable insights. They complement popular BI tools to guide analysts to swift, precise insights. For decades, firms have struggled to make BI work to drive business outcomes. Today, end users have access to more data than ever before but not the actionable insights necessary to help steer the business to the best possible outcomes. As Herbert Simon, Nobel Prize and Turing Award winner, noted back in 1971, “Wealth of information creates a poverty of attention and a need to allocate that attention efficiently among the overabundance of information sources that might consume it. “In 2013, Clayton Christensen (of Innovator’s Dilemma fame) et al. wrote in the Harvard Business Review article “Consulting on the Cusp of Disruption” that, “The big data company BeyondCore can automatically evaluate vast amounts of data, identify statistically relevant insights, and present them through an animated briefing, rendering the junior analyst role obsolete. “BeyondCore eventually became Salesforce Einstein Discovery and inspired the modern augmented analytics wave. The team behind BeyondCore has now started Aible, which takes autonomous analytics to the next level by marrying it with generative AI.

GENERATIVE AI-POWERED INSIGHTS DRAMATICALLY FASTER, WITH MINIMAL COST

Aible’s patented technology automatically explores millions of cuts of data in minutes, at a fraction of the cost of legacy of industries have published case studies showing the scalability, cost-efficiency, and speed of the Aible platform. A Google blog post entitled Aible’s serverless journey to challenge the cost vs. performance paradigm explains how Aible delivered such analytics efficiencies on BigQuery, but such efficiencies can be expected on any platform.

As an illustration of this capability, consider data for an outbound call-center that makes calls to offer co-branded credit cards to prospects. Aible enables autonomous analytics to use transaction data, enriched with raw operational data such as agent attributes like education, experience, call quality, and scores, to understand how and to precisely what extent they drive KPIs like conversion rates and offer a focused view into what can be done to address opportunities for improvement.

For example, it is helpful to combine agent attributes in the credit card call center illustration above into agent segments that can be used as engineered features in the analysis to better understand the precise extent to which cohorts drive outcomes. Aible automatically generates and evaluates such cohorts to determine the “net effect” of each combination on the KPI of interest. The most significant drivers of the tracked KPIs are reflected in a circular Sankey visualization, which shows the net effects of all variables at a glance. Alternatively, for business users, Aible can auto-generate traditional dashboards with the key charts organized in order of their impact on the KPI.

AUGMENT BI PLATFORMS

Autonomous analytics platforms can work in standalone mode but work best as complements to popular BI platforms. Aible automatically evaluates raw and engineered data, determines key insights, and auto-generates the KPI driver view. It can even export native BI tool dashboards, to be embedded into yet other BI tools such as Tableau and Power BI. In this design, we retain all the capabilities of the BI platform, with Aible’s analytics engine helping us find the key statistically-sound insights behind the scenes. This also suggests a new way for working for analysts and business leaders. The BI platform should be configured to monitor KPIs and alert analysts and/or business leaders about key insights related to the KPIs with additional information on which of the patterns surfaced are credible. The analysts can use these reports as starting points to pull further data. When used in this manner, the Aible engine can be thought of as driving BI for enterprise performance analytics. The BI platform should also be leveraged when analysts are looking to get reports in broader contexts than understanding KPI drivers.

“AUTONOMOUS ANALYTICS PLATFORMS CAN WORK IN STANDALONE MODE BUT WORK BEST AS COMPLEMENTS TO POPULAR BI PLATFORMS.”

DATA GRANULARITY IS ESSENTIAL

To get the most out of autonomous analytics platforms, the transaction data must be at the most granular level possible and then tagged with all raw and engineered attributes, such as from interaction and segmentation analysis, that make sense for that level. The autonomous analytics system can also auto-generate engineered features. Aible uses this data to generate key insights from the data and enable users to ask business questions such as “How can I improve sales to Gen Z customers?” instead of just analytical questions that must be translatable to SQL.

The engine will monitor KPIs, identify significant trends and shifts in the KPIs, and highlight statistically credible alerts. It also generates visuals to explain the single or multi-variate patterns in a matter suited to the user persona – from circular Sankey charts and mind maps for expert analysts, to dynamic dashboards and generative AI storytelling for business users.

The Aible AI engine provides a list of drivers in a circular Sankey chart, with any overlap clearly indicated. The same view also provides ordered lists of drivers and corresponding charts showing the exact impact of each on tracked KPIs. In addition to this, the AI engine also provides a view into the behavior shift and population shift for each driver for period-to-period results, where the rate effect reflects change attributable to the average change in the value of a cohort, whereas the mix effect reflects the change in the proportion of that cohort.

Aible includes a generative AI platform that uses foundation models such as PaLM 2 and GPT-4 to allow users to ask the Aible engine questions about the drivers and their precise extent of impact behind KPIs in plain English; these generate a well-articulated response, also in plain English. Such a system gives everyone at firms the power to interrogate the engine about business questions related to the KPI and the drivers behind KPIs or KPI changes.

A properly implemented platform can provide insights into which customer segments and/or employee segments are driving observed KPIs as applicable. Aible provides the necessary guardrails for enterprises to securely scale insights from generative AI responses with its ability to automatically doublecheck the output to reduce hallucinations (where generative AI creates inaccurate facts).

Aible can deliver insights in near real-time, allowing firms to respond to market threats and opportunities much faster, and in a very surgical fashion to optimize outcomes.

INNOVATION TAKEAWAYS

AI FIRST

An AI first approach automatically analyzes raw data across millions of variable combinates – group-by and drill-down charts– in a matter of minutes and costing cents.

THE ART OF STORY TELLING

Generative storytelling automatically highlights key insights in the data while double-checking the generative AI for hallucinations.

HAVE IT YOUR WAY

Insights can be consumed in multiple ways, from conversational interfaces to dashboards and mind maps.

Interesting read?

Capgemini’s Innovation publication,Data-powered Innovation Review | Wave 7 features 16 such fascinating articles, crafted by leading experts from Capgemini, and partners like Aible, the Green Software Foundation,and Fivetran. Discover groundbreaking advancements in data-powered innovation, explore the broader applications of AI beyond language models, and learn how data and AI can contribute to creating a more sustainable planet and society. Find all previous Waves here.

While telecoms networks were largely dependent on manual systems and processes before they had software-driven rules and automation built in, today, new innovations are set to drive tangible business benefits across the telecoms industry. Today, by leveraging machine learning (ML) and data and analytics, networks have the autonomy to take certain low risk, rules-based actions. However, telco networks look set to evolve once more, as the promise of a ‘self-serving’, ‘self-fulfilling’, and ‘self-assuring’, or autonomous networks draws closer.

The increasing demand for improved customer experience is placing pressure on telcos and their network resources. Despite investments in 5G and fiber broadband improving the reach and quality that networks provide, the increasing number of mobiles and devices connecting to networks, and the subsequent rise in volume of data means that telco networks are now more complex than ever. Essentially, the management of networks is increasingly going beyond the reach of manual operations. This is where autonomous networks can drive tangible business value for telcos, providing a springboard to improved customer service, as well as more efficient and more sustainable networks.

Investment in autonomous networks is on the rise

According to the latest research by the Capgemini Research Institute, telcos are expected to invest $87 million on average in autonomous networks over the next five years. [KA1] 

Today, according to TM Forum’s taxonomy of autonomous networks, the majority (84%) of telcos have either a level-1 or level-2 autonomous network. We know from our research that this is set to increase, with 61% of telcos aiming for at least level-3 autonomy over the next five years. Currently, Europe leads the way in overall network maturity, with over half (51%) of telcos at level 2 network autonomy, although North America has the greatest proportion of telcos at level 3, at 14%. Despite the clear eagerness showcased by telcos for greater network autonomy, the majority of the use cases are still at proof-of-concept stage. Of those surveyed, adaptive/dynamic network policies for changing conditions are the most popular use case at proof-of-concept stage (46%), followed slice optimization and SLA assurance in RAN/ORAN (40%).

Subverting the challenges

Despite the innovative use cases that autonomous networks can deliver, there remains considerable barriers to adoption. Chief among these is cultural issues, with just over half (51%) of telcos citing that employees don’t have the right mindset to undertake such a shift. Although, when you consider that just 17% of telcos have a well-defined autonomous networks transformation strategy, and fewer than one in five organizations have appointed a dedicated leader, it’s no surprise that cultural issues persist. However, the barriers to adoption aren’t limited to just cultural issues. In fact, we found that issues around technological maturity is slowing down telcos’ autonomous network journeys. Our survey found that 48% of telcos flagged technology integration was a noteworthy issue. Additionally, 33% and 25% of telcos flagged that technology maturity and the lack of skills among the internal workforce as significant barriers to adoption.

Gen AI and sustainability benefits front of mind

Over the past year, generative AI has risen from emerging technology status to the center of the boardroom conversation. All industries alike are assessing their operations to understand where they can integrate the technology, and telecoms is no different. According to our survey, three in five telcos are exploring generative AI for autonomous networks, and one in ten has implemented generative AI for networks at partial scale. Generative AI strikes that crucial balance for telcos as a cost reducer, and efficiency driver. We know from our research that the most popular use cases are complex event processing and dynamic bandwidth and path selection. On a more granular level, generative AI can assist telcos with translation, fraud resolution and model training.

However, it’s not just generative AI-related benefits that telcos are reaping and in fact, those telcos that are moving faster on their autonomous network journeys, are realizing the benefits. In fact, in just the past two years, telcos have on average achieved a 20% improvement in operational efficiency and 18% reduction in OPEX through autonomous networks. The survey also finds that telcos are expected to invest $87 million in autonomous networks over the next five years, but that this would amount to $150 million – $300 million in OPEX savings. And the benefits aren’t limited to simply cost savings, with the sustainability benefits front of mind for many telcos.

Today, it’s crucial that businesses have sustainability built into their core. Energy accounts for 30-40% of telco OPEX, with the Radio Access Network (RAN) accounting for 80% of network energy consumption. And those who transition to a higher level of autonomous network can expect a reduction of somewhere between 7.5%-15% reduction in their networks carbon emission. For instance, Telefónica Group has successfully reduced its energy consumption by 7.2% between 215 and 2022. While this initial number may seem low, when you consider that their network traffic has increased seven-fold over the same period, they have reduced their overall emissions by 51% over the period. As generative AI continues to cement itself as a key innovation across the telecoms landscape, we’re going to see these results improve much quicker.  

Accelerating the transition

As aforementioned, just 17% of telcos have a comprehensive autonomous networks strategy in place. Those who have this strategy in place can expect to realize the benefits of autonomous networks much sooner. With this in mind, I wanted to take a moment to explain what a comprehensive autonomous network strategy consists of.

• Strategy & roadmap: Here it’s critical that telcos establish the business case early on, so that they can secure the necessary finance and build a strategy that simultaneously resonates at both a global and local level.
• People: With innovations comes the need for new skillsets. Telcos should work to bridge the skills gap in areas such as AI by upskilling and reskilling the current workforce. As I mentioned earlier, the cultural shift presents one of the biggest barriers. By reorganizing systems, processes and tools, telcos can guide their organizations through to a new, more efficient operating model.
• Technology: Technology integration issues were high on the agenda of telco executives. To combat this, they should ensure they have an end-to-end view of their data landscape and leverage the cloud for virtualization where possible. Atop of this, telcos should invest time into establishing robust data-governance and data-management frameworks.
• Pace of transformation: As always, the pace of transformation depends on the maturity of the technology. For instance, beginner telcos should consider which network domains and use cases to prioritize, whereas those midway through their journey should double down on investment and focus on scaling.
• Innovation: Open and disaggregated networks open the door to new innovative use cases. Telcos should experiment with emerging technologies such as generative AI, metaverse and digital twins to ensure they enhance network efficiency.

The operating model of networks is going through a generational shift, from one managed by human operators, to an autonomous one whereby AI and data take center stage. While this shift requires significant investment, telcos should welcome it with open arms. Autonomous networks provide strike that all important balance of reducing costs, increasing efficiency, and contributing to a more sustainable future.

We look forward to meeting you at the Capgemini booth (2K21) in Hall 2 at MWC Barcelona from February 26th to February 29th

Do you want to know the fundamental concept behind the very impressive Sora Text to Video Generator by OpenAI?

At its heart lies this concept of the General World Models, or GWMs. This very concept also nudges us closer to the holy grail of AI: Artificial General Intelligence (AGI).

In my latest article, I delve deep into the essence of General World Models, exploring how they’re set to transform the AI landscape and what this means for the future of digital innovation. Check out and let me your thoughts.

In today’s fast-paced business landscape, artificial intelligence has shifted from a buzzword to a crucial part of organizational strategy. Recognizing the imperative to stay ahead of this curve, Capgemini launched its Generative AI Lab.

Capgemini’s Generative AI Lab is an initiative designed to navigate the evolving topography of AI technologies. This lab is not just another cog in the machine; it’s a compass for the enterprise, setting the coordinates for AI implementation and management.

The aim of the lab is straightforward yet ambitious: to understand continual developments in AI. It runs under the umbrella of Capgemini’s “AI Futures” domain, tasked with finding new advances in AI early on and scrutinizing their potential implications, benefits, and risks. The lab operates as a dual-pronged mechanism: partly analytical, providing a clear-eyed assessment of the AI landscape, and partly operational, testing and integrating new technologies.

A FORWARD-LOOKING INITIATIVE

The Generative AI Lab also homes in on multifaceted research areas that promise to elevate the field of AI. On the technical front, it ventures into realms like multi-agent systems, where an amalgamation of specialized large language models (LLMs) each excel in particular tasks, such as text generation or sentiment analysis. These are supplemented by model-driven agents rooted in mathematical, physical, or logical reasoning to create a robust, versatile ecosystem.

Also, the lab is innovating ways to make LLMs more sensitive to real-world context, helping to curate AI behavior that’s not just intelligent but also intuitively aware. Then, the lab devotes considerable attention to socio-individual and psychological dimensions, exploring how generative AI interacts with and affects human behavior, social norms, and mental well-being.

Finally, it takes a truly global perspective by investigating the geo-political implications of AI, like how advancements could shift power balances or impact international relations. By casting such a wide net of inquiry, the lab ensures a holistic approach to AI research, one that promises to yield technology that’s technically advanced, socially responsible, and geopolitically aware.

THE LAB’S FRAMEWORK FOR CONFIDENCE IN AI

A noteworthy achievement of the lab thus far has been the development of a comprehensive framework focused on instilling confidence in AI. According to the lab’s research, contemporary technology has only managed to implement four factors necessary for total confidence in AI applications, particularly in decision-making scenarios. The lab, in an initiative led by Mark Roberts, CTO, Hybrid Intelligence, has identified an additional eight factors that need to be considered, offering a 12-point blueprint that could serve as the industry standard for the ethical and effective implementation of AI.

Traditional considerations for AI usually highlight factors like “robustness,” ensuring the system performs reliably under a variety of conditions; “reliability,” so the performance is consistently up to par; and “stability,” meaning the system behaves predictably over time. These factors aim to ensure that an AI system works well, efficiently carrying out tasks and solving problems. However, the lab contends that this isn’t enough.

The lab brings into the limelight other equally important but often overlooked aspects, such as “humility,” the ability for the AI to recognize its limitations and not overstep its capabilities. Then there is the idea of “graceful degrading,” which refers to how the system handles errors or unexpected inputs; does it crash, start to hallucinate, or manage to keep some functionality? And, of course, “explainability,” which revolves around the system’s ability to articulate its actions and decisions in a way that’s understandable to humans and truthful to its contents.

For instance, a recommendation algorithm might be robust, reliable, and stable, meeting all the traditional criteria. But what if it starts recommending inappropriate or harmful content? Here, the lab’s factors like humility and explainability would come into play, supplying checks to ensure the algorithm understands its limits and can explain its rationale for the recommendations it makes.

The goal is to develop an AI system that’s not just effective, but also aligned with human needs, ethical considerations, and real-world unpredictability. The lab investigates and identifies solutions for integrating these more nuanced factors, aiming to produce AI technology that is both high-performing and socially responsible.

OPERATIONAL RESULTS

Since its start, the lab has made significant strides. It has found crucial issues surrounding generative AI, spearheaded the integration of “judgment” in AI systems across the organization, and discovered promising technologies and startups. It has also pioneered methods for using AI for heightened efficiency, superior results, and the facilitation of new organizational tasks. The lab has not only served as a think-tank but also as an operational wing that can apply its findings in a practical setting.

The lab’s workforce is its cornerstone. Composed of a multinational, global, cross-group, and cross-sector core team, it has the expertise to cover a wide array of disciplines. This team can dynamically expand based on the requirements of specific projects, ensuring that all needs – however specialized – are met with finesse.

REPORTING AND EARLY WARNING

Transparency and prompt communication are embedded in the lab’s operational DNA. It has an obligation for early warnings: alerting the organization about the potential hazards or advantages of new AI technologies. Regular reports support a steady stream of information flow, enabling informed decision-making at both the managerial and executive levels.

Another vital role the lab plays is that of an educator. By breaking down complex AI systems into digestible insights, it provides the leadership teams of both Capgemini and its clients with the tools they need to understand and direct the company’s AI strategy. It’s not just about staying updated; it’s about building a wide perspective that aligns with the broader strategic goals.

“BY ALIGNING A HIGH-PERFORMING AI SYSTEM WITH ONE THAT UNDERSTANDS AND RESPECTS ITS HUMAN USERS, THE LAB IS ON THE FRONTIER OF ONE OF THE MOST GROUNDBREAKING TECHNOLOGICAL SHIFTS OF OUR TIME.”

FUTURE OF AI

The Generative AI Lab at Capgemini has managed to go beyond the conventional boundaries of what an in-house tech lab usually achieves. It has successfully married theory with practice, innovation with implementation, and foresight with action. As the realm of AI continues to evolve unpredictably, the lab’s multi-dimensional approach – grounded in rigorous analysis, practical testing, and educational outreach – stands as a beacon for navigating the complex yet promising future of AI.

With this in-depth framework, the Generative AI Lab isn’t just looking at what makes AI work; it’s exploring what makes AI work well in a human-centered, ethical context. This initiative is a key step forward, one that promises to help shape the AI industry in ways that prioritize both technical excellence and ethical integrity.

By aligning a high-performing AI system with one that understands and respects its human users, the lab is on the frontier of one of the most groundbreaking technological shifts of our time. And as AI continues to weave itself into the fabric of our daily lives, initiatives like this are not just useful but essential, guiding us towards a future where technology serves humanity, and not the other way around. With its robust framework, multidisciplinary team, and actionable insights, the lab is indeed setting the stage for a more reliable, efficient, and ethically responsible AI-driven future.

INNOVATION TAKEAWAYS

NAVIGATIONAL COMPASS FOR AI

Capgemini’s Generative AI Lab serves as an enterprise compass, charting the course for AI application and management. This initiative is at the forefront of exploring multi-agent systems and enhancing LLMs with real-world context sensitivity. It’s about pioneering an AI landscape that’s not only intelligent but intuitively tuned to human nuance, reshaping how AI integrates into the fabric of society.

ETHICS AND UNDERSTANDING IN AI

The Lab’s 12-point Confidence in AI framework aims to set new industry standards for ethical AI implementation. By considering factors beyond the traditional – like AI humility, graceful degrading, and explainability – the lab aspires to create AI systems that are not just technically efficient but also socially responsible and aligned with human ethics, ensuring technology that can perform robustly and understand its boundaries.

FROM THINK-TANK TO ACTION

The lab transcends the typical role of an R&D unit by not only dissecting and developing AI advancements but also operationalizing these insights. Its multinational, multidisciplinary team has been crucial in integrating AI judgment across the organization and fostering technologies that enhance operational efficiency, elevating the lab’s role to that of both an innovator and an implementer.

Interesting read?Capgemini’s Innovation publication, Data-powered Innovation Review | Wave 7 features 16 such fascinating articles, crafted by leading experts from Capgemini, and partners like Aible, the Green Software Foundation, and Fivetran. Discover groundbreaking advancements in data-powered innovation, explore the broader applications of AI beyond language models, and learn how data and AI can contribute to creating a more sustainable planet and society.  Find all previous Waves here.

For organizations across the world, 5G has the potential to drive digital transformation and unlock new business opportunities, whether that’s connecting factories across vast areas or optimizing new energy sites like wind farms. However, as organizations embrace 5G, they must also address the escalating cybersecurity threats that come with this technological evolution.

Traditional security models, which rely on perimeter-based defenses and implicit trust built within the industrial network, are ill-equipped to handle the dynamic and distributed nature of 5G networks. This is where zero trust security principles come into play.

Defining zero trust

Zero trust assumes that an attacker may already be present within the network, and a constant cycle of validation needs to be in motion to prevent further infiltration and lateral movements. It offers a proactive and adaptive approach to security, emphasizing continuous verification and strict access controls to mitigate risks and ensure the integrity of 5G networks.

For Industrial enterprises deploying private 5G networks, a zero-trust approach means that all access to 5G networks should be explicitly authenticated, authorized, and monitored, and access privileges should be continuously reviewed. No access should be granted implicitly or by default.

Zero trust for Private 5G Networks

Our strategy for implementing zero trust in private 5G networks aligns with the vendor-agnostic Cybersecurity and Infrastructure Security Agency (CISA) Zero Trust Framework, addressing security across five pillars of zero trust. By leveraging the right technology and methodology, Capgemini recommends an approach based on zero trust principles for secure deployment of 5G technology in OT networks.

The implementation of robust protection and seamless operations needs to cover all five key pillars:

• Identity: Define and enforce granular access control policies for industrial users, allowing specific users to perform specific tasks on a specific asset. These policies should consider contextual requirements such as the time and location of a user’s request.
• Devices: Validate the end devices and ensure that the trust level of the devices is assessed. Policies need to be enforced to allow the segmentation of devices based on 5G network-specific identities e.g. Subscription Permanent Identifier (SUPI)/Subscription Concealed Identifier (SUCI). Policies should support device identity matching and context-based segmentation that allows the grouping of devices based on device type, cellular identities, location, along with the quality of service (QoS), latency, bandwidth, redirection, etc.
• Networks: Zero trust requires a clear separation of communication flows for network control and application/service tasks. For organizations using private 5G networks, a secure communication channel should be established when communicating with different locations, in addition to enabling secure remote access so that only those with the correct authentication and authorization credentials are allowed on the network.
• Applications and Workloads: The OT environment hosts various applications that are used for different purposes, such as data collection, process monitoring, and product creation. Access to these applications over a 5G network should be governed by access control policies that enforce application control, thereby minimizing the attack surface at ingress and egress. Policies should also be able to detect workload vulnerabilities and misconfigurations and enable application control based on operator-specific/standard slices.
• Data: Data protection should include measures to protect sensitive industrial information and prevent data loss. Data protection must be supported for both data at rest and data in motion, taking into account data classification and file types. Data flowing from one industrial site to another and to remote users must be inspected for data leakage, and measures must be taken to restrict access to unmanaged devices and unknown users.

Key industrial 5G security use cases aligned with the zero-trust framework include:

• Network segmentation: Improve digital perimeter resilience by enforcing, micro-segmenting, and grouping devices based on device type (CCTV, mobile), vendor, location, QoS, or 5G cellular identities (SUPI/SUCI).
• Secure remote access: Enforce zero trust-based access controls and offer secure remote access to industrial environments deploying 5G assets for internal and third-party users.
• Policy enforcement for slices: Apply security policies per network slice or group of slices assigned for various applications, based on their slice ID and thus prevent unauthorized data transfer and block various malicious activities inside industrial environment.
• Security monitoring using 5G SOC: Security monitoring of various 5G powered industrial devices (sensors, robots cameras drones end user phones, laptops) in a 5G security operation center (SOC) that offers centralized visibility along with other features such as incident management, vulnerability, and compliance management.

Conclusion

Implementing Zero trust for private 5G networks involves several key essential steps that include defining all the key assets to be protected such as applications, devices, data, etc, documenting the traffic flow over the 5G networks and defining fine grained policies that determine access to resources along with logging and monitoring, that provide key insights into network activity. Effectively implementing zero trust across all levels can greatly enhance the security posture of OT networks leveraging 5G technology.

By embracing zero trust principles and integrating them into the fabric of 5G networks, organizations can mitigate risks, protect sensitive data, and ensure the integrity of their networks in an increasingly interconnected and dynamic digital world.

You can learn more about our approach and our partners by joining us at Mobile World Congress in Barcelona between 26–29 February 2024.