This article is featured on FinanceDerivative and is written by Terence Tse, Bill McCahey and Danny Goh.

The Operational Challenge

Direct lending institutions face a critical bottleneck that constrains growth and profitability: financial spreading. This process—standardizing borrower financials for credit analysis—consumes disproportionate resources while creating delays in loan processing. For mid-sized lenders processing over 2,000 financial documents annually, traditional approaches require teams of analysts to manually review and enter data from diverse document formats, resulting in extended turnaround times, elevated costs, and inconsistent data quality.

Previous technological attempts to address this challenge through optical character recognition or robotic process automation have delivered disappointing returns on investment. These tools can read text but fail to understand context, often requiring as much human intervention as they were designed to eliminate—a frustrating outcome for organizations seeking efficiency.

The AI Solution: Capabilities and Business Impact

Today’s AI platforms represent a fundamental shift in approach, combining computer vision, natural language processing, and machine learning to transform financial spreading from an operational burden into a strategic advantage. The business impact is substantial and measurable across multiple dimensions.

From an operational perspective, leading lenders report processing time reductions from 3.2 hours to just 28 minutes per loan file—an 85% improvement. Error rates have declined by more than 70%, while document handling has become format-agnostic, allowing borrowers to submit information in whatever form they have available. When documentation is incomplete, AI systems immediately identify gaps, eliminating the costly back-and-forth that traditionally delays approvals.

For risk management, the advantages extend beyond efficiency. One direct lender discovered that AI-powered review identified significant financial discrepancies in 43 out of 500 loan files that human analysts had previously cleared—representing potential problem loans with real economic implications. Modern systems automatically calculate and benchmark financial ratios against industry standards while flagging concerning trends in borrower statements, ensuring consistent application of credit standards across all applications.

The revenue impact is equally compelling. A construction-focused lender increased loan volume by 62% without proportional staff growth after implementing AI-powered processing. This scalability allows institutions to pursue market opportunities without the traditional constraint of processing capacity, while faster turnaround times improve both the borrower experience and competitive positioning. Perhaps most importantly, by automating routine data extraction, lenders can redirect analyst talent toward relationship development and complex deal structuring—activities that directly drive new business acquisition.

Implementation Framework for Maximum Return

Business leaders considering AI implementation should focus on five key factors to maximize return on investment. First, integration with existing infrastructure is essential—systems must create seamless data flow between AI platforms and current loan origination systems to eliminate duplicate work and realize full efficiency gains.

Second, customization for specific business segments ensures the AI understands industry-specific financial presentations relevant to your lending portfolio. A tech startup’s financials differ substantially from those of a construction firm, and the system must accommodate these variations to deliver accurate results.

Third, robust security and compliance architecture with bank-grade encryption and comprehensive audit trails maintains regulatory compliance while protecting sensitive borrower information—a non-negotiable requirement in the lending industry.

Fourth, a thoughtful change management strategy develops clear analyst workflows that leverage AI for routine tasks while emphasizing human judgment for complex decisions. The most successful implementations position AI as an analyst enhancement rather than replacement.

Finally, establishing clear performance metrics for processing time, error rates, and analyst productivity creates accountability and quantifies the return on technology investment.

Strategic Advantages Beyond Cost Reduction

Forward-thinking lenders are leveraging AI not merely for efficiency but as a competitive differentiator. With streamlined processing capabilities, institutions can profitably serve smaller loan segments previously considered unprofitable due to operational costs. Advanced AI models identify borrower performance patterns, enabling proactive portfolio management and targeted cross-selling opportunities.

Enhanced borrower experiences represent another strategic advantage. Modern systems offer multi-channel assistance through text, voice, or video chat, providing tailored industry-specific guidance and interactive dashboards that display application status, document quality scores, and improvement recommendations.

Looking ahead, emerging AI technologies will further transform lending through predictive cash flow modeling incorporating real-time market trends, conversational interfaces for borrower communication, alternative data integration, and automated fraud detection. These capabilities will enable lenders to make faster, more personalized decisions at scale.

The financial spreading process hasn’t disappeared, but its transformation through AI creates a foundation for precision lending that delivers superior results for both institutions and borrowers. As competition intensifies, those who embrace this technology will position themselves to lead the market through operational excellence and enhanced decision-making.

Terence Tse is Professor Finance, Hult International Business School and cofounder and Chair in Fintech & Business AI at The Chart ThinkTank

Bill McCahey is Capital Markets Technology Leader, Client Partner at Infosys

Danny Goh is CEO of Nexus FrontierTech and is cofounder and Chair in Frontier Technology & Innovation Ecosystems at The Chart ThinkTank

This article is featured on Tech Policy Press and is written by Mark Esposito, David De Cremer

Forget crystal balls and hazy predictions – we’re on the precipice of an era where the future isn’t merely predicted; it’s actively engineered. Advances in generative artificial intelligence (AI) are fusing with strategic foresight methods to fundamentally change how people and organizations plan for what lies ahead. In traditional scenario planning, experts typically envision a handful of possible futures. Now, AI systems can rapidly generate and simulate countless scenarios, giving decision-makers a tapestry of possible futures to explore. Generative AI can serve as a co-creator in foresight exercises, responding in real time to stakeholder input and spinning out a volume of nuanced scenarios that would have been difficult to craft by human imagination alone. The result is a potentially transformative surge of human creativity enabled by machine computation that stands to redefine decision-making across industries and governments.

Yet alongside this promise come new governance challenges. As AI-driven “agentic” systems take on more decision-making, policymakers must confront questions about oversight, transparency, accountability, and inclusion sooner rather than later. Who sets the parameters for an AI that can shape critical decisions? How do we ensure these AI systems are transparent about their reasoning and accountable for their recommendations? And how do we include diverse voices so that the futures being engineered reflect broad societal values and cultural sensitivities? These issues are no longer abstract – they are pressing concerns as we integrate powerful AI agents into decision-making processes that affect entire communities.

Generative AI meets strategic foresight

At the heart of this shift is the blending of generative AI with strategic foresight practices. In the past, planning for the future involved static models and expert intuition. Now, AI models (including advanced neural networks) can churn through reams of historical data and real-time information to project trends and outcomes with uncanny accuracy. Crucially, these AI-powered projections don’t operate in a vacuum – they’re designed to work with human experts. By integrating AI’s pattern recognition and speed with human intuition and domain expertise, organizations create a powerful feedback loop. AI proposes scenarios and forecasts; humans review these outputs and provide feedback or new inputs; the AI refines the scenarios further. This iterative cycle enables a form of augmented foresight far more dynamic than anything before. Researchers have even found that such human–AI collaborative frameworks can significantly boost decision-making efficiency – one study reported a 10–20% improvement in efficiency and user satisfaction when real-time human feedback was incorporated into AI-driven decision processes.

Rather than treating AI as an oracle, leading organizations treat it as a strategic partner. A generative AI system can sift through millions of data points to suggest, for example, how a geopolitical event might ripple through supply chains or how consumer preferences might shift in a pandemic. Human decision-makers then apply judgment to these AI-generated insights, weeding out scenarios that are implausible or undesirable and probing the ethical implications of those that remain. If carefully designed to account for the known biases and common failure points in generative systems, this blend of machine-driven analysis with human values and critical thinking can yield a more robust decision-making process – one that is fast and data-driven yet remains anchored by human oversight. It can be a potent weapon against uncertainty, allowing leaders to navigate complexity with greater confidence. Indeed, companies that bolster their organizational learning with AI tools are significantly better equipped to handle uncertainty from technological and market disruptions than those that rely on intuition alone. In policy terms, this suggests that governments and institutions embracing AI-enhanced foresight may be better prepared for shocks and surprises, from financial crises to public health emergencies.

Rewriting the rules across industries

The fusion of generative AI and foresight isn’t confined to tech companies or futurists’ labs – it’s already reshaping industries. For instance, in finance, banks and investment firms are deploying AI to synthesize market signals and predict economic trends with greater accuracy than traditional econometric models. These AI systems can simulate how different strategies might play out under various future market conditions, allowing policymakers in central banks or finance ministries to test interventions before committing to them. The result is a more data-driven, preemptive strategy – allowing decision-makers to adjust course before a forecasted risk materializes. Early adopters in the financial sector have found that AI-enhanced forecasting helps them anticipate everything from interest rate fluctuations to credit risks, informing more resilient policy measures.

Similar transformations are occurring in healthcare (with AI predicting disease outbreaks and optimizing hospital responses), in urban planning (with AI simulating infrastructure projects and their long-term impacts), in tourism (for forecasting demand personalization of itineraries), and beyond. In each case, the rules of the game are being rewritten. Decisions that once relied on hindsight and educated guesses are now increasingly informed by forward-looking simulations and analytics. AI can crunch complexity – whether it’s climate data, economic indicators, or social media trends – revealing patterns that humans might miss. Armed with these insights, leaders in both the public and private sectors can craft policies and strategies that are proactive rather than reactive. Crucially, AI-driven foresight doesn’t eliminate the role of human judgment; it amplifies it with better evidence. A data-driven approach to strategy means nothing is left to superstition or wishful thinking – assumptions can be tested in simulations, and the consequences of decisions can be visualized before they happen. For policymakers, that means the potential to “wind-tunnel” test policies (from housing programs to emergency responses) in immersive simulations, refining them for maximum benefit and minimum risk.

Decision theaters: where collaboration happens

These advances are not happening in isolation on engineers’ laptops; they are increasingly playing out in “decision theaters” – specialized environments (physical or virtual) designed for interactive, collaborative problem-solving. A decision theater is typically a space equipped with high-resolution displays, simulation engines, and data visualization tools where stakeholders can convene to explore complex scenarios. Originally pioneered at institutions like Arizona State University, the concept of a decision theater has gained traction as a way to bring together diverse expertise – economists, scientists, community leaders, government officials, and now AI systems – under one roof. By visualizing possible futures (say, the spread of a wildfire or the regional impact of an economic policy) in an engaging, shared format, these theaters make foresight a participatory exercise rather than an academic one.

In the age of generative AI, decision theaters are evolving into hubs for human-AI collaboration. Picture a scenario where city officials are debating a climate adaptation policy. Inside a decision theater, an AI model might project several climate futures for the city (varying rainfall, extreme heat incidents, flood patterns) on large screens. Stakeholders can literally see the potential impacts on maps and graphs. They can then ask the AI to adjust assumptions – “What if we add more green infrastructure in this district?” – and within seconds, watch a new projection unfold. This real-time interaction allows for an iterative dialogue between human ideas and AI-generated outcomes. Participants can inject local knowledge or voice community values, and the AI will incorporate that input to revise the scenario. The true power of generative AI in a decision theater lies in this collaboration.

Such interactive environments enhance learning and consensus-building. When stakeholders jointly witness how certain choices lead to undesirable futures (for instance, a policy leading to water shortages in a simulation), it can galvanize agreement on preventative action. Moreover, the theater setup encourages asking “What if?” in a safe sandbox, including ethically fraught questions. Because the visualizations make outcomes concrete, they naturally prompt ethical deliberation: If one scenario shows economic growth but high social inequality, is that future acceptable? If not, how can we tweak inputs to produce a more equitable outcome? In this way, decision theaters embed ethical and social considerations into high-tech planning, ensuring that the focus isn’t just on what is likely or profitable but on what is desirable for communities. This participatory approach helps balance technological possibilities with human values and cultural sensitivities. It’s one thing for an AI to suggest an optimal solution on paper; it’s another to have community representatives in the room, engaging with that suggestion and shaping it to fit local norms and needs.

Equally important, decision theaters democratize foresight. They open up complex decision-making processes to diverse stakeholders, not just technical experts. City planners, elected officials, citizens’ groups, and subject matter specialists can all contribute in real time, aided by AI. This inclusive model guards against the risk of AI becoming an opaque oracle controlled by a few. Instead, the AI’s insights are put on display for all to scrutinize and question. By doing so, the process builds trust in the tools and the decisions that come out of them. When people see that an AI’s recommendation emerged from transparent, interactive exploration – rather than a mysterious black box – they may be more likely to trust and accept the outcome. As one policy observer noted, it’s essential to bring ideas from across sectors and disciplines into these AI-assisted discussions so that solutions “work for people, not just companies.” If designed well, decision theaters operationalize that principle.

Governance, transparency, and inclusion – a policy balancing act

As AI takes on a more agentic role in shaping decisions, governance, and ethics can no longer be an afterthought. The power that makes AI-driven decision theaters attractive – the ability to rapidly chart courses of action and foresee outcomes – could also lead us astray if not guided by strong principles and oversight. Policymakers should consider establishing clear governance frameworks for how AI is used in strategic decision-making contexts. This includes setting standards for transparency (AI systems should be able to explain why they suggest certain futures or decisions) and accountability (human officials must ultimately be responsible for choices made with AI input). Encouragingly, we see movement on this front. The European Union’s new AI Act, for instance, explicitly emphasizes the importance of trust, transparency, and accountability in the deployment of advanced AI. By adopting a risk-based approach, the AI Act aims to ensure that, as AI systems become more autonomous and influential, they are still aligned with fundamental rights and subject to human oversight. This kind of regulatory ethos will be crucial for decision theaters: the AI tools guiding group decisions must be trustworthy, and their operations must be visible to participants and regulators alike.

One practical governance step is requiring algorithmic transparency in public-sector AI tools. If a city uses an AI-driven model in its urban planning decision theater, its assumptions, data sources, and known limitations should be audited. Likewise, outputs should be recorded – which scenarios were generated and on what basis – so that there is an audit trail linking decisions back to evidence. This would help address questions later if a decision is called into question (“Why did we choose Policy X? What information was it based on?”). An ethical framework for AI-assisted decision-making can guide what kinds of scenarios are explored; for example, intentionally avoiding options that, even in simulation, blatantly violate ethical norms or human rights. Think of it as drawing bright lines in the sandbox: some “unacceptable” AI-suggested actions should be off-limits, just as the EU AI Act bans certain high-risk AI practices outright.

In parallel, policymakers must ensure that the inclusivity of decision theaters isn’t just an aspiration but a reality. If only elites or other homogeneous groups have access to such foresight tools, we risk reinforcing existing biases and blind spots in policy. Therefore, guidelines or even mandates for multistakeholder participation could be established. For national-level foresight exercises, that might mean having representatives from different regions, social groups, and expertise areas “in the room” (physically or virtually) when AI-driven scenarios are being discussed. A recent multistakeholder forum noted that including voices from civil society and diverse communities in AI policy dialogues is essential to ensure outcomes serve the public interest. The same holds true for AI-guided planning: inclusion is a safeguard against error and inequity. Diverse participants can spot cultural blind spots or value conflicts in AI models that developers might have missed. They can also raise concerns about how different social groups might be affected by a given scenario, prompting the exploration of more inclusive alternatives.

Finally, existing governance models in related domains can provide a template. For example, frameworks developed to oversee AI in high-stakes fields like healthcare or autonomous driving could be adapted to the context of strategic planning. One study suggests that aligning AI deployments with Environmental, Social, and Governance (ESG) principles can help businesses navigate the ethical and societal challenges of AI. A similar approach could inform the governance of decision theaters, ensuring that the AI’s use aligns with societal values (social responsibility, fairness, sustainability) and that an overriding ethical framework guides how scenarios are generated and evaluated. In practice, this might involve an oversight board reviewing major AI-informed policy decisions or scenario sets, evaluating them against criteria like fairness and sustainability. It could also mean updating public sector ethics rules to cover the usage of AI in analysis and decision support. The key point is that policy infrastructure must keep pace with technical infrastructure. Just as we invest in AI capabilities to improve decision-making, we must invest in the rules, norms, and institutions that ensure this new decision-making paradigm remains worthy of the public’s trust.

A future of empowered decisions – if we get it right

The advent of AI-enhanced decision theaters represents a paradigm shift in how societies can plan for the future. This new model holds extraordinary promise: more informed strategies, fewer unintended consequences, and a capacity to navigate uncertainty with clarity that past leaders could only dream of. In a sense, we are coding the future – using algorithms to chart pathways through the fog of the unknown. This can empower organizationsand communities to take proactive stances on everything from climate adaptation to economic development. Rather than being blindsided by events, those using these tools will have rehearsed many tomorrows in advance.

But realizing that promise requires a conscious effort to marry innovation with governance. Policymakers and strategists should see themselves not just as consumers of AI foresight tools but as shapers of the ecosystem in which those tools operate. The rules of the game are still being written. By instituting strong transparency requirements, accountability mechanisms, and inclusive processes now, we can ensure that the “game” yields wins for society at large and not just a tech-savvy few. It bears remembering that AI, for all its computational genius, lacks a moral compass or a sense of public duty – those must be provided by us, the humans in the loop.

In the coming years, expect to see decision theaters pop up in government agencies, international organizations, and corporate strategy departments. They will likely become as indispensable as conference rooms and Zoom calls – the place you go when you need to tackle a tough, complex decision with input from many angles. A new generation of policy professionals and leaders will be as comfortable interrogating an AI-driven simulation as they are reading an Excel chart. Their success, however, will hinge on the guardrails we set up today. With thoughtful policy and an insistence on ethics, we can harness agentic AI to widen decision horizons without ceding values. In doing so, we affirm that while the future may indeed be coded, humans still write the instructions.

This article is featured on The Economic Times and is written by Amit Kapoor and Mohammad Saad, ET CONTRIBUTORS.

AI’s rapid integration into the IT sector has raised concerns about job displacement, particularly in India’s outsourcing hub. While AI tools enhance coding speed, over-reliance could erode essential problem-solving skills, limiting innovation.

Artificial Intelligence (AI) has become the defining buzzword of the 21st century. It’s all around us, integrating into our lives more rapidly than we could have imagined just a few years ago. While AI’s advancement is set to transform the way we work, concerns about job displacement are understandably widespread. However, a closer look at these trends suggests that, in the long run, the rise of AI may produce counter-intuitive effects on the labour market.

AI pioneers envision a future where AI plays a central role across nearly all sectors of the economy. Although we may not be there just yet, the IT sector’s vulnerability to rapid AI advancements is already becoming apparent. AI systems today are becoming increasingly capable of writing and debugging computer code. Anecdotal evidence confirms that many professionals in the IT industry are already leveraging large language models (LLMs) to assist with coding. Although human oversight remains necessary, the growing sophistication of these tools have sparked legitimate concerns. Fears are ripe that with time, simpler, & repetitive coding jobs will be the first to be replaced by AI. Looking ahead, people argue that continued advancements may eventually lead to the automation of even more complex software development roles. In such a scenario, the remaining jobs will only be for AI/ML engineering and maintaining AI systems. While there is considerable support for this viewpoint, a closer examination of how AI works, suggests that coding jobs are not going to completely disappear.

We know that AI tools can generate code in various programming languages mainly because they have been trained on vast amounts of publicly available code from sites like Stack Overflow (a popular forum where developers share ideas, ask questions, and exchange knowledge). However, these tools don’t “understand” code in the way humans do; instead, they generate plausible solutions based on patterns learned from existing examples. A crucial point to recognize is that AI relies on existing code to solve problems. In other words, it struggles with problems that haven’t already been addressed in its training data. If the trend of relying on AI for coding continues, we may eventually reach a point where the internet’s repositories of new and original code are updated more slowly. That is because fewer people may retain the skills or motivation to write and share novel solutions independently. This growing reliance on AI tools could erode essential problem-solving and debugging skills, particularly among newer developers. In his blog, The Pragmatic Engineer, Gergely Orosz highlighted a sharp decline in the number of questions posted on Stack Overflow, suggesting that the site is becoming increasingly irrelevant as more developers turn to AI for coding help. This shift suggests a potential weakening of hands-on coding abilities and raises concerns that online code repositories may not become obsolete but could see slower updates as fewer developers actively share new solutions. This poses a significant limitation for AI: since these tools depend on existing code to generate solutions, they struggle to address novel challenges. If the pool of skilled coders continues to shrink, AI will lack the fresh examples it needs to stay effective.

Ironically, this could lead to a market correction. As basic coding roles are automated out of existence, those who retain strong problem-solving and programming skills will become increasingly valuable. In the future, we may see a smaller number of highly paid developers who are capable of tackling complex, unsolved problem; an outcome few anticipate today.

Now, proponents of AI might argue that while AI may struggle with completely novel problems, it excels at performing permutations and combinations of existing data and, therefore, has the capability to ‘generate novelty’. However, a deeper philosophical reflection shows that AI lacks true understanding. It doesn’t know what it’s doing as it simply maps input patterns to output patterns. So, what appears to be “novelty” in AI outputs is often just the result of recombining known elements. Although the set of plausible new combinations that AI can produce is extremely large and that space may contain many useful ideas, but it won’t include all truly new ones, especially those that require abstraction or insights. Therefore, one cannot help but accept that AI will always likely stay a step behind the collective mental prowess of humankind.

Despite these dynamics, the global impact of AI on the workforce will still be significant, and in India’s case, it is likely to be even more pronounced. India is home to one of the world’s largest populations of developers and has long served as a hub for outsourced IT services. In September 2024, it was reported that GitHub had added 2.2 million new Indian developers to its platform, bringing the total to 15.4 million. Around the same time, an RBI survey noted that India’s software services exports rose to $205.2 billion in 2023–24, up from $200.6 billion the previous year. However, these very services are among the most vulnerable to automation. A large segment of India’s tech workforce is already performing tasks that are increasingly being assisted or replaced by AI. A recent GitHub report found that developers using GitHub Copilot completed coding tasks 55% faster than those who did not. This growing presence of AI could have a double-edged impact on India’s job market. Traditional coding roles may decline, and at the same time, the workforce might struggle to shift toward innovation-driven work due to growing dependence on AI. If AI systems eventually reach limits in solving new or complex challenges, countries with a more deeply skilled developer base may be better positioned to lead in software innovation.

For India to maintain its leadership in IT service outsourcing, it must ensure that future entrants into the tech workforce use AI as a productivity enhancer, without compromising their core coding skills. As AI increasingly takes over routine tasks, human strengths like problem-solving and innovation will become even more valuable. The Indian government’s efforts to prepare the workforce for an AI-driven future must emphasize developing these higher-order cognitive skills and cultivating a mindset ready to tackle novel challenges. At the same time, India should not remain merely a consumer of new technologies but strive to become a global leader in innovation, shaping the next generation of tools and platforms. Ultimately, AI should serve as a tool that enhances human intelligence, not one that replaces it.

Amit Kapoor, is Chair and Mohammad Saad, is Researcher at Institute for Competitiveness .

This article is featured on The Economic Times and is written by Amit Kapoor and Kartik, ET CONTRIBUTORS.

India is working towards becoming an AI leader by launching initiatives for infrastructure, innovation, and skill development. The government has allocated funds and set up high-end computing facilities. They focus on training and retaining talent. Private sector collaboration is needed. Comprehensive AI education is required to prepare future generations.

If geopolitics were like a poker game, AI chips would be the aces up every nation’s sleeve. The only exception today is that you can’t just draw one; you must build the whole deck. While key global players worldwide are busy stacking their decks, India is still figuring out how to print the cards. Responding to this imperative, India launched its AI mission last year. Even though the efforts of the government are steps in the right direction, a more holistic effort is needed across various stakeholders in the coming years to make India an AI powerhouse.

Under the mission, India has launched initiatives addressing various aspects of the Innovation landscape, from infrastructural capacity building to spurring innovation. India has allocated approximately 10,300 rupees under this mission over the coming five years. A major focus of the mission is to build high-end common computing facilities equipped with 18,693 Graphics Processing Units. To ensure access to abundant computing architecture, the Government has introduced an open GPU marketplace, enabling startups, researchers, and students to access high-performing computing architecture. Although the Indian government is making strides in fostering AI investments, the private sector’s contribution is comparatively weak, a conclusion supported by India’s ranking within the Stanford AI Index report. India ranked 8th in AI private investment, attracting $1.39 billion in 2023, compared to $67.22 billion in the US and $7.76 billion in China. Regarding AI concentration, India remains behind Israel and Singapore despite its talent pool growing by 263% since 2016. While India’s AI startup ecosystem is expanding, ranking 8th globally with 45 new AI start-ups in 2023, it still trails established innovation hubs.

Till now, India’s focus has been heavily focused on catching up rather than pioneering innovation. While countries like the US and China invest heavily in fundamental AI research, India’s AI ecosystem is dominated by service-based applications rather than groundbreaking AI model development. This over-reliance on IT services rather than fostering deep-tech innovation limits India’s ability to set global AI trends. The lack of innovation in the field is asserted by the fact that India contributes only 0.23% of the AI patents filed globally, far behind China’s 61.13%. To induce innovation in the space and build globally competitive AI models, the Indian government called for proposals from startups and researchers to build AI models specifically trained on Indian data sets. As part of this initiative, the Indian government received 67 proposals from academia and industry to build such models. Additionally, the government has developed the IndiaAI dataset platform to train the AI models, providing platform builders access to a unified repository of high-quality anonymized datasets to train their models. Building upon the government’s initiatives, a shared responsibility exists for the private sector to contribute equally to AI advancement possibly through increased collaboration with academia through joint research programs and internships to nurture fresh talent within their organization, inducing further innovation in the industry.

Beyond fostering innovation, developing an AI powerhouse demands a highly skilled workforce. Equipping this workforce with the necessary skills at all educational levels is the critical responsibility of the nation’s education system. For this very purpose, India has launched the Future Skills Prime program. The program focuses on reskilling and upskilling workers in the IT industry by offering 119 courses on AI and other technologies such as Robotic Press Automation, Augmented/Virtual Reality, and many more. To provide a platform to students, especially in Tier 2 and Tier 3 cities, the government has established Data and AI labels offering India AI fellowships to students pursuing undergraduate and postgraduate courses. However, India’s AI mission lacks initiatives to integrate AI education into the school system, a critical gap in building a future-ready workforce. Without early exposure, students miss foundational skills essential for advanced AI careers. China has been aggressive in this regard. Most recently, Beijing announced AI courses for primary and secondary school students, ensuring early exposure to AI concepts. From September 2025, schools are set to introduce AI-focused curricula, incorporating AI into after-school programs, research projects, and extracurricular activities. More than 500 universities in China already offer AI courses, and top institutions like Peking University are expanding enrolment in AI-related fields. This structured approach reflects a long-term commitment to AI dominance, ensuring that future generations are equipped with the skills necessary to compete in an AI-driven global economy.

Although equipping students with cutting-edge AI skills is crucial, India grapples with an even more pressing challenge which is on retaining its high-skilled workforce. Between 2015 and 2022, 1.3 million Indians left the country for higher salaries and advanced research infrastructure. Brain Drain significantly impacts AI development in the country as it depletes its pool of skilled professionals. This exodus weakens domestic innovation as India loses the expertise required to drive cutting-edge AI projects. This leads to companies facing talent shortages in the sector, delaying product developments and reducing global competitiveness. The onus of tackling the issue of heavy brain drain lies heavily on the private sector. The private sector must focus on by creating a conducive environment for talent retention and growth, which would curb the ongoing brain drain from the country. In addition to providing competitive salaries and career advancement opportunities, companies should foster a culture of research and experimentation within their organization. To promote R&D, firms should establish in-house innovation labs and ensure hassle-free access to cutting-edge tools for all employees.

As the global poker game intensifies, where AI is the ace, India recognizes that merely printing cards is insufficient. A full house, a winning hand, requires more than just government initiatives; it demands a concerted effort from the private sector to amplify investments and staunch the brain drain. To ensure future generations are dealt a winning hand, AI education must be integrated early on, a strategic play to secure future aces. By strategically combining these moves, and building on this solid foundation, India can transform its nascent deck into a formidable force, ready to compete at the highest table.

Amit Kapoor, is Chair and Kartik, is Researcher at Institute for Competitiveness.

This article is featured on The European Business Review and is written by Aurélie Jean, Mark Esposito, Terence Tse and Danny Goh.

Artificial intelligence and its rapid emergence have proven to be a double-edged sword. While its benefits are numerous, it also has the potential to bring grave consequences. In this essay, four experts from the fields of technology, AI, and digital transformation provide a framework for making the most of AI with low risks.

Today’s rapid emergence of AI for research and development (R&D) has opened up a myriad of use cases across different industries. We stand to gain the benefits of AI in general, and with the right level of project planning, we can deliver high business performance improvement. These are exciting times for AI. These are also times of contradiction: every story covering unprecedented successes is matched by other articles of concern over threats of unintended consequences. We want to develop AI to improve our lives. But we also want to protect ourselves from the bad things.

So, how can we build better tools to live better lives and use our tools to prevent life’s catastrophes? In the modern world, the very best time to do that is in research and development planning. We need ways to guide our innovative thinking and planning. Here, we are offering one such possibility.

The Recursive “More 3P” AI Transformation Framework

The definition of a recursive algorithm continuously seeks to break down the problem that the software is being asked to solve into smaller instances of the same problem. Instead of a solution to the problem, our proposed “More 3P” AI Transformation Framework intends to be a comprehensive disciplined approach to solution development.

The “3P” refers to the three-dimensional realms considered in the R&D guidance: Prediction, Personalization, and Precision.

Prediction:

Predictive data analytics help ensure that organizations can, and in near-real-time, forecast conditions, trends, outcomes and behaviors more accurately, leading to informed proactive planning and decision-making. When considering how to apply AI in any given field, predictive simulations come first to mind. Indeed, by capturing the signals and patterns within the underlying mechanisms and logic of past events or a given phenomenon, algorithms can anticipate (or predict) a similar event in the near or long-term future.

In past decades, ever-better processing and collection of data have enabled real-time gathering and automated curation of datasets that can be used to train AI.

Of note is that one could pay attention to the model’s capability to predict unique, singular, or new events resulting from the simulations, either autonomously or with human subject matter experts (SMEs) oversight in the given field or industry sector.

Such “predictives” can also help identify threat risk potential hidden – for example technology discrimination and environmental impact – within the opacity of the algorithms driving the simulations, again with SME human oversight.

Personalization:

Personalization, on the other hand, indicates a potential for customizing services, products, and experiences that are tailored to individual users (consumers, customers, clients, and/or value-chain partners) based on demographics, preferences, and behavioral data to enhance product or service provision, satisfaction, and loyalty.

In past decades, ever-better processing and collection of data have enabled real-time gathering and automated curation of datasets that can be used to train AI. Additionally, up-leveling advances made on AI models make possible highly personalized recommendations or actual modifications of the delivery of products and services in general. Such increased capacity for personalization brings new business opportunities in improved design, production and delivery of more personalized products and services to customers.

How to Embrace the Recursive 3P AI Transformation Framework Approach

We expect More Predictive, More Personalized and More Precise business, work or production processes by employing formal design and systems thinking approaches that fully anticipate and consider AI transformation opportunities.

A use-case-driven approach should be used when adopting the 3P Frame by asking the following questions: What are the business transformation opportunities envisioned? What are the problems that can be anticipated? What are the threat risks?

We offer the More 3P AI Transformation Matrix, as depicted in Figure 1, which presents a list of all tasks for each of the three Ps. Here are the steps to building and using an AI Transformation planning matrix:

1. List all the tasks in which AI is already being, will be or can be applied;

2. For each task, list how this task can become more precise and/or more predictive, and/or more personalized;

3. Identify tasks that have substantially similar or common ways to more of each P dimension;

4. Identify tasks that have the most influence on precision, prediction and personalization to identify the tasks that will bring the most to the business;

5. Reiterate this matrix over time while moving forward to the AI transformation of the target process;

6. Tell the stories of what went right, what went wrong, and what came next.

During the matrix planning process, it is useful to remember that this is simply an R&D guidance mechanism for model development. The simpler, the better. This enables development teams to actively and iteratively apply The More 3P Matrix to make ever-increasingly smarter, and more effective planning decisions.

The Promise of Business-Transforming Performance and Return on Investment

A simple framework and matrix means it can be more rigorously applied. In its simplicity, thoughtful detail, and rigor lies its power as an important planning tool. “More” performance improvement within the context of the 3P Framework Matrix may also mean improvement in planning for abeyance and mitigation of downside risk instances such as technology discrimination, for example, techno-racism, and algorithmic biases.

This is made possible by measuring the “more” quotient levels in each of the 3P realms and their evolutionary stages over time. Indeed, being more precise, more personalized and more predictive over baseline measures enforces project concepts as a matter of science and engineering.

The More 3P AI Transformation Framework is a proactive approach to continuous process improvement that voluntarily compels the full and willing participation of designers, developers (both data scientists and process engineers), and business-side leaders.

Technology development teams and business leaders must collaborate. They need to collaborate at every stage to design, develop, test, and operationalize. The underlying machine learning operations and/or generative artificial intelligence operations, as applied to the algorithmic models being developed, require business leadership to help ensure that the model’s aims, in terms of “More 3P AI,” meet all design criteria represented in the matrix.

If not, the systems may not run correctly with a resultant drag on business performance and return on investment for the company as a consequence. Poor governance may well result in the unintended consequences of imprecise (loose) AI model development. The More 3P AI Matrix, at its best application, prevents that.

Forward-Leaning

The More 3P AI Transformation Framework is a proactive approach to continuous process improvement that voluntarily compels the full and willing participation of designers, developers (both data scientists and process engineers), and business-side leaders.

This matrix must be planned and implemented with no-option fidelity as a “mission critical” matter to ensure these software systems meet stringent performance guidelines, specifications, and necessary policy and governance guardrails, all geared toward achieving true AI-led business transformation.

The More 3P AI Framework Matrix should likely lead the development of a business case for the proposed AI project models. This may apply particularly in mature corporate software development organizations. What makes it to the matrix and is fully listed and described will inform the business case criteria. For the AI start-up or early-stage venture, the 3P Framework Matrix may represent a clear, easy-to-understand way to introduce to all stakeholders what the business of the business is.   

When pursuing AI Transformation within an organization, it’s useful to envision it through this Framework lens as it demonstrates how the organization can empower itself to take the best advantage of AI. The faster this can be fully articulated and demonstrated to leadership, the faster competitive advantage can be attained.

Furthermore, delineating how the business can evolve to become a More 3P organization within the business case cum an applied 3P AI Transformation Framework during the AI transformation provides a straightforward narrative of AI’s impact. This storyline goes beyond just the expected growth, profitability and efficiency performance gains. This narrative powerfully conveys AI’s business value, resonating with all stakeholders—employees, customers, value chain partners, and investors—by showcasing tangible business operations and enhancements to outcomes.

Embedding Effective Governance

Demonstrating near bullet-proof effective algorithmic governance in the 3P AI Transformation Framework is crucial as it addresses various challenges, such as data privacy, sovereignty, ethical considerations and guardrails, and regulatory compliance. By integrating algorithmic governance, the organization can mitigate risks associated with AI while optimizing its benefits. This governance should be dynamic, evolving with changing AI capabilities and societal norms to ensure that the transformation is innovative, continuously responsible, and sustainable.

Conclusion

A well-considered AI transformation course, viewed through the “3P” lens can confer substantial advantages for organizations. By rigorously applying 3P AI Transformation Framework strategies, organizations can realize significantly enhanced operational effectiveness and efficiencies, business performance improvement, ROI and competitive advantages – all governed by robust recursive algorithmic development and management practices that safeguard against potential risks.

Aurélie Jean, PhD, computational scientist, entrepreneur and author. Aurélie Jean has close to 20 years of experience in computational science applied to a broad range of disciplines. After 11 years of academic research, Aurélie is now running two companies, including a deep tech AI startup in the early detection of breast cancer. She is the author of several bestseller non-fiction titles on algorithmic science, as well as a columnist on science and technology. Aurélie is teaching algorithmic science in executive education and is a research fellow at the Hult Business School and The Digital Economist. She is also an investor and a board member of several companies in the United States and in France.

Dr. Mark Esposito is a professor of economics and public policy with appointments at Hult Int’l Business School and Harvard University. He equally serves as an Adjunct Professor of Public Policy at Georgetown University’s McDonough School of Business. 

At Harvard, he serves as a social scientist with affiliations at Harvard Kennedy School’s Center for International Development; Harvard University’s Institute for Quantitative Social Science (IQSS); the Davis Center for Eurasian Studies and he is an incoming faculty affiliate of the Berkman Klein Center for Internet and Society at Harvard.

He co-founded the Machine Learning research firm, Nexus FrontierTech and the EdTech venture, The Circular Economy Alliance. He has equally co-founded The Chart ThinkTank and The AI Native Foundation. He was ranked by Thinkers50 in 2016 as one of the 30 rising business thinkers in the world and got shortlisted for the Breakthrough Award in 2019 and for the Strategy Award in 2023. He holds a doctoral degree from Ecole des Ponts Paris Tech and lives and works across Boston, Geneva, and Dubai.

Terence Tse is a globally recognized educator, author, and speaker. He is a Professor of Finance at Hult International Business School and co-founder of Nexus FrontierTech, an AI company. He is also a visiting professor at ESCP Business School and Cotrugli Business School. His latest co-authored book, The Great Remobilization: Strategies and Designs for a Smarter Global Future, was nominated for the 2023 Thinkers50 Strategy Award. Terence co-authored two Amazon bestsellers, The AI Republic and Understanding How the Future Unfolds. The DRIVE framework from the latter earned a nomination for the Thinkers50’s CK Prahalad Breakthrough Idea Award. Terence also authored Corporate Finance: The Basics, now in its second edition. He has appeared on numerous media platforms as well as ran workshops and consulted for global brands. He holds a doctoral degree from the Cambridge Judge Business School and has a background in investment banking and consulting.

Danny Goh is a serial entrepreneur and educator with a significant impact in the fields of innovation, sustainability, and digital transformation. He currently holds the position of Entrepreneurship Expert at the Saïd Business School, University of Oxford, and serves as a Senior Fellow at the Centre for Policy and Competitiveness at the École des Ponts Business School. Additionally, Danny lectures on entrepreneurship and digital transformation at various universities, including ESCP and Ecole des Ponts Business School, and provides his expertise as an advisor and judge for various technology start-ups and accelerators, including Microsoft Accelerator, Startupbootcamp IoT, and LBS Launchpad.

Danny is also a sought-after speaker, having presented at prominent conferences such as TEDx, UK High Commission events and BCGx. He co-authored the international best-seller on Amazon, The AI Republic: Building the Nexus Between Humans and Intelligent Automation, alongside Dr. Mark Esposito and Dr. Terence Tse. He is also a co-founder of the AI Native Foundation and The Chart Thinktank, a bipartisan non-profit organisation based in London. These organisations collaborate with global entities, including the United Nations and various governments, to advance the use of technology in the fourth industrial revolution.

As the founder and CEO of Nexus FrontierTech, Danny leads an AI research lab that leverages computer vision and NLP technologies to create digital transformation solutions for businesses across all sectors. The lab has successfully assisted Fortune’s top 100 enterprises like HSBC, UBS, MUFG and Toyota, as well as government bodies in the UK and Singapore, in integrating AI into their operations, thereby driving business growth and maintaining a competitive edge in an ever-evolving digital landscape.

This article is featured on California Management Review and is written by Melodena Stephens, Mark Esposito, Raed Awamleh, Terence Tse, and Danny Goh.

Defining open source AI and a framework for understanding the ambiguities of the term.

“Open source” has been a buzzword in the tech community for decades. The term first became popular with software in the 1970s, when operating systems were incompatible and early programming suffered when older systems were replaced (see Richard M. Stallman’s (2002) work on free software). However, there has always been ambiguity and speculation about what exactly “open source” means, a debate that recently reopened when Elon Musk sued Open AI in March for reneging on its mission to be open source (Gent, 2024). In this article, we offer a definition of open source AI and a framework for understanding the ambiguities of the term and offer recommendations for the safe and efficient use of open source AI. 

Understanding open source AI

Open source AI refers to a combination of what is free in terms of resources of the AI model (API, code, data, hardware, IP), processes (development, testing, feedback, patching), or effects (knowledge, education, products). In general, open-source AI involves the algorithms, code, and data used for training an AI model being made publicly available. The goal in doing so is often to foster collaboration and allow for users, developers, and researchers to build upon and improve the AI model in question (Shrestha et al., 2023). Open source AI is thus a deliberate strategy regarding the access and usage terms of the AI at hand (Fukawa et al., 2021).

Our framework posits that open source AI is not limited to software but can include combinations of software, hardware, data, or knowledge (see Figure 1). It is crucial to note that some projects claim to be “open source” when they only release the neural network model’s weights (its pre-trained parameters) while not providing other elements, such as the original dataset or training code (Ramlochan, 2023). Non-profits have worked on developing projects that fully open up the AI model training process, such as the Allen Institute for AI’s (2024) Open Language Model OLMo. 

Even with truly open source software, misconceptions arise. The term does not mean the software is freely available, i.e., the source code (the parts of software copied on computer/device), the code behind this, or the kernel (the part of the software that ties the entire system together) may not be free. In many cases, this misunderstanding is creating confusion about the actual transparency of AI companies and their altruism (Liesenfeld & Dingemanse, 2024). For example, software that is not free in terms of money but free in terms of usage could be considered open source. For example, Red Hat Software Inc., a publicly traded company, sells subscriptions for Linux-based products, considering that Linux is an open source free software. 

Figure 1: Open source AI in terms of hardware, software, data, and knowledge

Open source AI can allow you to study, use, access, copy/make, modify, distribute, and collaborate, sometimes with strings attached – like a licensing fee, copyright distribution terms, fees for tech support, or other hidden fees (like data storage and access fees once you move to larger data volumes). 

Figure 2: The Open Source Spectrum

Using open source AI safely and efficiently

When you choose open source AI projects—for transparency, research collaboration or commercialization—read the fine print. Firstly, investigate who the owner of the project is. The owners of many open source AI projects have a poor track record of keeping promises. When the upkeep of the project becomes expensive, many owners also restrict new model upgrades using a fee structure. It is also crucial to look at how active the community is around the open source project. Are they a dying breed, or are they growing and keeping an eye on each other? 

For safety reasons, parts of open source AI code are proprietary. You need to understand the implications of this, especially if you are going to use open source AI as part of a professional or business venture. Software support will also be needed down the line for functioning and cybersecurity. Open source AI can, for example, be embedded with malicious code (see Harush, 2023). If you do not have the necessary cybersecurity expertise yourself, be sure you know how to get it and if you will have to pay for it. If you are building on top of an open source project and planning to commercialize, consider the risks if the project gets corrupted or shifts – how will it impact the functioning of your business? Often, open source means you should also open source derivatives of the project unless you state it specifically, so always read the open source license terms.

Open source AI is often presented as a way to democratize AI development and training. However, open source AI has the same ambiguities and poses the same risks as any other tech trend. Taking the time to understand the variety of open source AI projects out there, how they approach the concept of open source in terms of hardware, software, data, and knowledge, and how to use them safely and efficiently is crucial in order to reap the benefits of this tech while effectively mitigating its risks. 

References

Allen Institute for AI. (2024). Open Language Model: OLMo. Retrieved May 13, 2024, from https://allenai.org/olmo

Fukawa, N., Zhang, Y., & Erevelles, S. (2021). Dynamic capability and open-source strategy in the age of digital transformation. Journal of Open Innovation: Technology, Market, and Complexity, 7(3), Article 175. https://doi.org/10.3390/joitmc7030175 

Gent, E. (2024, March 25). The tech industry can’t agree on what open-source AI means. That’s a problem. MIT Technology Review. https://www.technologyreview.com/2024/03/25/1090111/tech-industry-open-source-ai-definition-problem/ 

Harush, J. (2023, November 27). The hidden supply chain risks in open-source AI models. Checkmarx. https://checkmarx.com/blog/the-hidden-supply-chain-risks-in-open-source-ai-models/ 

Liesenfeld, A., & Dingemanse, M. (2024). Rethinking open source generative AI: Open-washing and the EU AI Act. In Seventh annual ACM conference on fairness, accountability, and transparency (ACM FAccT 2024) (pp. 1-14). Association for Computing Machinery. https://pure.mpg.de/rest/items/item_3588217_2/component/file_3588218/content 

Ramlochan, S. (2023, December 12). Openness in language models: Open source vs open weights vs restricted weights. Prompt Engineering & AI Institute. https://promptengineering.org/llm-open-source-vs-open-weights-vs-restricted-weights/ 

Shrestha, Y. R., von Krogh, G., & Feuerriegel, S. (2023). Building open-source AI. Nature Computational Science, 3(11), 908-911. https://doi.org/10.1038/s43588-023-00540-0 

Stallman, R. M. (2002). Free software, free society: Selected essays of Richard M. Stallman. GNU Press.

Melodena Stephens

Melodena works on Metaverse governance, AI ethics and sustainability, agile government, and crisis management at a policy and strategy level. She is a member of several IEEE SA Ethics committees, and a co-author of the upcoming book, ‘AI Enabled Business: A Smart Decision Kit.’

Mark Esposito

Mark Esposito is Professor at Hult Int’l Business School and Harvard University’s Division of Continuing Education and works in public policy at the Mohammed Bin Rashid School of Government. He directs the Hult Futures Impact Lab. He co-founded Nexus FrontierTech and the Circular Economy Alliance. He has written over 150 articles and edited/authored 13 books. His next book, “The Great Remobilization” will be published by MIT University Press in the course of 2023.

Raed Awamleh

Raed Awamleh is the Dean of the Mohammed Bin Rashid School of Government leading executive education, policy research, and Master programs. Prior to 2015, Raed was Middlesex University’s (UK) Pro Vice Chancellor and Dubai Campus Director. He was also the Dean of Academic Affairs at the University of Wollongong (Australia) in Dubai.

Terence Tse

Terence Tse is Professor of Finance at Hult International Business School and a co-founder and executive director of Nexus FrontierTech, an AI company. He is also a co-founder of Excellere, a think tank with the goal to help people explore and release their potential through new technologies. He has worked with more than thirty corporate clients and intergovernmental organisations in advisory and training capacities. He has written over 110 articles and three books including The AI Republic: Building the Nexus Between Humans and Intelligent Automation (2019). His next book, The Great Remobilization, will be published by MIT University Press, in the course of 2023.

Danny Goh

Danny Goh is a serial entrepreneur and an early stage investor. He is the partner and Commercial Director of Nexus Frontier Tech, an AI advisory business with presence in London, Geneva, Boston, and Tokyo to assist CEOs and board members of different organizations in building innovative businesses that take full advantage of artificial intelligence technology.

Abstract

The launch of the generative artificial intelligence (gen AI) application ChatGPT by OpenAI launched artificial intelligence into public discourse and led to a wave of mass uptake of this technology in organizations in the private sector. At the same time, AI is increasingly incorporated into government functions and the public sector. We propose that governments and the public sector can set an example for the responsible use of AI technologies by following the principles of algorithmic governance traditionally recommended to the private sector. Algorithmic governance has traditionally been defined in the literature as governance by algorithms, or how artificial intelligence is used to make governance decisions and affect social ordering. However, we take an alternative approach; instead, we conceptualize algorithmic governance as the governance of algorithms. We begin by summarizing the risks of generative AI use in government, then outline algorithmic governance principles, a step-by-step approach to implementing algorithmic governance into government or public sector projects, opportunities for inter-sector collaboration, and final conclusions.

Read the full text.

This article is featured on LSE and is written by Aurélie Jean, Guillaume Sibout, Mark Esposito, and Terence Tse.

In the era of artificial intelligence, information is collected and processed automatically on a large scale. However, translating that into innovation is a challenge. Aurelie Jean, Guillaume Sibout, Mark Esposito and Terence Tse write that opening datasets and computer source codes may help solve the problem. Sharing information accelerates the pace of co-innovation, facilitating inter- and multi-disciplinary research and study, as well as expanding and propagating scientific knowledge and research results.

The quote Information is power relates our level of influence and power to the amount and quality of information we own. Today, more than ever, and thanks to technology, we share data and information to reach greater influence and power to optimise decisions, detect key insights to a given phenomenon or customise user experience. (Big) data allow us to get insights and counter-intuitive information, or to track in time and space the evolution of information on a phenomenon, for instance.

In the COVID-19 pandemic, open-data platforms, open research and open-source software programs demonstrated the super power of the opening paradigm: sharing information to overcome a large-scale challenge such as forecasting the virus propagation based on worldwide data collection and collaboration.

A year ago, the rise of technologies like ChatGPT resulted in discussions about protecting human rights or freedom and democracy by sharing information on how some algorithms are developed. More recently, a global discussion on open AI, which led to an open letter, has focused on how to take greater advantage of open-source algorithms in order to accelerate and challenge any piece of implemented algorithms to the benefits of all, while possibly significantly decreasing any threats.

The legacy paradigm

Information has been part of the economy and has been used as leverage to increase the power of an individual or institution. Scientific globalisation was made possible with the Gutenberg press in the 15th century and the Watt steam engine in the 18th century. These two innovations made it possible to share knowledge, discoveries and theories. The first countries that took advantage of the novelties were able to own the knowledge and the innovations, and increase their power. The same applies to companies or individual inventors.

Now, with big data and artificial intelligence (AI) feeding algorithmic models, information is collected, structured and processed automatically on a large scale to provide understandings, predictions or answers to specific questions. Despite the obvious benefits in many fields, AI presents threats we need to fight, such as discrimination and environmental impact. Additional challenges are the limitations on the size of datasets and the talent pool we need to access to come up with novel technologies. Opening some datasets and computer source codes can help us overcome these limitations and develop next-generation breakthrough innovations while protecting our fundamental rights.

Sharing information

Sharing information makes it easier for anyone to overcome the most challenging obstacles by accelerating the pace of co-innovation, facilitating inter and multi-disciplinary research and study, as well as expanding and propagating scientific knowledge and advanced research results.

During the COVID-19 pandemic, many countries shared their health statistics to feed predictive models and get relevant insights on the pandemic in a short time, which accelerated research in AI applied to healthcare and increased interest in accelerating the academic peer review publication process. Sharing information generally enables large-scale data-driven decisions to manage crises.

Some AI-based technologies require diversified large-scale datasets that often can be retrieved only by accessing open-data sources such as ImageNet, a platform used to train image recognition algorithms. Finally, training datasets from large databases enriches and diversifies perspectives by offering greater diversity and representativeness, thus decreasing the likelihood of bias and guaranteeing the inclusiveness of the resulting innovation.

This new paradigm based on sharing information also enables us to protect the fundamental rights of people as it encourages key players to share how they built technologies that can have a significant, and in many cases negative, impact on free will and democracy. The accelerated propagation on social media of conspiracy theories and fake news demonstrates the urgent need to make publicly available the recommendation algorithms on platforms such as X, Facebook, TikTok and ChatGPT.

Concrete examples

Open-research publication platforms such as Science Open, Open Access, ResearchGate or Welcome Open Research enable the sharing of research results and methods, thus accelerating and facilitating academic research and developments. They confront outputs, helping improve consensus, scaling solutions to practical problems faster and translating them more easily to industrial applications.

Open-source software and libraries are enabling faster co-developments by providing developers, scientists and engineers with ready-to-use computer programs and software functionalities with access to the source code (open-source software) or without (libraries and application programming interfaces (APIs)). The Python library named TensorFlow is commonly used by anyone implementing machine-learning algorithms.

Open-data platforms such as the World Bank Open Data or the World Health Organization’s open-data repository make possible the creation of representative training datasets to analyse an issue or to increase the accuracy of statistical metrics. This allows us to create more efficient algorithmic models to solve large-scale and complex problems. We could also mention the US Census Bureau, the Bold Open Database by Veuve Clicquot, or more recently Météo France, which will soon share publicly their data to leverage competencies from talented individuals for the analysis of climatology and real-time weather data.

How to act

As an actor in the private sector, you need to distinguish between the algorithmic technologies and data that are key in your intellectual property and business model, and the secondary ones that eventually support the first ones. You can also share specific pieces of your source code in order to take advantage of the open-source paradigm, including code and model benchmark, algorithmic bias detection or general improvements, while preserving your intellectual property for a given period.

In addition, without opening some of the source code of your technology, you can envision sharing and opening some or the entire dataset you used to build the algorithm(s) embedded in your technology.

Sharing the best practices that define part of your algorithmic governance might make teams and companies more competitive, since they become more trustworthy and therefore more attractive to users, consumers, the public and markets. Finally, sharing your mistakes, your failed attempts as well as your learnings is also critical in the openness paradigm, which will provide every actor with a safe space to share, discuss and challenge each other.

Conclusion

There is a growing discussion around “openness”, which is likely to become a standard vision for public and private institutions. Next-generation expectations include building and deploying a concrete and specific open strategy by defining the innovation components to open, such as the data, the algorithm and the source code, as well as the conditions for sharing. This is part of data and algorithmic governance.

About the author

Aurélie Jean is a computational scientist, entrepreneur and author, specialized in algorithmic modelling. She has a PhD in Material Sciences and Engineering, option computational mechanics and mathematical morphology, from Mines Paris, PSL University, France.

Guillaume Sibout is a consultant with In Silico Veritas (ISV), a consultancy specialising in strategies and governance of algorithms. He has an Executive Master’s degree in the digital humanities, innovation, transformation, media and marketing from Sciences Po.

Mark Esposito is Professor of Strategy and Economics at Hult International Business School and Director of the Futures Impact Lab. He is equally a Harvard social scientist, with appointments at the Center for International Development at Harvard Kennedy School; the Harvard’s Institute for Quantitative Social Science and the Davis Center for Eurasian Studies.

Terence Tse is a professor of finance at Hult International Business School.

With the hype around large language models such as ChatGPT, many firms are rushing to adopt the technology and integrate it in their operations. However, businesses should be cautious. Not all AI is created equal. Terence Tse, David Carle and Danny Goh pose three questions to guide companies in their AI adoption decisions.

To stay ahead of the competition, you have integrated artificial intelligence (AI) into your company. However, are you sure you have truly harnessed technology’s potential, or have you inadvertently stumbled into empty hype? Have you got the AI right or, more importantly, the right AI?

Considering the potential hazards that come with AI integration, we encourage businesses to answer three important questions prior to embracing this technology. By contemplating these questions in advance, companies can gain valuable insights into the potential advantages and drawbacks of implementing AI.

Is output accuracy a priority?

In the realm of “traditional AI”, or non-ChatGPT-type AI, accuracy serves as a beacon that guides the creation of effective AI solutions. However, with generative AI, typified by its ability to mimic human-like creativity, accuracy as a performance evaluation metric has seemingly lost its vital role. It is easy to fall for the allure of generative AI’s human-like touch, causing us to prioritise experiential value over strict accuracy.

By way of comparison, this is similar to desiring a string of dates without valuing the quality of the interactions. However, such preference can be risky, considering that generative AI has a reputation for generating deceptive content called hallucinations. This makes AI outputs unreliable, which can lead to wasted time and resources and expose companies to reputational and financial hazards.

From our front-row seats working with our clients, we have devised three methods to mitigate the hallucination problem. One method involves implementing a so-called “large language model (LLM) adaptor”, which relies on the use of multiple AI models rather than solely depending on a single one. This specific adaptor ensures effortless compatibility between different LLMs and empowers these models to select the route that will deliver the best accuracy and performance.

Incidentally, this also serves as a back-up plan. If, for instance, one of the LLMs has suddenly stopped working, the AI solution can continue to function on the feed from other models.

Another method is to implement a technique called “retrieved augmented generation”. This option involves imposing certain restrictions on the generative AI  model that can lead to more precise output, while also mitigating any instances of distorted information.

The third approach is to obtain the ability to track the origins of data.

Can your AI system trace the origins of data?

Picture a scenario where the HR team in your company recruits individuals just relying on the credentials they assert, neglecting to verify their CVs. The team completely ignores objective evaluations and makes recruitment decisions solely based on personal preferences. Would you perceive it as odd, if not outright unjust? Yet, it is how we embrace generative AI. We place trust, sometimes without question, on the output of generative AI.

While we may not have a foolproof solution for problem, we can implement measures to guarantee the AI’s information is sourced reliably. For instance, there is a huge difference between a model that generates a figure and a model that produces the same figure along with the details of the data origin: the name of the report, the page number and the exact location on the page where the data sit.

Data traceability is no fantasy. In our previous client works, we have been able to trace the original materials 100 per cent of the time. We strongly believe that only this way we can instil full confidence in our clients to depend and rely on generative AI.

Is the data secured?

In AI integration, safeguarding client data is a non-negotiable priority. Companies need to be mindful of the risks of relinquishing control over sensitive data. This is often due to the fact that many AI vendors currently require their clients to use cloud-based solutions. This can be dangerous as it means that not only the client’s data, but often the data of the client’s client will also have to sit in the cloud. This can put all the parties involved at unnecessary risk.

A much better approach is to deploy AI solutions using clients’ on-premises data centres instead. However, this may introduce more complex development and implementation processes. Yet, we believe the long-term benefits far outweigh the short-term troubles. To make the solution development part easier and faster, it is possible to train, develop and test many AI models using just small or even synthetic datasets and then migrate them to on-premises systems of the clients.

While embracing AI promises unprecedented opportunities, careful considerations are paramount. Not all AI are created equal, and navigating the business AI landscape demands caution. By asking the right questions and prioritising accuracy, data traceability, and data security, companies can chart a course toward AI integration that is both safe and sound.

About the Author

Terence Tse: a professor of finance at Hult International Business School.

David Carle: US Managing Director at Nexus Frontier Tech.

Danny Goh: a partner and commercial director of Nexus Frontier Tech, a co-founder of the technology group Innovatube and an expert with the Entrepreneurship Centre at Said Business School, University of Oxford.

Alors que la frénésie autour de ChatGPT s’est calmée, les entreprises se demandent comment exploiter réellement la puissance de l’IA générative.

Près d’un an et demi après la sortie de ChatGPT 3.5, les entreprises et les particuliers se sont précipités pour explorer les technologies d’IA générative (GenAI). Pour beaucoup, il y avait une peur palpable de passer à côté de la prochaine grande nouveauté, d’être dépassés par des concurrents capables de révolutionner leur entreprise, ou d’être pris au dépourvu par un changement radical à l’échelle de leur secteur. Rapport après rapport, certains experts ont vanté le pouvoir transformateur du GenAI dans tous les secteurs et ses implications sur l’avenir du travail (« Artificial intelligence’s promise and peril », de Hervé Tourpe, FMI, 2024).

Pour ajouter de l’huile sur le feu, nombre de médias nous ont rappelé que des emplois allaient probablement être perdus à grande échelle et rapidement.

Aujourd’hui, la frénésie GenAI semble s’être calmée – du moins marginalement. De nombreuses entreprises sont toujours confrontées aux mêmes questions qu’il y a un an : comment tirer parti des économies de coûts promises et des gains d’efficacité substantiels que GenAI est censé offrir ? Comment réellement procéder pour l’utiliser à des fins professionnelles ?

L’IA générative : Prometteuse, mais complexe

Beaucoup d’entreprises semblent être en difficulté. Il y a plusieurs raisons à cela.

• Premièrement, de nombreuses entreprises, peu importe leur taille, se demandent encore comment intégrer les IAs précédemment déployées (comme les algorithmes basés sur des règles explicites ou de l’apprentissage automatique) dans leurs opérations. Au mieux, ils sont dans une phase exploratoire avec ces IAs, et au pire ils se sentent simplement perdus. Une étude récente suggère que plus de 70 % des grandes entreprises interrogées se demandent encore comment tirer parti des avantages potentiels que l’IA peut offrir (« Becoming an AI-fueled organization Deloitte’s State of AI in the Enterprise », Deloitte AI Institute et Deloitte Center for Integrated Research, Deloitte Insights, 2023).
• Deuxièmement, la GenAI est beaucoup plus complexe que les autres IAs et est aujourd’hui conçue pour répondre à des objectifs spécifiques. Bien qu’elle soit capable de rédiger un rapport de 5 000 mots en un rien de temps, elle ne peut pas, par exemple, effectuer une tâche de saisie de données de base, comme extraire et classer les données du permis de conduire, que les autres IAs peuvent effectuer facilement et efficacement. C’est pourquoi les entreprises doivent réfléchir en profondeur à la rentabilité des GenAI en découvrant leurs avantages concrets et pratiques en comparaison aux autres IAs. Naviguer à travers l’IA, c’est comme naviguer dans des eaux agitées avec un navire à la pointe de la technologie mais quelque peu encombrant, et les GenAIs ajoutent plus de tonnage, de puissance et une mer encore plus turbulente. Une entreprise encore instable avec le premier aura bien sûr du mal avec le second.
• Troisièmement, les implications à long terme de l’adoption de GenAI – telles que les coûts à long terme et les impacts de la réglementation actuelle et future – sont encore incertaines. Pour nous, la situation actuelle nous ramène juste avant le millénaire. Même si les entreprises de l’époque avaient peut-être compris la nécessité de créer des sites Web, rares étaient celles qui voyaient clairement les rôles spécifiques que l’Internet au sens large jouerait en tant que partie intégrante des stratégies omnicanales, sans parler des appareils et des applications téléphoniques.

Compte tenu de tout cela, il est logique que la plupart des entreprises soient toujours à la recherche d’une voie à suivre (même si l’on a l’impression que tout le monde l’a compris). Cela ne veut pas dire que la recherche est une folie. Voici comment les entreprises peuvent s’orienter et déterminer la marche à suivre.

Le marché de GenAI

La première décision que la plupart des entreprises doivent prendre est de savoir quel produit GenAI elles souhaitent utiliser. À l’heure actuelle, il existe de nombreux fournisseurs de GenAI – à la fois des titans de l’industrie tels que Meta et Alphabet et de nouveaux venus comme Hugging Face, Anthropic et Stability.ai. Ce marché est appelé à devenir encore plus encombré, avec des sociétés riches en données telles que Bloomberg et JPMorgan Chase signalant leur intention d’entrer dans la mêlée, et Apple travaillant sur sa propre offre, appelée Ajax. Les entreprises devraient prendre en compte quelques facteurs.

D’une part, Open AI et ses rivaux actuels se disputent désormais le premier choix des développeurs de solutions GenAI, et les retardataires ont peut-être déjà raté le coche. L’introduction récente par OpenAI d’un outil simple pour créer des applications basées sur ChatGPT est probablement une tentative de consolider sa position, car les utilisateurs habitués à un système sont susceptibles de l’utiliser à nouveau dans leurs projets futurs. Avec la GenAI la plus grande – et sans doute la meilleure – du marché, OpenAI est la mieux placée pour établir un écosystème.

Cela dit, les développeurs de solutions ne voudront probablement pas prêter allégeance à l’un des fabricants de GenAI afin de conserver la possibilité de sélectionner GenAI pour différents projets. Cela a donné naissance à des boîtes à outils telles que LangChain, une plate-forme open source conçue pour permettre aux utilisateurs de travailler simultanément sur différentes GenAI.

La concurrence qui se joue entre les différentes sociétés GenAI ressemble un peu aux premiers jours du duel entre iOS et Android. Un vaste écosystème permettrait à OpenAI de rester le leader du marché (qui rapporte de l’argent) pendant quelques années jusqu’à ce que ses concurrents parviennent à unir leurs forces. Cela ne signifie pas nécessairement qu’Apple et Google s’uniraient pour concurrencer Microsoft. Plus probablement, nous verrions des concurrents s’entendre sur la même norme sur laquelle collaborer afin de s’opposer à la domination d’OpenAI. Nous pensons que cela n’est pas sans rappeler la situation de 2015, dans laquelle les partisans d’Android ont finalement réussi à établir un écosystème significatif pour rivaliser avec iOS. À mesure que le marché GenAI se consolide, nous pouvons nous attendre à voir deux à trois grandes factions se faire concurrence. Attendez-vous à ce que davantage d’entreprises, grandes entreprises technologiques et startups, redoublent d’efforts pour être au cœur de ces systèmes.

Considérations clés pour tirer parti du GenAI

Compte tenu de la situation actuelle, comment les entreprises pourraient-elles intégrer du GenAI ? Voici quelques suggestions :

➤ Choisissez la performance plutôt que la nouveauté

De notre expérience avec GenAI, ses performances ne proviennent pas de réponses textuelles de type humain de manière conversationnelle ou d’un modèle formé sur une grande quantité de données. Pour tirer le meilleur parti de GenAI, vous devez vous demander s’il s’agit de la bonne technologie pour une tâche ou un objectif particulier.

Par exemple, bien que ChatGPT soit (pour le moment) meilleur dans le traitement des mots et des langues, nous avons constaté que les modèles traditionnels d’apprentissage en profondeur donnent de bien meilleurs résultats dans le traitement des images. Autre découverte : dans un produit que nous construisons, nous avons constaté que ChatGPT-4 est plus efficace pour « comprendre » les requêtes des utilisateurs, tandis que la version 3.5 est plus rapide et plus efficace pour convertir les résultats traités en réponses aux utilisateurs.

En d’autres termes, au lieu d’adopter sans réserve les dernières technologies d’IA, les entreprises doivent comprendre les problèmes commerciaux qu’elles tentent de résoudre et trouver l’outil d’IA le plus approprié en fonction à la fois des forces et des faiblesses de chacune des options disponibles.

➤ Combinez le GenAI avec la puissance des bases de données vectorielles

Il s’agit d’une nouvelle forme de base de données spécialisée dans la récupération des données les plus proches statistiquement afin de répondre au mieux à des requêtes spécifiques (par opposition aux bases de données traditionnelles qui contiennent simplement les données collectées). Les entreprises peuvent utiliser une GenAI telle que ChatGPT pour décomposer les requêtes des utilisateurs, puis utiliser une base de données vectorielles pour rechercher les meilleures réponses correspondant à ces paramètres.

Prenons une analogie : si vous passiez un entretien pour un emploi, ChatGPT et ses concurrents offriraient la possibilité de « lire la salle », en analysant la posture, les expressions faciales, les choix de mots et les tons des intervieweurs. Les bases de données vectorielles, en revanche, agiraient comme des banques de mémoire et de sagesse, constituant la capacité de proposer les meilleures choses à dire.

En d’autres termes, GenAI à lui seul n’est peut-être pas suffisant. Selon les problèmes à résoudre, cela ne peut représenter que la moitié de la solution technologique. La nécessité d’une base de données vectorielles pour rendre GenAI vraiment utile signifie que les entreprises doivent s’attendre à faire face à encore plus de complexité et à des délais de mise en œuvre plus longs lors de la mise en place de la solution.

➤ N’oubliez jamais l’humain (human-in-the-loop)

Comme toujours, quelle que soit la puissance des technologies d’IA, leurs capacités dépendent de l’implication des humains. Ce constat est similaire pour les GenAIs. Les humains jouent un rôle essentiel en guidant les GenAIs vers les objectifs commerciaux, en gérant les interactions au sein des systèmes informatiques, en concevant les actions requises pour que les données entrent et sortent des modèles d’IA ainsi qu’en atténuant les hallucinations – les informations inventées ou carrément fausses produites par GenAI – cela reste aujourd’hui un problème majeur de GenAI.

➤ N’oubliez jamais l’explicabilité

Exploré depuis de nombreuses années, le domaine du XAI (eXplainabiligy AI) s’applique également à l’IA générative et à ses applications. Cela dit, lorsqu’on utilise une telle technologie, on devrait être capable d’expliquer certains (sinon la plupart) résultats, et erreurs possibles (telles que les hallucinations ou les discriminations technologiques) en appliquant les méthodes statistiques développées de XAI. Il convient de noter que XAI est un domaine en constante évolution qui progresse à un rythme rapide dans les laboratoires universitaires, les départements de R&D privés et les startups. Cela implique que chaque acteur utilisant de telles technologies doit mettre à jour ses connaissances.

➤ Tracez vos données

Même si le problème des hallucinations reste encore omniprésent, il est important d’établir une trace claire depuis la source de données jusqu’aux utilisateurs finaux. La traçabilité permet aux utilisateurs de connaître la source originale des données, ce qui renforce la fiabilité et la fiabilité des résultats de GenAI, créant ainsi une base plus solide pour une prise de décision éclairée.

Les entreprises doivent s’assurer que le traçage des données constitue une caractéristique importante à la fois dans leurs piles technologiques ainsi que dans leurs processus et flux de travail. Ce n’est qu’ainsi que les entreprises pourront être pleinement conscientes qu’elles utilisent les bonnes données.

Ayez des attentes réalistes

GenAI est un navire rapide avec beaucoup de choses qui se passent sous le pont. Il est difficile de savoir exactement quoi, dans quelle mesure et à quelle vitesse les entreprises GenAI peuvent réaliser. Croire avec conviction qu’elle peut produire des résultats immédiats et des rendements financiers exceptionnels conduira très probablement à des déceptions. Les dirigeants doivent reconnaître que le parcours exploratoire et expérimental de GenAI sera probablement long.

L’utilisation des technologies GenAI dans les opérations commerciales transcende un simple investissement technologique ; c’est un impératif commercial. Aussi difficile que cela puisse être en tant qu’entreprise, intégrer du GenAI dans les opérations de l’entreprise nécessite de comprendre les nuances des développements actuels de GenAI et d’avoir une conscience aiguë des défis présentés. Pourtant, pour les entreprises qui parviennent à utiliser GenAI avec succès pour atteindre leurs objectifs commerciaux, les récompenses ne peuvent être qu’à la fois prometteuses et énormes.

Aurélie Jean: Docteure en sciences, entrepreneure, et autrice, spécialisée en modélisation algorithmique, vit et travaille entre les Etats-Unis et la France. Elle a fondé deux entreprises : une agence de conseil et développement en data et en algorithmique, et une start-up de deeptech en algorithmique dans le domaine médical. Elle enseigne en formation continue, est investisseure business angel et est chercheuse invitée à la Hult Business School. Aurélie Jean est auteure de plusieurs essais sur la science algorithmique et contribue à de nombreux médias grand public dans le domaine des sciences et des technologies.

Terence Tse: Professeur, auteur et conférencier, il est le cofondateur et le directeur exécutif de Nexus FrontierTech, qui accélère les entreprises en créant des solutions d’IA. Professeur à la Hult International Business School, ses recherches ont donné lieu à de nombreuses communications et publications dans des revues et journaux professionnels. Il est titulaire d’un doctorat obtenu à la Judge Business School de l’université de Cambridge.

Mark Esposito: Professeur en gestion et en économie à la Hult International Business School et à l’université Harvard depuis 2011. Il est stratège en socioéconomie et connaisseur des changements liés à la quatrième révolution industrielle. Auteur de bestsellers sur les grandes tendances, les modèles économiques innovants et la compétitivité, il a cofondé Nexus FrontierTech, une entreprise de création d’IA, et Circular Economy Alliance. Il est également expert global du World Economic Forum depuis 2014 et senior advisor chez Strategy (PwC). Mark Esposito est aussi titulaire d’un doctorat à l’Ecole des Ponts ParisTech. 

Danny Goh: Entrepreneur et investisseur, il est associé et directeur commercial de Nexus Frontier Tech, un cabinet de conseil en intelligence artificielle présent à Londres, Genève, Boston et Tokyo. Il a aussi cofondé Innovatube, un groupe technologique qui comprend un laboratoire de R&D dédié aux logiciels et à l’intelligence artificielle, un fonds d’investissement et un incubateur. Expert à l’Entrepreneurship Centre de la Saïd Business School de l’université d’Oxford, il est aussi chercheur au Center for Policy and Competitiveness, à Paris.

Paul Lee: Paul Lee est cofondateur et PDG d’Aumeo Audio, d’ACE Communications et de Ximplar.