Weaponized electronics: A wake-up call for supply chains
The deadly blasts from tampered devices in Lebanon highlight the fragility of supply chains, making it clear that better security measures are urgently needed....
by José Parra Moyano Published 13 December 2023 in Technology • 7 min read • Audio available
In a world where AI is increasingly shaping the future of industries, the key to gaining a competitive edge lies not just in the algorithms that give rise to the AI, but in the quality, diversity, and integrity of the data that powers those algorithms.
The world is not short of data. Thanks to the rapid digitization of our economies and growing prevalence of smart devices, the amount of data produced globally is large and rapidly increasing. This represents a huge economic opportunity. The European Commission estimates that the EU’s data economy alone will be worth €829 bn in 2025, accounting for around 6% of regional GDP.
Yet many firms struggle to gain access to the high-quality data that is needed to train generative AI models. Those that rely on off-the-shelf AI solutions find themselves at a disadvantage since these generic models lack the tailored insights and specific nuances necessary for a true competitive edge.
Thus, organizations looking to get ahead need to train their AI with proprietary, unique data that will reveal insights that none of their competitors can gain (simply because they don’t have that proprietary data). In other words: those who control proprietary, high-quality data will control the AI systems that generate the most value.
The training of an AI with proprietary, unique data can be done either by fine-tuning existing models or by developing proprietary ones from scratch. Fine-tuning involves adjusting pre-trained models on specific data relevant to an organization’s unique needs, allowing for more accurate, business-specific outcomes. Developing proprietary AIs from the ground up with specialized datasets implies building the AI exclusively within the boundaries of the organization, using proprietary data and algorithms.
However, the amount of data required to fine-tune an existing AI, let alone to train a proprietary one, is non-negligible, and thus beyond the reach of many small, medium, or even average-sized organizations. This represents a challenge for companies willing to harness the power of AI to gain a competitive advantage, as they often find themselves in a position where they simply cannot get the data they need to better compete. To use a mechanical analogy, they may have the car, but they lack the fuel to be able to drive it.
Even some of the world’s most prominent generative AI companies like OpenAI, the company behind ChatGPT — which until now have acquired large parts of their data from scraping the web — may find it increasingly hard to get their hands on this valuable resource. OpenAI is facing a slew of copyright challenges, while owners of intellectual property are starting to take proactive steps to protect their content.
Against this backdrop, a new approach is emerging that allows companies to improve their AI capabilities through collaborative efforts, while complying with strict data privacy standards. Instead of amassing reams of data to feed the algorithm, you can bring the algorithm to the data.
One such approach is secure, multi-party computation, a method that allows for the training of AI models using decentralized data. In this model, the algorithm is sent to various data sources, learns from them locally, and then only the insights or model improvements are aggregated centrally, without ever transferring the raw data. For example, several hospitals can train an AI to diagnose a lung disease without the need for any of those hospitals to share the personal data of the patients who have the diseases. This enables the collective training of the AI in a way that secures the privacy of the patients.
This methodology not only helps to safeguard the privacy and security of each organization’s data, but also enables the AI to learn from a diverse and vast array of datasets, which might otherwise be inaccessible due to privacy concerns or regulatory restrictions. By leveraging secure, multi-party computation, organizations can mutually benefit from each other’s data, enriching their AI models with a wider spectrum of knowledge while upholding the principles of data confidentiality and privacy.
Since this approach preserves privacy, organizations can also use it to monetize their data by using it to train other organizations’ algorithms. Given the demand for AI to fine-tune, or to develop proprietary AIs, that data is in high demand, and thus extremely valuable.
As Frederic Pont from Tune Insight (a Swiss company developing an encrypted computing platform) says: “Novel approaches that combine federated learning and privacy-enhancing technologies release the tension between data utility and privacy, even for the most sensitive types of data in health or financial sectors, facilitating the training and validation of AI models without the need of direct data access, but also enabling the valorization of a company’s own data, without ever transferring or disclosing it.”
The data to train the algorithms doesn’t have to just come from organizations; it can also be owned by individuals. This is creating a new environment that empowers citizens, in part because of privacy laws in many areas such as the General Data Protection Regulation (GDPR) in the European Union and the California Consumer Privacy Act (CCPA), both stating that a person’s data belongs only to that person.
Some of these citizens are starting to form data cooperatives. Simply put, these are collaborative organizations where members voluntarily share and manage their data collectively for mutual benefit and decision-making. To monetize the data that the cooperatives collect, and yet respect the privacy of their members, cooperatives can (and are starting to) implement the notion of “bringing the data to the algorithm.”
One example of such a cooperative is MIDATA, a Swiss-based cooperative focusing on personal health data. This cooperative uses secure multi-party computation techniques to ensure that individual health data remains private, while still providing valuable insights to healthcare researchers and professionals.
Interestingly, data cooperatives break the traditionally existing zero-sum game between getting insights from data, and respecting the privacy of those whose data is analyzed.
Another cooperative is Swash, which serves as a data cooperative by enabling data monetization through secure multi-party computation. Users can install a Swash browser plug-in which lets them earn money as they browse. At the same time, businesses can gain access to high-quality data without having to go through third parties. Swash represents a shift towards the generation of passive income generation for citizens, while ensuring data privacy.
Rita, a data cooperative focused on advertising and marketing, represents yet another such example. These cooperatives illustrate a growing trend where the value of data is harnessed in a way that respects individual privacy and offers a collective benefit, showcasing the practical application of secure data-sharing methods like multiparty computation. In a sense, such cooperatives are generating returns while (or precisely because) they respect the privacy of their members. This is a form of monetizing privacy.
Data cooperatives are on the move and the World Economic Forum has set up guidelines to design sustainable data cooperatives. In that vein, the Institute for Human-Centered Artificial Intelligence at Stanford University is studying how data cooperatives can give consumers more power over their data. The relevance of such cooperatives is set to increase in the future, partly because of the augmented interest of organizations in using the data that cooperatives can gather, to train their very own AIs.
Given the increasing demand for data, motivated by organizations’ need to improve the competitiveness of the AIs that they use, data cooperatives are set to become very relevant players in the digital economy. Interestingly, data cooperatives break the traditionally existing zero-sum game between getting insights from data, and respecting the privacy of those whose data is analyzed. It remains to be seen how these initiatives will unfold in the upcoming years, but what is certain is that the current “reshuffling of the competitive landscape” that is taking place because of the possibilities —and threats— of the use of AI will force business leaders to keep an eye on these organizations.
Professor of Digital Strategy
José Parra Moyano is Professor of Digital Strategy. He focuses on the management and economics of data and privacy and how firms can create sustainable value in the digital economy. An award-winning teacher, he also founded his own successful startup, was appointed to the World Economic Forum’s Global Shapers Community of young people driving change, and was named on the Forbes ‘30 under 30’ list of outstanding young entrepreneurs in Switzerland. At IMD, he teaches in a variety of programs, such as the MBA and Strategic Finance programs, on the topic of AI, strategy, and Innovation.
17 October 2024 • by Öykü Işık, Carlos Cordon in Technology
The deadly blasts from tampered devices in Lebanon highlight the fragility of supply chains, making it clear that better security measures are urgently needed....
20 September 2024 • by Jerry Davis in Technology
Information technology can reduce the transaction costs of using markets. For labor markets, however, the costs have merely been transferred from business to workers....
23 August 2024 • by Luc de Brabandere , Jonas Leyder, Lina Benmehrez in Technology
In today's fast-paced world, guidance is needed not on what to think but how to think....
13 August 2024 • by Julia Binder in Technology
The legacy of these Olympics extends far beyond the final medal count, providing a model for how global events can contribute positively to both the environment and society....
Explore first person business intelligence from top minds curated for a global executive audience