Innovations and challenges of AI in education, society, and work. Ethics, human-AI collaboration, PA technologies, and investments in Italy, in an exclusive interview with Paola Pisano.
Our journey into the deep universe of technology starts with understanding AI’s history and its potential impact on human activities.
The world is in a new era of inevitable changes. In this delicate transformation, we want to analyze closely the innovations and the challenges of AI in crucial areas of society.
Few technologies have integrated into society as rapidly as Artificial Intelligence (AI). As we envisage the economy of the future and the geopolitical landscape, AI stands as its cornerstone.
In this time, the implementation of these technologies and the necessary investments are the center of debate, especially in Italy.
Today we’re surfing through innovation with Paola Pisano, Former Minister of Innovation and Digitalization of the Italian Republic, and Associate Professor at the Department of Business and Innovation Management at the University of Turin.
Can you outline the key stages of your journey in artificial intelligence that you’ve either observed or been engaged in throughout your career?
Artificial Intelligence (AI) has traversed a journey marked by four pivotal phases. I’ve had the privilege of closely observing and participating in the latter two stages, both during my tenure in government and as a professor in the Department of Economics and Statistics at the University of Turin.
The inaugural phase, spanning from the 1950s to the 1970s, heralded AI’s emergence as a distinct branch of computer science. This pioneering era laid the groundwork, introducing seminal concepts such as machine learning and neural networks, which remain indispensable to the field today. The subsequent phase, unfolding during the 1980s and ’90s, witnessed the advent of the first expert systems. Notably, technologies like the acclaimed MyCin system employed extensive rule sets to diagnose infectious diseases based on patient symptoms, showcasing AI’s capacity to address tangible challenges within specific domains.
The third act begins at the dawn of the new millennium and continues until 2020. These years, marked by the invention of the web browser and the spread of the internet, radically transformed our way of living, creating a more digital and interconnected world. This scenario enabled the generation of massive amounts of data, catalyzing the development of machine learning systems. Although AI began to permeate the devices we use every day, such as smartphones with their search algorithms and recommendation systems, its presence remained almost invisible to the public.
The last phase, beginning in 2020, is characterized by the advent of generative artificial intelligence. Today, this technology is not only a concept accessible to researchers but has become an integral part of our everyday life. Thanks to dialogue interfaces that use natural language, AI is now within everyone’s reach, making it possible for anyone to use advanced tools without the need for specialized computer science skills.
What do you consider to be the most noteworthy innovations and formidable challenges in recent years within the realm of AI development?
Three groundbreaking innovations have emerged as pivotal in recent years: word embedding, the attention model, and the Hugging Face platform.
Word embedding, introduced in 2013, heralded a paradigm shift in natural language processing. This innovation converts words into numerical vectors, effectively encapsulating the semantic essence of each term. By doing so, it empowers machines to comprehend and manipulate language with unparalleled precision. As a result, we’ve witnessed remarkable advancements in machines’ capacity to interpret and generate text in a coherent and contextually relevant manner.
Introduced with the Transformer model in 2017, the selective attention technique further revolutionized AI. This approach allows models to focus on specific parts of the input, significantly improving the quality of responses and the efficiency of the learning process. This mechanism is fundamental for processing complex data sequences and optimizing performance in tasks such as automatic translation, text understanding, and content generation. Launched in 2019, Hugging Face has democratized access to artificial intelligence technologies. The site facilitated knowledge transfer in the AI field, allowing researchers and developers to build on the latest discoveries rather than starting from scratch. The platform has greatly accelerated the innovation cycle, making AI resources more accessible to a global community.
However, AI development has also presented formidable challenges, both economically and socially. The costs for training advanced models, as highlighted in Stanford’s “Artificial Intelligence Index 2024,” have grown exponentially from $930,000 in 2017 to $191.4 million in 2023 spent by Gemini Ultra. This trend limits the development of these models to a few economically solid entities. Moreover, this rise in costs raises questions about the economic sustainability of the technological race towards increasingly complex models.
At a social level, AI challenges by making cheaper skills of creativity, reasoning, and problem-solving, traditionally seen as human prerogatives. The issue of how artificial intelligence can coexist with humans, supporting and expanding our capabilities without overpowering or diminishing them, is a crucial debate for the future.
How do you foresee artificial intelligence reshaping our societal framework in the next decade?
The integration of artificial intelligence (AI) into our society looms as one of the defining narratives of the coming decade. However, it brings forth an inherent risk: if not managed carefully, AI could exacerbate existing disparities. Technological advancements have the potential to disproportionately benefit those already endowed with wealth and technological proficiency. This trend is evident at the national level, with the United States leading with 109 foundational AI models compared to China’s 20. Such technological supremacy, concentrated in a few countries and private entities, may exacerbate global economic inequalities, granting them a competitive advantage across various sectors.
At the individual level, this technological disparity could translate into unequal job opportunities. Consequently, societies are confronted with the formidable task of crafting policies that ensure equitable AI access. This entails democratizing its benefits across all societal strata and nations, thereby potentially narrowing—not widening—the global inequality gap.
How should AI be incorporated into education to adequately prepare future generations?
As the digital age surges forward, universities worldwide are increasingly introducing degree programs focused on artificial intelligence (AI), often housed within their computer science departments. A groundbreaking illustration of this trend can be found in the United Arab Emirates, where the world’s first AI-focused university graduates 200 students annually.
This laudable initiative underscores the critical importance of integrating AI knowledge beyond traditional technical pathways. The versatility of AI technology extends across diverse academic disciplines, including Arts, History, Philosophy, Literature, Economics, and Law. Furthermore, there is an urgent need to introduce AI education at the secondary level or even earlier. Notably, thirty U.S. states now mandate computer science courses in their secondary curricula, likely incorporating AI topics.
At the University of Turin, within courses such as Innovation and Management of Economics, a significant portion—approximately 20 out of 50 hours—is dedicated to AI exploration. This encompasses not only functional explanations but also an in-depth examination of its applications and implications for businesses and nations. Moreover, students benefit from personalized support provided by an AI agent, adeptly trained on the course material, serving as a valuable resource akin to a tutor in resolving doubts and uncertainties.
Nevertheless, the integration of AI into education fuels an ongoing debate, particularly regarding the adaptation of student assessment methods. As educational institutions navigate this evolving landscape, addressing these challenges will be crucial to ensure the effective incorporation of AI knowledge into academic curricula.
What are the most promising research and development areas in AI for the future?
We are in an era of heightened interest, where companies across all sectors envision every conceivable use case and activity as a potential application for artificial intelligence.
Among the sectors where AI promises the most significant impact is the pharmaceutical industry. From identifying new, hard-to-detect diseases to developing novel drugs (with Isomorphic Lab and Deepmind leading the charge), from personalized medicine determining the correct drug dosages and administration timings for patients to predicting epidemics (specialist firm Airfinity), AI is revolutionizing healthcare. It also aids in diagnosing diseases like skin cancer, liver, renal and urological disorders, brain conditions, and Alzheimer’s disease.
The financial sector, too, has embraced AI technology across three main areas reflecting the endeavors of nearly all enterprises: pattern recognition in data for service personalization or fraud detection and prevention; process automation like customer-interacting chats; and prediction using vast data amounts to forecast stock or sector performance.
In Italy, Intesa SanPaolo is among the first banks to leverage AI for cost reduction and revenue generation. The company has identified 15 new AI use cases, established an AI unit for technical and economic due diligence before case development, and collaborates with startups (such as Aptus.AI) and tech giants like Google Cloud and Microsoft for ongoing technological advancements. Internationally, while Bloomberg can afford to create its large language model, Morgan Stanley employs prompt tuning and Retrieval-Augmented Generation (RAG) techniques to customize Chat GPT 4 for financial advisors, providing them rapid access to relevant knowledge for tackling client problems. The prompt training system is deployed on a private cloud, exclusively accessible to Morgan Stanley employees. In its first month, the model answered 25,000 questions at a cost of $0.002 each, totaling $3,000 – not including the recruitment of about 20 managers in the Philippines, continually analyzing and evaluating documents on multiple criteria.
Consulting firms are also making substantial use of generative AI, with companies developing a variety of use cases experimenting with performance improvements among consultants. In the programming domain, AI, as the first technology developed to adapt to humans rather than the other way around, is not only narrowing the gap between laypeople and programmers but also enabling programmers to work more swiftly and scalably.
What is your stance on the role of ethics in AI, and how can we ensure responsible AI development?
The incorporation of ethics into artificial intelligence (AI) goes beyond a moral obligation tied to upholding fundamental human rights and fostering a democratic ethos in technology. It’s also a fundamental aspect of operational efficiency. An AI system devoid of ethical principles risks faltering in its intended tasks. For instance, within the context of candidate selection, an unethical system may overlook the most qualified applicants, unfairly discriminating against individuals based on unjust criteria.
I am convinced that the development of explainable AI (XAI) systems is paramount. These systems provide clarity to users on the variables that heavily influence the AI’s decisions, thereby increasing transparency and trust in AI usage. At the AI and Data Lab of the Department of Economics and Statistics at the University of Turin, which I have the honor to lead, we implement XAI systems in all our AI projects.
Furthermore, the recent European legislation on Artificial Intelligence (EU AI Act) stresses the need for controls to ensure ethical principles are upheld. This includes transparency regarding the data used for training systems and an understanding of the capacities and limitations of the models. These requirements are fundamental to ensuring that AI is used in a manner that is responsible and beneficial to society.
In your view, how should we address the issue of job displacement with automation and AI?
When addressing the issue of job displacement resulting from automation and AI, I foresee three distinct scenarios. First, there are instances of replacement, where AI substitutes certain tasks or functions currently performed by humans. Second, there are cases of enhancement, where individuals can overcome skill deficits by leveraging AI, such as in language learning. Lastly, there are instances of transformation, where new roles and tasks emerge, necessitating adaptation and redefinition of work processes.
A recent study (Eloundou et al., 2023) suggests that approximately 50% of activities could be partially automated with Large Language Model (LLM) systems. The crucial factor lies in understanding the pace of this automation. If it occurs gradually, our workforce will have the opportunity to undergo training and adapt, as seen in past economic revolutions. However, if automation accelerates rapidly, the disruptive force of AI could have adverse effects on employment.
Therefore, it is imperative to develop appropriate policies at both the national and corporate levels by comprehensively understanding the speed of this transformation. This understanding should also consider the economic incentives associated with replacing human workers with AI systems.
How do you envision the optimal collaboration between human and artificial intelligence in the coming years?
In the coming years, I envision the optimal collaboration between humans and artificial intelligence characterized by an efficient allocation of competencies. Ideally, AI would undertake tasks that are repetitive, dangerous, or less stimulating, thereby relieving humans of these burdens. This liberation would enable us to focus on more creative, rewarding, and high-value activities, enriching the quality of our work and allowing us to explore and develop our unique capabilities further.
Furthermore, AI can serve as a powerful tool to extend or augment our natural abilities, providing assistance in areas where we may lack expertise. For instance, despite my lack of innate drawing talent, I have personally experienced how AI, through tools like MidJourney, can enable me to create visual works in ways that would otherwise be impossible.
What are the critical technologies that Public Administration (PA) needs right now?
At this juncture, the Public Administration (PA) stands to benefit greatly from prioritizing certain technologies. Foremost among these is cloud computing, which offers the potential to significantly reduce operational and maintenance costs associated with traditional data centers while simultaneously enhancing flexibility and agility in managing vast volumes of data and services. Moreover, within the cloud infrastructure, the PA can leverage Artificial Intelligence (AI) and Machine Learning (ML) tools to optimize decision-making processes, analyze extensive datasets to predict trends, and enhance service personalization.
For instance, my lab collaborates with the government on predictive analytics, particularly in conflict forecasting. We’ve developed an early warning system that monitors extensive databases, providing insights into the probability and various factors contributing to potential violence in a country. Given the escalating cyber threats, it is imperative for the Public Administration to implement advanced cybersecurity solutions to safeguard sensitive data and critical infrastructures from external attacks.
Additionally, digital service platforms play a pivotal role in improving accessibility and efficiency for citizens and the administration alike. By facilitating access to online public services, these platforms streamline processes, reduce time and costs, and ultimately enhance overall service delivery. This aligns with the technological policy I championed during my tenure as Minister of Digitalization and Innovation, where initiatives such as the cloud project, the establishment of a national cybersecurity authority, and the development of the IO platform by pagoPA aimed to consolidate services onto an unified platform, fostering greater integration and efficiency across the board.
What significant contributions is Italy making in the field of AI?
In 2020, Italy embraced a national strategy for AI, geared towards fostering the development and responsible deployment of artificial intelligence across the country. The strategy encompasses six overarching goals: enhancing both basic and applied research, mitigating sector fragmentation, bolstering support for the public sector, and attracting and retaining talent in pivotal industries such as manufacturing, education, agri-food, tourism, culture, healthcare, and green infrastructure.
Italy has further demonstrated its commitment to AI advancement through substantial investments. A dedicated investment fund of one billion euros, overseen by Cassa Depositi e Prestiti’s (CDP) Venture Capital unit, has been earmarked for AI development initiatives. Additionally, the recent launch of Frontech—an accelerator program focusing on startups pioneering innovative digital solutions in AI, as well as the metaverse and/or web 3.0—underscores Italy’s dedication to nurturing AI-driven entrepreneurship. With a projected funding allocation of 7 million euros, this program aims to fuel the growth of cutting-edge AI ventures.
In the private sector, Italian companies are increasingly integrating AI technologies to drive productivity enhancements, personalize services, and spearhead innovation across various sectors including manufacturing, fintech, healthcare, and retail. Furthermore, a burgeoning ecosystem of startups specializing in AI customization is thriving. Notably, Fastweb is embarking on an ambitious endeavor to develop a wholly Italian artificial intelligence, rooted in the nation’s rich history and culture. This initiative reflects Italy’s aspiration to carve out a distinctive niche in the global AI landscape while leveraging its unique heritage and expertise.
What advice would you offer young researchers aspiring to pursue a career in artificial intelligence?
Dispensing advice is always a nuanced task, but here’s an attempt.
First and foremost, prioritize building a robust foundational education. Subjects like mathematics, statistics, and computer science form the bedrock of AI, making degree courses in computer science, engineering, applied mathematics, economics, and statistics excellent starting points. However, recognize that AI is not confined to a single discipline; it thrives on interdisciplinary collaboration. Consider integrating AI studies with other fields like chemistry or biology to cultivate a unique and versatile skill set.
International exposure is invaluable in broadening your horizons and expanding your professional network. Engage with researchers and professors worldwide to gain diverse perspectives on AI’s potential and applications. Embrace curiosity and discipline as guiding principles. AI research often involves experimentation and occasional setbacks before breakthroughs occur. Maintaining a spirit of exploration and resilience is crucial.
In both sports and science, success often hinges on unwavering commitment and perseverance. Remember that every significant discovery begins with a question and the relentless pursuit of answers over time. Stay curious, stay disciplined, and embrace the journey of discovery with unwavering determination.
With a background in Economics, an MBA in Finance, and a PhD in Economics and Innovation, Paola Pisano currently holds the position of Professor of Economics and Business Management at the University of Turin. Here, she also serves as the Director of the AI and Big Data laboratory, while holding the position of Deputy Director at the Dish interdepartmental center for digital Humanities.
Paola Pisano played pivotal government roles, serving as the Minister of Technological Innovation and Digitization during the Conte II Government and as advisor to the Minister of Foreign Affairs and International Cooperation and the Minister of Public Administration in the Draghi Government.
