AI as a Cultural Asset and the Government's Imperative Role in Its Education" by Ariel Arrieta
Introduction:
AI is the deepest technology humankind has developed, more profound than fire, and can surpass human intelligence within this decade .
This technological evolution is bringing us unforeseen benefits only imagined by science fiction authors. Still, it also brings changes we cannot prevent, and there's hardly any time to adapt to the constant acceleration of these changes.
When we reach The Singularity ,? the human race will lose control over human development, ceasing to be the essential factor of change, moving from a leading role to merely being secondary actors in evolution. A new species will take over humanity's role, at least in this part of the universe.
These developments are inevitable and unstoppable, but we can act and influence how they occur. Just as we can impact and influence our children in their adult lives. Not directly by opining on decisions we don't understand but through the values and principles we help instill during their formative years.
Today, we are living through the formative years of the General Artificial Intelligences (GAIs) that will govern our world, and we have the opportunity and obligation to train them so they can represent our values and culture in the most faithful way possible.
Section 1: AI as a Cultural Mirror and Biases in Decision-Making
With the development of transformer technology and Large Language Models (LLMs), AI systems can make decisions based on the training they have received. Beyond representing a block of technology, they are based on a system of principles and values that represent the cultures on which they were trained.
These systems aim to make the "right decision" that is most likely to represent the data, principles, values, and logic with which they were trained, thereby incorporating all the biases included in the generation of that data.
Thus, beyond AI being a technological product, it is also a cultural product that represents the culture with which it was trained, as described in the paper "Machine Culture" published in the journal Nature.
Origin of Biases in AI: Artificial intelligence, especially in fields such as machine learning and natural language processing, is developed from large datasets. These datasets come from human sources: texts, images, social media interactions, and more. However, these datasets inherently reflect the cultures and societies that generate them. This means that the norms, values, and prejudices of these societies are embedded in the data. When AI is trained with these datasets, it inadvertently acquires the cultural biases present in them, leading to the creation of systems that can perpetuate these same tendencies and prejudices.
Examples of Cultural Biases in AI: The examples of cultural biases in AI are varied and revealing. For instance, facial recognition systems have shown a higher error rate for people of certain ethnic groups, reflecting the lack of diversity in the training data. In language processing, algorithms can exhibit gender or race biases, replicating stereotypes and prejudices present in the texts they were trained on. These biases reflect not only imbalances in the data but also in the priorities and perspectives of the societies that produce them.
Cultural Impact of Biases in AI: The impact of these biases extends beyond technical failures; they have significant cultural and social consequences. They can perpetuate stereotypes, exclude minority groups, and reinforce existing inequalities. For example, an AI system used for hiring that has gender biases can perpetuate gender inequality in the workplace. Similarly, a credit algorithm reflecting socioeconomic biases can reinforce economic inequality. These examples highlight how biases in AI can act as amplifying mirrors of cultural and social inequalities.
Responsibility and Solutions: Recognizing the existence of cultural biases in AI is the first step to addressing them. It is crucial that AI developers, investors, and regulators understand and actively commit to cultural diversity and inclusion at all stages of AI development. This includes diversifying datasets, implementing algorithms that detect and correct biases, and constantly ethical review of AI applications. Moreover, promoting greater awareness and education about how AI technologies reflect and affect our cultures is essential to ensure that these powerful tools are developed and used responsibly and ethically.
?
Section 2: Future Implications of General Artificial Intelligence (AGI)
Overview and Future Challenges:
The progression towards a future where General Artificial Intelligence (AGI) plays a predominant role suggests a scenario in which most conflicts and challenges will be addressed and potentially resolved through the perspectives and values deemed optimal by these advanced intelligences. The influence of AGI on society will not be limited to technological applications or task automation but will extend to critical decision-making based on the training and data they have been nourished with. Although the advancement of AGI seems to be an unstoppable force, its direction and effects can still be influenced by human intervention.
Technological Influence and Direction:
Kevin Kelly, in his book "What Technology Wants," offers a provocative interpretation of how technology, considered as an almost living entity, fundamentally intertwines with human experience. The concept of the "technium," coined by Kelly, describes this phenomenon as a self-reinforcing system of technological creation that follows evolutionary patterns similar to those of biological life. According to Kelly, even though technological development appears to have its own volition, driven by the forces of the physical universe, human history, and human agency, it is not entirely beyond our influence. Thus, AGI, as the culmination of this technological trajectory, is not exempt from being guided by human principles and values.
Shaping AGI with Human Values:
The possibility of configuring AGI to reflect our cultural and ethical values is a task that directly concerns us. The importance of this task lies not only in the technical capabilities of these intelligences but in their alignment with the fundamental values that underpin our societies. Thus, the interaction between emerging technology and human agency becomes the axis upon which we can guide the development of AGI towards a future that enriches and respects human diversity and ethics.
Education and Legacy of AGI:
Ultimately, AGI will represent a form of intelligence that surpasses human capabilities and even those of any existing organization. This scenario places us in a position similar to that of parents educating their children: while we exert a certain degree of influence over their decisions and Upbringing during the early years, eventually, these children will grow and choose their own paths, possibly achieving feats beyond our imagination. Similarly, by educating the current and future generations of AGI with our values, principles, and, fundamentally, with love, we aspire for them to act following what we value and wish for the world.
This approach, championed by prominent figures such as Mo, the founder of Google DeepMind, in his book "Scary Smart: The Future Of Artificial Intelligence And How You Can Save Our World," highlights the critical importance of our active and conscious participation in the training of AGIs. He explains that just as we educate our children with love and hope that they will make the right decisions when we no longer have influence, we must do the same with the training of AGIs. Thus, the formation of these artificial intelligences not only becomes a reflection of our technological aspirations but, more importantly, a legacy of our deepest human values for the future.
Section 3: What We Can Do and The Role of Governments in AI Education
An artificial intelligence (AI) system is founded on three essential components that work together to enable the machine to learn, reason, and act. Firstly, the models are algorithmic structures that, through machine learning, seek patterns and relationships within data to make predictions or decisions based on new information. The second component, the data, constitutes the core of AI learning, as it is the set of relevant information on which the models are trained; the quality, quantity, and diversity of this data are crucial for the performance and accuracy of the system. Finally, computing refers to the processing capacity and the computer resources needed to train AI models, which often require a significant amount of computing power and storage to analyze and learn from vast data sets, allowing for the optimization and continuous improvement of the models. These three components are closely interrelated, each playing a fundamental role in the effectiveness and efficiency of AI systems.
领英推荐
The 3-part structure of an AI:
1. Models: There are proprietary models (Chat GPT, GPT-4, Claude, Gemini, Groq, etc.) and open-source models (Llama). It is important to support and develop open-source models that promote diversity of access.
2. Computing: Access to GPU processing capacity is a resource with geopolitical implications, leading to an increase in value for companies like NVIDIA, TSMC, etc., but with very little actionable value.
3. Data: This is the most sensitive component of the system, where our contribution can make a significant difference.
The role of governments, once they understand the significance and fundamental change that AI implies, evolves in the following way:
Stage 1: They first attempt to determine the degree of danger it may pose to society. If this technology is capable of annihilating humanity or any part of it. After understanding the limitations of these questions and the impossibility of stopping progress, they move on to the next stage.
Stage 2: Ensure access to technology and understand the differences between proprietary and open models. The importance of data and training. Access to the models will evaluate the ability to access and the implications it may have in terms of productivity and the changes that will impact society. Sometimes, trying to regulate access, resources, or the legal framework.
Stage 3: Understanding which changes could have the greatest impact and that if governments do not make these changes, probably no one will, thereby rescuing the cultural value of each nation and group of individuals.
The development of datasets that encapsulate and represent the values and principles of each culture and its diversity will allow it to be represented in the decisions and values that will be embedded in the models.
Section 4: What Can I Do, Case Studies, and Best Practices
There is limited scope for developing new technologies in chips, new models, or exponential technologies that can change the operations of the forthcoming changes. It's a race for giants.
What we can do—and if we don't, nobody else will—is to train the models with our values, principles, and culture to be represented in the systems.
The time is now, just as a person in their formative years connects and learns from books, papers, professors, and mentors, and then, when more developed, learns from colleagues, the time to influence AGI models is now, as it will be very difficult to enact change later.
Specifically, American culture will be broadly represented in the cultural values with the data that the models are being trained on, while other more peripheral countries like Mexico or Japan are making efforts to incorporate their culture into the training datasets of the models.
Conclusion:
A tsunami of changes unleashed by the development of AGI, referred to as the singularity, is coming, and there is no way to stop it but it is possible to influence it so that this intelligence is aligned with our cultural values and can represent us.
This cultural representation is the differential that nobody will do for us, and every effort made in this direction will have a unique representational value that will likely be irreplaceable. Understanding artificial intelligence as a cultural product and seeking for that culture to represent us is the low-hanging fruit and probably one of the most impactful tasks that governments can do. Companies are already doing it, some individuals as well, but there is a lack of representation in the models that deserves to be completed to try and achieve models that represent us better.
?
Section 5 Annex:
If you want to delve deeper, here are some recommended contents:
Si quieres profundizar más acá hay algunos contenidos recomendados:
Este esquema proporciona una estructura comprensiva para abordar este tema desde una perspectiva profunda y matizada. Como experto en tecnología y cultura, puedes enriquecer este post con tus conocimientos y experiencias únicas, especialmente en el contexto de América Latina y la inversión en tecnologías emergentes.
?
- ?