Scaffolding Secure Development using AI

AI can release the imagination when put at the service of human intelligence. We can do more, better, and faster, using AI as a scaffold for development in Latin America and the Caribbean (LAC). As a growing reservoir of smart agency, the larger the number of people who can access and enjoy the opportunities and benefits of such a reservoir, the better our societies will evolve. AI can scaffold our capacity to look through the windows of our actual circumstances and to bring as-ifs into being.

Across LAC, AI can be a driver of Secure Development. AI can nurture diverse understandings of multi-layered points of view allowing us to imagine fresh possibilities. With the imagination, AI can fire a pathway to development. People look differently and they see the same thing differently.

The requirement is that we must look long enough to see. In painting, we use line and light to turn attention away and to carry the viewer away from merely “looking at” a thing. There is an awakening when we learn to notice what is there to be noticed; when we attend to what cries out to be attended to.

If developed thoughtfully, AI artefacts in LAC can offer opportunities to improve and multiply the possibilities of human agency.  Human agency can be supported, refined, and expanded by embedding indigenous knowledge into “facilitating frameworks” designed to improve the likelihood of morally good outcomes in this hemisphere. This could in turn amplify and strengthen distributed morality in human-to-human systems across the member states of the OAS. AI systems can also manage the coordination complexity of co-owned and co-designed artefacts.

Cutting and pasting the ontological and conceptual blocks of modernity is what AI-Assemblages do best when it comes to its impact on LAC. Design in this frame becomes the task of taking advantage of the affordances and constraints in view of solving problems like intergenerational immobility and inherited inequality.

Therefore, the cleaving power of AI decreases constraints on reality and escalates affordances. Design therefore becomes the art of solving problems through the creation of artefacts that take advantage of constraints and affordances, to satisfy some requirements, given a goal in view.

However, we cannot overlook the risks. The potential for algorithms to improve individual, and social welfare outcomes comes with significant ethical risks. In LAC it is assumed that outputs of translation and search engine algorithms are objective. However, they frequently encode language in gendered ways.

Algorithmic advertising of job opportunities for high-paying jobs in science and technology targets men more often than women. Prediction algorithms used to manage the health data of millions of patients worsen existing social problems in the USA. For these reasons, discourses around fairness, accountability, and transparency abound.

Studies of “inconclusive evidence” show how deterministic ML algorithms produce results in probabilistic terms. These models identify correlations between variables but not causal connections. As such they nurture the practice of “apophenia”: seeing patterns where none exists because massive amounts of data can offer connections that radiate in all directions.

The patterns produced may be the result of inherent properties of the system modelled by the data, the data sets themselves, or properties neither of the model nor the system. It is the case that trends observed in different data sets reverse when the data are aggregated. In other cases, poor data quality leads to “inconclusive evidence” to support policy decisions and human actions.

This gives rise to serious ethical risks. Insights from algorithmic data processing can be uncertain, incomplete, and time-sensitive. Lack of transparency is also an inherent feature of self-learning algorithms, which alter their decision-making logic. This makes it difficult for developers to comprehend why some changes are made.

Today, the conceptual space for the ethics of algorithms is fenced by six ethical parameters. Three are epistemic: inconclusive, inscrutable, and misguided evidence. Two are normative: unfair outcomes and transformative effects. One, traceability, is at the intersection of the two previous boundaries.

The epistemic elements highlight the justifiability of the conclusions that algorithms reach. This is critical since these shape morally loaded decisions for the environment, societies, and individuals. The normative concerns relate to the ethical impact of algorithmically driven actions and decisions, including lack of transparency (opacity), unfair outcomes, and unintended consequences.

Together, these six concerns form a new conceptual matrix that can shed light on ethical problems that algorithms pose at the micro-ethical level. This can spotlight how stubborn issues, data and responsibilities are intertwined, and the need for a macro-ethical approach as part of a wider conceptual AI space, namely digital ethics for LAC.

Ethics is not the preserve of one people or one place. The task is for AI developers to incorporate a more socially, geographically, and culturally diverse array of perspectives.  The Rawlsian view of justice as fairness brings into focus the correcting of past wrongs including eliminating unfair discrimination, fostering inclusion, promoting diversity, and barricading the reinforcement of biases, or the rise of new threats to justice.

Today, a tiny fraction of humanity is developing a set of digital tools that have already transformed the lives of everyone else. And so the future remains open to more suggestions from LAC.