MIT white logo

Ethics in Artificial Intelligence: Toward Systematic Empirical Analysis

Matters of ethics are becoming more salient at all level of politics, almost everywhere. In the scientific community, ethics in AI is increasingly gaining attention. The fact is that the rate of change in AI innovations and applications are growing much faster that our general appreciation or of our understanding of content or of consequences. 

Context

There are a large number of statements, but few ethical practices by countries, corporations, and individuals about that is desirable in the ethics domain for the broad area of Artificial intelligence. Far less frequent—if at all—are operational applications of ethics code in the innovation, practice, and policy of AI. To date, the focus of attention is on scientific and technical advances as well as enhanced computational advances.

Especially important in this connection has been the pervasive “ethics-in-AI gap,” that is, the near absence of attention paid to matters of ethics in Artificial Intelligence. At this point, there is a growing recognition that ethical issues cannot be ignored. Some have issued formal statements expressing their corporate position. Others, like Amazon, Google, Facebook, IBM, and Microsoft are collaborating to develop best practices in AI. Governments are gradually turning to these issues as well.

Of relevance here are the OECD’s AI Policy Observatory and the European Commission’s High-level Group on Artificial Intelligence, as well as the United States Commission on Artificial Intelligence.

As part of our background investigations, we have (a) identified and reviewed a large number of statement-as-formal-policy on ethics in AI in both the private and public sectors worldwide and (b) systematically recorded their central features.

We have also (1) identified countries with formal AI policies and (2) systematically recorded central features.

Yet to be examined is any relationship between AI policy on the one hand, and AI capability on the other. Of importance is capturing systematic relationship (if any) between:

  • AI policy and AI capability; 
  • AI capability and level of economic development; and 
  • AI capability and content of AI activity.

Proposition, Purpose, Premises

The basic proposition underlying this initiative is that the parameters for agreement on ethics in artificial intelligence remain largely uncharted and fraught with diverse types of unknowns. Our purpose is to anchor ethics in AI in a multidimensional context to facilitate alignment of AI ethics and AI practice.

Basic premises include (i) ethics as a central feature of any emergent international agreement on AI (ii) no constraints on research and exploring the “unknown rights (iii) Close review and assessment of machine-brain interactions (or interface systems) and (iv) attention must be given to culturally-based ethical considerations.

To reduce the dangers of undue simplification, or the trap of “one size fits all,” and to avoid implicit bias, we use four distinct, but interconnected, imperatives as a “basic checklist “and methodological guidance, to ensure that we remain on course, working towards an integrated and coherent system of “Ethics-in-AI.” These are:

  • Dimensions of Analysis, 
  • Domains of Interaction, 
  • Levels of Analysis, and 
  • Fundamentals of Foundations

Jointly these features provide solid basis for a robust framing of “Ethics-in-AI”

Program Activities & Expected Value

The proposed initiative consists of five activities, each designed to yield specific value added:

  1. AI General “State of the Art” to yield a comprehensive and “best review” on Ethics-in- AI. The value lies in creating a “system boundary” for the substantive inquiry and issues raised in the program design as a whole.
  2. Situating Ethics to identify the content of stated Ethics-in-AI and create of a database organized by key variables and sources. The value-added is distinguishing between aspirational and operational postures on ethics.
  3. Policy Analytics for Ethics-in-AI to generate an empirical database of AI policy structure, substantive content, as well as embedded features. The value-added lies in aggregating results to identify central tendencies, outliers, and other features,
  4. Ontology of Ethics-in-AI to create an empirically-based structured ontology of ethics-in-AI the database. The value-added is a searchable knowledge repository. 
  5. Ethics in AI and National Profiles to identify the statistical relationships (if any) between AI policies and state profiles (i.e., empirical features). The value-added lies in locating potential national propensities for particular Ethics-in-AI configurations.

Reference:

  • Choucri, N. (2022). Ethics in artificial intelligence: Toward systematic empirical analysis [Unpublished manuscript]. MIT Political Science.