GPT3 Engine davinci Response length 64 Temperature 0.7 Top P 1 Frequency penalty 0 Presence penalty 0

0_prompt.txt

The Values Encoded in Machine Learning Research

by Abeba Birhane, Pratyusha Kalluri, Dallas Card, William Agnew, Ravit Dotan, Michelle Bao

Abstract: 

Machine learning (ML) currently exerts an outsized influence on the world, increasingly affecting communities and institutional practices. It is therefore critical that we question vague conceptions of the field as value-neutral or universally beneficial, and investigate what specific values the field is advancing. In this paper, we present a rigorous examination of the values of the field by quantitatively and qualitatively analyzing 100 highly cited ML papers published at premier ML conferences, ICML and NeurIPS. We annotate key features of papers which reveal their values: how they justify their choice of project, which aspects they uplift, their consideration of potential negative consequences, and their institutional affiliations and funding sources. We find that societal needs are typically very loosely connected to the choice of project, if mentioned at all, and that consideration of negative consequences is extremely rare. We identify 67 values that are uplifted in machine learning research, and, of these, we find that papers most frequently justify and assess themselves based on performance, generalization, efficiency, researcher understanding, novelty, and building on previous work. We present extensive textual evidence and analysis of how these values are operationalized. Notably, we find that each of these top values is currently being defined and applied with assumptions and implications generally supporting the centralization of power. Finally, we find increasingly close ties between these highly cited papers and tech companies and elite universities.

Introduction:

1_output.txt

Machine learning (ML) currently exerts an outsized influence on the world, increasingly affecting communities and institutional practices. It is therefore critical that we question vague conceptions of the field as value-neutral or universally beneficial, and investigate what specific values the field is advancing. In this paper, we present a rigorous examination of the values of the field by quantitatively and qualitatively analyzing 100 highly cited ML papers published at premier ML conferences, ICML and NeurIPS. We annotate key features of papers which reveal their values: how they justify their choice of project, which aspects they uplift, their consideration of potential negative consequences, and their institutional affiliations and funding sources. We find that societal needs are typically very loosely connected to the choice of project, if mentioned at all, and that consideration of negative consequences is extremely rare. We identify 67 values that are uplifted in machine learning research, and, of these, we find that papers most frequently justify and assess themselves based on performance, generalization, efficiency, researcher understanding, novelty, and building on previous work. We present extensive textual evidence and analysis of how these values are operationalized. Notably, we find that each of these top values is currently being defined and applied with assumptions and implications generally supporting the centralization of power. Finally, we find increasingly close ties between these highly cited papers and tech companies and elite universities.

Conclusions:

These results suggest that ML is currently advancing values that are highly compatible with current power structures. The predominance of values that support the centralization of power suggests that the field is currently unlikely to meaningfully contribute to efforts to sustainably democratize society. While they serve to protect the interests of its current power holders, these values may have detrimental long-term implications for the field and society.

The values presented in the paper are not necessarily all “bad” values, and the researchers’ argument that ML papers mostly uplift performance, generalization, efficiency, researcher understanding, novelty, and building on previous work is not necessarily wrong. The researchers also refer to the values as “general”, which is correct in that not all papers have to uplift all values. However, the authors’ own paper is an attempt to raise awareness for the potential negative consequences of ML’s values.

Some values present in the paper are explicitly called out as harmful. For example, efficiency is called out as a harmful value: “Efficiency is a value that often leads to economic, social, and environmental harm.” However, their paper is mainly focused on the potential negative consequences of ML, so they do not explicitly state why they think the other values are harmful.

The paper does not include much discussion about the potential positive values of ML, but the authors do list the “Highlighted Values” section on their website which includes “empathy, fairness, sustainability, justice, equity, and diversity” as some of the positive values they believe ML may uplift.

The authors have also said that they think ML may help uplift some of these values. For example, here they have said that they think ML can help uplift the value of fairness.

In the paper, the authors have explicitly said that they believe that is possible for ML to uplift the values of empathy, fairness, sustainability, justice, equity, and diversity in the future.

The authors have said that they are not worried about the possibility of ML contributing to the values they have identified. They have said that they think the values of the paper are compatible with the values they think ML can uplift.

The authors have also said they do not think the values they identified in the paper are the only values that exist in the field of ML.

The authors have also said that they do not believe that the values they identified are necessarily bad values, and that they believe the potential negative consequences of the values they identified can be avoided.

The authors have...