Superethics instead of Superintelligence: Know Thyself, and Apply Science Accordingly | Pim Haselager, Giulio Mecacci, AJOB Neuroscience 2020, Vol. 11, No. 2, 113-119.

@ethics @superintelligence

Published in open source. Light read with a very sensible main message. There's quite a bit of repetition in there.

Abstract:

As the smartest species on the planet, we don’t need more intelligence. Since we appear to possess an underdeveloped capacity to act ethically and empathically, we rather require the kind of technology that enables us to act more consistently upon ethical principles. The problem is not to formulate ethical rules, it’s to put them into practice. Cognitive neuroscience and AI provide the knowledge and the tools to develop the moral crutches we so clearly require. Why aren’t we building them? We don’t need superintelligence, we need superethics. (113)

"Intelligence brought us technology and, recently, technology closed the circle and brought us intelligence." (113)

Thesis statement:

"The thesis of this short essay is that we need a more profound reflection on what our science enables us to understand about ourselves, and what our technology allows us to do with that knowledge." (114)

Specifically, CNS (analytically) and AI (synthetically) lead to insights about ourselves.

Claim: "Not only do they demonstrate the impressive level of scientific and technical mastery that we have achieved, but they also show what we want to do with those insights and technologies. They reflect our moral outlook, the kind of values that we are willing to pursue most of all, and those that we are willing to sacrifice." (114)

I'm not entirely convinced this is demonstrated. Does this claim not moralize technology already from the outset?

The potential apex of the AI pyramid currently being built is “superintelligence” (SI). As increasing funds go into the creation of Artificial General Intelligence (AGI), the risks that an SI may develop increases (Bostrom 2014; Russel and Norvig 2009; Schmidt 2018).

I'm not sure this apex is currently being built. So far we could mimic Nietzsche, saying "narrow, all too narrow".

References for friendly AI:

  • Alba 2015
  • Cellan-Jones 2014
  • Lewis 2015
  • Rawlinson 2015
  • Russell 2019
  • Tegmark 2015)

The article starts off from the realistic insight that technological innovation in AI is driven by economic considerations. But why spend millions on systems that are smarter than us and then again having to invest to keep them friendly? (114)

A.k.a. "What itch does SI scratch?" (114)

Yet another step in an arms race?

I really appreciate these two remarks:

  1. "But there is little to no discussion whether we really need increased intelligence at such great costs. " (114)
  2. "But it is hard to find that type of self-reflection regarding AI. To the extent that it is there, it gets bowled over in a feverish headlong pursuit of money." (115)

What are we missing?

So, from that long-term retrospective of future historians, the puzzle would be why we didn’t apply our science and technologies on improving our capacities to care and share. Why, after a rather tumultuous 20th century with two world wars and barely escaping the third, did we not draw the lesson that we should aim our best science and technology toward strengthening fairness and empathy, instead of intelligence?

The authors argue we should draw from work in CNS and social psychology to answer questions, e.g. related to Arendt's banality of evil?

What are the mechanisms that make us, healthy ordinary people, persevere in everyday life practices we know are collectively damaging?

Why is it so hard to act upon ethical principles that we already have?

They do discuss some drugs (the amphetamine "Captagon") and their potentiality of stimulating ethical or unethical behavior.

What's missing is the role technology may be able to play here.

What are the authors not suggesting? (p. 115)

  • No artificial moral agents that take over ethical decision making
  • No "ethical doping" (although then their discussion of above drug is slightly misleading)
  • Basically, in no form a way to delegate morality to technology

Instead, we need technology as moral crutches. They label this type of technologically applied ethics super ethics, because it transcends our current ethical (or ethically challenged) behavior:

We use the label “superethics” (alternatively “supra-ethics”) to indicate that we could use our science and technology to go beyond the ethical behavior we, as a species, generally display. But AI technology, based on recent neuroscientific insights, that supports us in ethical action is hard to find. There are so many startups in neuromarketing these days, where are the startups in applied neuroethics? (116)

Moral crutches

In short, what is missing is "technology that supports us in promoting and endorsing those values." (116) Attempts focus on embedding values into technological design to prevent unethical application:

Aristotle's virtue ethics points towards a possible solution: "ethical conduct depends on habit formation" (116).

In a non-ethical context, fitness apps already exploit this type of habit forming.

"A lot of their functionality aims to support habits by sustaining motivation, often using one’s social network to cheer you on." (116)

The authors ask why there is no such app for ethical habit forming?

"But what we are missing are the daily life moral fitness apps, based on applications for personalized use of artificial intelligence that help us in our moral self-improvement." (116)

(I suspect that whereas you have clearly defined goals in fitness, these are more ambiguous when it comes to ethics ).

But indeed, then the authors admit the large amount of moral frameworks, difference in "ethical sensibilities", or the reasons to use such "moral support systems".

A moral habit-forming app would need to support moral habit forming in a variety of contexts and for a variety of reasons.

"That will require additional research into how, when and why people are aware of a need or desire for moral self-improvement, e.g. when the pangs of remorse or shame hit them, or when the satisfaction of having done something good is still lingering on." (117)

They thus suggest to go the CNS route rather than the utilitarian route where an app explicitly scores your behavior? The paper says nothing about how such an app would really look like.