Which are the suitable strategies for effective human-algorithm interaction?
Effective beneficial human-algorithm interaction strategies require inclusive, democratic participation by humans in decision making. Algorithms should be subordinated to humans, serving them in such tasks as finding facts, proposing options and counting votes rather than making decisions that, because not collectively arrived at, cannot but be under-informed and biased.
Proportional gender representation is particularly key to intelligent decision making³ and in general propitiates success for both sexes4.
A good strategic bet for beneficial algorithmic interaction with humans might be to find and implement good automated methods for participatory and proportionally representative decision making, and adopt laws mandating governments to use these tools to obtain representative and specific mandates from the people as frequently as needed- at least for the most important questions that affect us all. The state-of-the-art in electronic voting allows us to ground democratic power in actual choice and consent. There is no excuse to continue the obsolete system of voting only once every X years for the blanket, static, and not even binding platform of some political party which often represents a minority when it wins.
This single strategy of algorithmically enabling and legally mandating proportionally representative decision making, if successful, could then serve to generate all other necessary strategies to ensure that algorithms serve all humanity, by democratic and representative, automated universal vote: Should flawed and/or unaccountable algorithms be allowed to replace humans? Should robo-signed mass actions against citizens be legal? Should algorithms pay taxes and benefits like the humans they purport to replace would? Should our laws allow for results of publicly-funded research to be used against the public, e.g. by creating unemployment? Should algorithms contribute to a fund for retraining humans into new jobs? Should our right to information be cancellable by the use of black box algorithms?
Even more importantly, this single strategy could also serve to generate the strategies that are urgently needed to push our Doomsday clock back from the two minutes to midnight it just hit: Should we all endorse the U.N.’s decision to ban nuclear weapons? What measures should be enacted towards ending violence, inequity, poverty, dominance, war, oppression, militarization, ecocide, climate change, etc?
We may not have the political will, as a society that has not quite reached true and representative democracy, to collectively and representatively induce the intelligent decision-making processes needed. But at least the necessary tools for reliably and efficiently mechanizing these processes are within our reach. As a community of scientists, we can, at this point of urgent need, decide to develop them and promote them into use in interaction with all other concerned groups besides ordinary citizens: scientists, educators, health care professionals, governmental agencies both national and international, legislators, judges, politicians, grass-root organizations, etc. Regardless of what means are chosen, our scientific community is already taking action, cf. their
3 https://futureoflife.org/2016/06/13/collective-intelligence-of-women-save-world/
4 https://news.ubc.ca/2014/09/30/gender-equality-olympics
campaign urging a U.N. treaty against killer robots, or STEM professionals’ initiative to more effectively manage technology and other resources crucial to human welfare.
6 http://demilitarize.org/milex-sign-new-statement-climate-change-military-spending/
10 Summary of the Panel Discussion “Algorithms’ Impact on Human Behaviour”
Verónica Dahl
Simon Fraser University
This discussion tied together Henk Scholten’s talk on “Digital Transformation and Governance on Human Societies”, Nicole Dewandre’s on “AI as an interesting leverage point to rethink humans’ relations to machines... and to themselves” (see Section 7); Carlos Castillo’s on “Algorithmic Discrimination” (see Section 8); and Fabien Giraldin’s on “Experience Design in the Machine Learning Era” (see Section 9). The moderator’s brief position statement stressed as urgent the need to regulate algorithms, to ensure in particular that the wonderfully powerful tool that AI represents is used only for beneficial impact on human lives and behaviours.
The main questions discussed were:
- The need to debunk the view of rationality as the highest human capability, establishing in political terms the relational self as both free and social, so as to approach AI in a more human-centered (as opposed to control, malecentered) way.
- The need to systematically develop a proper vocabulary and mindset that will allow us to define what we need to do, for whom, and with appropriate measures of how, if adopted, it will lead to a better state of the world.
- The need to develop a vision of fairness in the digital world, and of what it means to evolve with AI in a socially-mindful, rather than interest-led, way.
- The need to correctly conceptualize notions of fairness and privacy, which are sometimes incorrectly invoked for the sake of the rational subject’s interested wishes. While the separation line between using personal data for society’s benefit and protecting it as private might be sometimes unclear, a good rule of thumb might be that someone’s rights end where the rights of others begin, e.g. hiding behind encryption for criminal acts would warrant losing one’s “right” to such privacy. Where the separation line is really blurry, laws designed for partial compliance (as exist already in Europe) are usually preferable to algorithmically enforcing total compliance.
- The need for transparency and accountability, defined as giving the public the ability to challenge an algorithm’s decision (N.B. this is different from having all details about the algorithm, which may be useless in terms of challenging it). At present, Software Engineering cannot verify whether our complex, often statistically-based, unpredictably but speedily evolving AI systems will behave as planned, but perhaps a way will be found in the future. Meanwhile it may be prudent to legally disallow algorithms that cannot deliver transparency and accountability where it is due.
- The question of why are we being so suspicious or untrusting came up: it goes back to the degradation of human-to-human relationships under our globalized neoliberal economic politics, which normalizes human instrumentalization. For as long as “progress” means to use less people (improve labour productivity), and machines are being built to dispense with humans, it is “natural” for humans to end up wondering when machines will become “human”. A way out might be to legislate that the uses of (publicly funded in particular) AI must benefit, not hinder, the public, e.g. by making algorithms (in fact, those operating them) pay taxes and benefits, and contribute also to a fund for retraining into new jobs the people they’ve disloged.
- The need to decide as a society, and legislate, who controls the results of AI, what is done in AI, and for whom. The absence of adequate laws for common good endangers modern political order, since a few very powerful global companies can profoundly influence many areas of daily life unchallenged, potentially generating
| informational dictatorships able to manipulate the behaviours of humans and organizations alike, and even to erode representative democracies and world peace. | |||
|---|---|---|---|
| Some possible strategies were also put forward: | |||
| • | through valuing what is widely shared. | We should define machine intelligence not by what exceptional people can do, but | |
| • | Protect the attentional sphere of the users from the demands of multiple systems. | ||
| • | Educate governments and the public about AI’s fallibility and limitations, e.g. neural net based systems cannot be totally autonomous given that they can have catastrophic failures embedded which cannot be anticipated until they occur. | ||
| • | 7 generating the most intelligent decisions | Conjure as much, and as representative, citizen participation as possible for decision- making on how to instrumentalize AI and technology in general for social benefit (e.g. ). Many frameworks are possible, which must be balanced for efficiency and to not cause fatigue, e.g. as part of municipal bills of rights to be developed, or as machine assisted collective vote for the more critical issues ( the framework, proportional representation stands out as the crucial ingredient for 9. | See more details in8). Whatever |
| • | Work with legislators to bring about 8 George Monbiot (2017) Out of the Wreckage. Verso. | main dangers of unregulated algorithms (such as technological unemployment, algocracy, killer robots), and will ensure that they are used to help solve humanity’s present problems for universal benefit rather than for that of a privileged few. 9 [https://futureoflife.org/2016/06/13/collective-intelligence-of-womensave-world | the](https://futureoflife.org/2016/06/13/collective-intelligence-of-womensave-world |
| 32 |