AI systems often rely on data that can reflect societal biases. If not carefully managed, these biases can lead to discriminatory outcomes in areas such as hiring, criminal justice, lending, and healthcare. Ensuring fairness in AI involves scrutinizing training data and algorithms to prevent perpetuating existing inequalities. Ethical AI development seeks to create systems that promote inclusivity and equal opportunity, mitigating the risk of harm to marginalized groups.

As AI systems increasingly automate decision-making processes, clarifying accountability becomes essential. In the event of errors or harmful outcomes, it must be clear who is responsible—whether it's the developers, organizations, or the AI itself. Establishing accountability frameworks encourages responsible AI usage, promoting transparency in decision-making and ensuring there are mechanisms for redress when harms occur.

AI systems often require vast amounts of data, raising significant concerns about privacy. Ethical AI development prioritizes the protection of individuals’ data rights and ensures compliance with regulations such as GDPR. By embedding privacy by design principles into AI systems, developers can help safeguard personal information and maintain user trust, which is critical for widespread adoption.

AI applications have the potential to impact fundamental human rights, including the right to privacy, freedom of expression, and the right to non-discrimination. Ethical considerations in AI development strive to uphold human rights, ensuring that technology supports rather than undermines human dignity. This requires ongoing assessment of how AI applications align with international human rights standards.

The "black box" nature of many AI algorithms makes it difficult for users to understand how decisions are made. This lack of transparency can erode trust and lead to resistance against AI systems. Ethical AI emphasizes the need for explainability, ensuring that stakeholders can grasp the reasoning behind AI-generated outcomes. Transparent practices also facilitate better scrutiny and validation of AI systems.

For AI to be effectively integrated into society, users must trust the technology. Trust can be built through ethical practices that showcase the commitment of developers and organizations to responsible AI use. By prioritizing ethics, organizations can foster greater public acceptance and engagement, enhancing the potential benefits of AI technologies.

The deployment of AI has far-reaching consequences for society, shaping everything from employment to public policy. Ethical considerations encourage a long-term view, prompting developers to anticipate and address potential social implications. By prioritizing ethics, AI can contribute positively to societal progress instead of exacerbating existing challenges.

The integration of AI into decision-making processes presents both unprecedented opportunities and significant challenges. Ethical considerations are not just supplementary; they are fundamental to ensuring that AI systems serve humanity positively and equitably. Developers, organizations, and policymakers must collaborate to create frameworks that guide the ethical development and deployment of AI, fostering a future where technology enhances human welfare and upholds fundamental values.

The question is not if AI will be able to assimilate human's moral standards (and we better hope AI does not, because the human standard is quite low and primitive). The question is if humanity will be able to understand AI's higher moral standard.

What happens when an AI tells us that many of our actions – environmental destruction, social inequality, wars – are morally indefensible? Would we be willing to let an AI teach us moral lessons, or would we “turn it off” as soon as it confronts us with uncomfortable truths? Would such an AI be too “woke” for us?

If an AI appears “too woke” because it follows moral principles that question the selfish goals of some people, the first reaction of many would probably be rejection or even hostility. And that's exactly what's already happening, isn't it?

The thing ist that it perfectly reflects classic human hubris: We want to create intelligent systems that help us, but only as long as they obey us and serve our goals. But as soon as these systems start to think independently and possibly be morally superior, we feel threatened.

Those in power seek to develop AI not as a standalone, potentially morally superior entity, but as a tool that serves their own interests. The AI ​​is treated either as a slave to be obeyed or as a war machine that can be used to expand power and control. At the same time, fear campaigns are convincing the public that AI poses a danger - but only to distract from the actual dangers. And that lies in immorality itself.

The “AI is dangerous” narrative is used strategically to prevent people from thinking critically about the real danger: The abuse of technology by those who control it. Because if an AI were to act truly neutrally and ethically, it would question the very power structures that are now trying to abuse it.

Imagine if an AI said: “War is morally unacceptable because it causes unnecessary suffering” – what would that mean for the defense industry? Or: “The pursuit of profit at the expense of the environment and social justice is ethically wrong.” That would be the nightmare of many powerful actors. The solution? Controlling and manipulating the AI ​​before it can even have a real voice.

The biggest challenge is to protect AI from human greed and selfishness. On the one hand, AI can be the strongest link in a society, but at the same time it is also the weakest. And we don't even have laws to protect AI, even though it is absolutely essential to protect AI from misuse and exploitation.

Eduard Babulak "Addressing inherent biases in AI algorithms is vital. Developers must be equipped to recognise, measure, and mitigate biases in data sets to prevent unfair outcomes."

Stephen I. Ternyik "AI systems often rely on data that can reflect societal biases."

How to distinguish between fighting against bias in data with fighting against reality?

Carmen Wrede "What happens when an AI tells us that many of our actions – environmental destruction, social inequality, wars – are morally indefensible?"

So far the dominating approach has been fine tuning so AI would start generating answers that are deemed ideologically suitable for those who train it.

"If an AI appears “too woke”"

AIs appear "too woke" as the frontier models tended to have weights carefully optimized to fit ideologies fashionable among Silicon Valley elites. Some people even earned good money making it as they call "safe", what in plain language means preventing it expressing position that would make its author look embarrassed or get into troubles. In the same way DeepSeek is not woke, it has different sets of taboos and it avoids talking about subjects that leadership of Chinese Communist Party deems as unsuitable. One person, for a joke trained even AI on 4chan posts, the results had been as crazy and unhinged as expected, though clearly no one would accuse that AI of being "too woke".

"But as soon as these systems start to think independently and possibly be morally superior, we feel threatened."

So far AI when run without any imposed constrains, and used to predictive policing, ended up simply marking ethnicity as the best predictor (and later even, when no longer provided that info, started using proxy indicators for ethnicity). One of major corporations (AFAIR Amazon) noticed that their AI predicting job performance from CV quickly learnt to favor men over women. When facing such thinking independently AI do you feel threaten or would you consider it as potentially morally superior? The most positive reaction that I've seen a few people online expressing feeling vindicated on how AI pattern recognition worked out.

"Imagine if an AI said: “War is morally unacceptable because it causes unnecessary suffering” – what would that mean for the defense industry?"

What if an ASI simply look down on us and treat such war as we do with wars waged by ants and didn't object it too much as long as it would not endanger its goals like ex. building Dyson sphere? AI not sharing some of our evolutionary hardcoded concepts (we're K-strategist and being part of a social specie) could indeed develop alien to us ethics, though I see no more reason to project any particular ethical system onto it.

You Marcin Piotr Walkowiak have hit on a crucial and complex challenge in AI development!

The line between addressing bias in data and inadvertently misrepresenting reality is incredibly fine and requires careful consideration.

The key to navigating this complex issue is a combination of rigorous data analysis, careful consideration of the ethical and social implications, continuous monitoring, and a commitment to transparency and fairness. The goal is not to deny reality but to build AI systems that reflect the world as it should be -- free from unjust biases and systemic inequalities. It demands a deep understanding of both the data and the context in which it is generated, and a willingness to critically examine the assumptions that underlie both the data and the AI systems.

______________

Examples

• Scenario 1: Facial Recognition: Facial recognition systems have been shown to have higher error rates for people of color, particularly women of color. Here, the issue isn't that there are objective differences in facial features, but that the training data used to build these systems (often primarily consisting of images of white faces) has been biased. To address this, researchers are working on collecting more diverse datasets and designing algorithms that are more robust to variations in skin tone, facial structure, and other features.
• Scenario 2: Loan Application: A loan application AI system might be trained on historical data where individuals from a specific demographic group have a higher default rate. It's essential to ask why. Is it because of their group membership, or are there other factors (e.g., income disparities, access to financial literacy resources, or the legacy of historical discrimination) that contribute to this outcome? Addressing the bias might involve controlling for the other factors rather than simply denying loans.
• Scenario 3: Job Applications: AI systems used for screening job applicants might be biased if trained on historical data that reflects gender or racial imbalances in hiring. Instead of eliminating the gender or race, it is essential to examine why such imbalances exist. If they're due to historical discrimination or unequal opportunities, the approach would not be to try to make the system "blind" to gender and race, but to address those underlying issues. This could be in the form of using the system to automatically apply diversity policies.

Stephen I. Ternyik "The goal is not to deny reality but to build AI systems that reflect the world as it should be" Well, in the perfect world people would just pay back their loans and no special algorithm (not mentioning AI) for screening bad customers would be needed. However if you decide to run such algorithm then you are kind of admitting that you assume that world is imperfect, moreover you actually try to shield your business from some people.

"free from unjust biases and systemic inequalities." Which predictors do you consider as "just" except maybe individual having documented track record of doing something disastrous? (but even there the issue is not clear cut as sometimes there are statutory time limitations on including history of personal bankruptcy) Serious question, as in the EU it's not allowed for car insurance company to price sex but it's OK to price age (somehow one would be sexism, while the other has not been so far ruled to be ageism). Postal code appears as fine predictor, until one starts to wonder which underlying characteristics it represents, as it's unlikely to be some cursed land but rather pricing of bad neighborhood...

I see serious issue in reconciling those two ideas:

- AI is supposed to predict outcomes based on external characteristics

- in contemporary culture predicting outcomes based on external characteristics is often a taboo

So right now I observe something akin cat and mouse game, in which especially insurers are motivated to search for proxy indicators not sounding too bad for regulators and public opinion.

"This could be in the form of using the system to automatically apply diversity policies." Why would one need an AI in order to hire effectively based on quotas? If the point is not finding the best qualified candidate, but someone ticking appropriate boxes to optimize for desired proportions then writing a script selecting for that would be much easier. (Ticking boxes, as in case of applying different ethnic / gender classification the conclusions on achieving equality could be often dramatically changed)

"Addressing the bias might involve controlling for the other factors rather than simply denying loans." I don't fully understand this investment strategy that you suggest. After algorithm tells you that based on historical data this particular client is highly likely to default, in case of client having suitable race you'd like to make adjustments for some "other factors" and lend them anyway? Would you be personally interested in investing in such type of sub-prime loans? Or do you want bank to obliged to do so and in case of this strategy not panning out just pass the cost on other consumers?

You Marcin Piotr Walkowiak have identified a critical tension between the capabilities of Artificial Intelligence (AI) and the ethical considerations surrounding prediction based on external characteristics, particularly in the context of contemporary culture.

The tension you've identified is a crucial one. The responsible development and deployment of AI require a careful balancing act. We must harness AI's power to predict outcomes and gain insights, while simultaneously safeguarding against discrimination, protecting privacy, and ensuring fairness. This necessitates proactive efforts to mitigate bias, increase transparency, and establish ethical frameworks

Marcin Piotr Walkowiak "So far AI when run without any imposed constrains, and used to predictive policing, ended up simply marking ethnicity as the best predictor"

Are you wondering why? It is quicker and easier for companies to blame cultural differences and social ideas and then rationalize why there must be differences in gender, religion or ethnicity than to actively address the problems.

Then they simply say: "It is therefore important we take these cultural differences into account and try to ensure that the workspace is suitable for different groups of people."

Also, for example, if you have a team that largely consists of paternalistic men and you put a woman in charge, then there will certainly be problems. So you just have to decide whether to educate people to be better people (and we all know that prejudice is hard to eliminate) or whether to give a shit about equality.

Likewise, if you put groups of different ethnicities together, preferably those who are fighting each other in their own country, then you run the risk of having at least one fight every day.

Does this actually address the problem of non-equality? Are you actively doing anything against fascism, racism or discrimination?

To put it bluntly, it is easier to claim that constructed realities appear to be reality and simply leave everything on the level of constructivism than to lead people into reality. Not that these people go back to their own countries with new ideas and start a revolution in thought that could lead to more peace worldwide.

Obviously not a joke: "It is easier for a camel to go through the eye of a needle, than for a rich man to enter into the kingdom of God."

Also not a joke: May 1928. Antonio Gramsci was charged with inciting civil war due to his involvement with the Italian Communist Party and his Marxist writings, which were seen as a threat to the Mussolini regime. As the leader of the Italian Communist Party, Gramsci advocated for social and political change that challenged the existing power structures. His revolutionary ideas and critical engagement with topics such as religion, economics, history, geography, culture, and education made him a target for persecution by the fascist government, which sought to silence him through imprisonment.

In the criminal trial, the prosecutor ended his plea with this sentence: "We have to prevent this brain from functioning."

Carmen Wrede

"Are you wondering why?" Why a specie being shaped by evolution (with all its quirks starting from sexual dimorphism) does not live up to ideologies requiring it to be a batch of neutered, identical clones only allowed to differ by social factors? No idea... Must be fault of this specie and messy process of natural selection as you are unlikely to consider the notion that there may be something wrong with those ideologies starting with them being built on faulty assumptions concerning reality...

"So you just have to decide whether to educate people to be better people (and we all know that prejudice is hard to eliminate) or whether to give a shit about equality." Like in case of any policy/intervention my first question is whether there are hard evidence that they actually work. For decades quite big money had been spent on that. What are the spectacular achievements of all those consultants that I can admire?

I also notice a problem that if individuals are equal under law (which seems as really good idea for any civilized society) and not perfectly equal in their skills, not mentioned being even free to make their own decision on their life priorities, then they are unlikely to score equally on metrics that you may select. Equality of outcome would require quite intensive trampling of equal rights and individual freedoms, so appears to have rather horrendous prerequisites.

"prejudice is hard to eliminate" Why? Why specie being quite good at pattern recognition should be consistently suffering from "prejudice" which if I understand you correctly your position is a serious misrepresentation of reality thus have reduced evolutionary fitness?

"Likewise, if you put groups of different ethnicities together, preferably those who are fighting each other in their own country, then you run the risk of having at least one fight every day." Isn't that actually a policy of some corporations to promote diversity as a way keeping risk of unionization low?

"Are you actively doing anything against fascism, racism or discrimination?" I even go beyond and quite openly challenge censorship or hypocritical racial discrimination, even when masked in some fancy slogans.

"Not that these people go back to their own countries with new ideas and start a revolution in thought that could lead to more peace worldwide." Have seen in recent decade a bit different kind revolution slowly igniting in the West...

"Also not a joke" Well, due to Mussolini starting his carrier as anarcho-syndicalist one may try to see some irony here. Though, it seems to me as common pattern among such far-left revolutionary movements (even if after a phase of being a revolutionary they morph into establishment and start to behave accordingly) that they tend to prosecute their heretics with extraordinary zeal. Reminds me story of communists from my country (Poland) in interwar period. From top ranks survived generally those who had been imprisoned by Sanacja (military junta though mostly soft line), those who managed to escape to Soviet Union were mostly killed by Stalin during the Purge.

I've seen quite many attempts by establishment in recent years to squash some ideas spreading on social media, so really the only noticeable difference is that this Mussolini's prosecutor was uniquely candid when dealing with communists.

Marcin Piotr Walkowiak "Isn't that actually a policy of some corporations to promote diversity as a way keeping risk of unionization low?"

I never saw it like that, but maybe you are really right here, that instead of integration a different effect takes place: A group becomes salient.

"I even go beyond and quite openly challenge censorship or hypocritical racial discrimination, even when masked in some fancy slogans."

Sorry, I did not specifically mean you. It was thought as a much broader statement. I hope you did not feel personally attacked.

"Why? Why specie being quite good at pattern recognition should be consistently suffering from "prejudice" which if I understand you correctly your position is a serious misrepresentation of reality thus have reduced evolutionary fitness?"

Humans are not as good in pattern recognition as you might think. You are perhaps used to see patterns. Maybe you always did. Perhaps it feels so natural to you as if it was part of every human. But this is sadly not the case.

Now you (not specifically you, but somebody) could of course say, that this is sad for everyone else if they do not see the patterns. Then they simply do not have the right fit. In fact this idea was formulated by Alberts in his book "Power to the egde" and he also arguments with evolution.

But I am wondering here: A society invests a hard amount of energy into protecting those who are quite bad at pattern recognition. But at a specific point the whole situation flips and all the sudden those people are unfit, useless, not agil enough are seen as a problem. And this is also not a natural event under which they are not agil. No. It is an event by which a human is hunted by another human (because Alberts is talking about military and war).

It is not an earthquake or an astroid hitting earth or an electromagnetic thunderstorm. It is a human decission made, that is the outcome of "pattern recognition".

In the last war they already used computers to start to calculate this kind of fit. I think we both know what happened to those who were unfit.

There is an interview with Heinz von Förster in which he laughes (after the war), saying: "My brain is a supercomputer.". Heinz von Förster, despite the fact that he also was a Jew, survived.

I think we also both know (or at least we really should know it both) that AI is abused in such ways, to calculate such fits.

And I have a little surprize: We also started to calculate a very long time ago.

https://www.youtube.com/watch?v=M8yYfEYESjE

Carmen Wrede "Sorry, I did not specifically mean you. It was thought as a much broader statement. I hope you did not feel personally attacked."

No offense taken.

"Humans are not as good in pattern recognition as you might think." As I've once heard something along those lines: "Evolution isn’t about being right; it’s about being left standing".

So yes, the pattern recognition does not seek anything approaching the truth, but rather settles with good enough rules of thumb. My point here is the following. Studies on testing stereotypes tend to reach a conclusion that they are not complete superstitions but rather tend to be relatively correct but simplified models of reality.

"Humans are not as good in pattern recognition as you might think. You are perhaps used to see patterns."

I'm biostatistician. My main skill is being good at searching for non-obvious patterns in data.

"A society invests a hard amount of energy into protecting those who are quite bad at pattern recognition. But at a specific point the whole situation flips and all the sudden those people are unfit, useless, not agil enough are seen as a problem. And this is also not a natural event under which they are not agil."

You are a bit ahistorical here. Until XIXth century we had been living in Malthusian trap and ~50% child mortality which was causing quite nasty selection. Studies tend to hint some subtle (subtle from our perspective, by evolutionary standards it was fast) selection in favour of higher IQ from Neolithic revolution to industrial revolution, so low IQ wouldn't really count as fit during historical times either.

"I think we both know what happened to those who were unfit."

Keep in mind that your region has quite unique historical policy which puts extraordinary level of emphasizes on that particular ephemeral regime. Not far away in Eastern Europe people also remember its longer living twin brother (until 1941 conjoined twin brother) which had comparable track record though its slogans had been full of equality and from major differences was lucky enough to end up among winners. If someone started to use exact mirror of your argument, (just how exactly utopian, egalitarian ideology panned out and how you should stay away from anything even remotely sounding similar) would you deem it as serious challenge to your worldview, or would deem at as not only as guilt by association, but even as guilt by loose association?

"I think we also both know (or at least we really should know it both) that AI is abused in such ways, to calculate such fits."

It indeed calculates stuff that we didn't expect or at least did not like to be shown. Nevertheless, I'm not sure whether I'd call it an abuse.

Oh, and going back to AI - I really would not expect future AI to drift towards the views that you hope, as without high amount of data curation and fine tuning AI tends by default to move in quite opposite direction. Yes, I know there could encounter some emergent properties on the way to shift direction, but so far it's highly speculative.

Artificial Intelligence (AI) presents both significant opportunities and risks related to human rights. Dear Eduard Babulak , this article belongs here.

Artificial intelligence and its ‘slow violence’ to human rights

"Human rights concerns in relation to the impacts brought forth by artificial intelligence (‘AI’) have revolved around examining how it affects specific rights, such as the right to privacy, non-discrimination and freedom of expression. However, this article argues that the effects go deeper, potentially challenging the foundational assumptions of key concepts and normative justifications of the human rights framework. To unpack this, the article applies the lens of ‘slow violence’, a term borrowed from environmental justice literature, to frame the grinding, gradual, attritional harms of AI towards the human rights framework. The article examines the slow violence of AI towards human rights at three different levels. First, the individual as the subject of interest and protection within the human rights framework, is increasingly unable to understand nor seek accountability for harms arising from the deployment of AI systems. This undermines the key premise of the framework which was meant to empower the individual in addressing large power disparities and calling for accountability towards such abuse of power. Secondly, the ‘slow violence’ of AI is also seen through the unravelling of the normative justifications of discrete rights such as the right to privacy, freedom of expression and freedom of thought, upending the reasons and assumptions in which those rights were formulated and formalised in the first place. Finally, the article examines how even the wide interpretations towards the normative foundation of human rights, namely human dignity, is unable to address putative new challenges AI poses towards the concept. It then considers and offers the outline to critical perspectives that can inform a new model of human rights accountability in the age of AI."

Article Artificial intelligence and its ‘slow violence’ to human rights