How Do We Define “Good” Algorithms?

Artificial intelligence is transforming our world at an astonishing pace, but it also brings many profound ethical questions. From the decisions made by self-driving cars to content recommendations on social media, every action of AI can have a profound impact on human society. So, how do we ensure that these algorithms are “good”? Or, more fundamentally, what does it mean for an algorithm to be “good”?

A few years ago, I read a news story about a self-driving car facing an emergency situation—it had to choose between hitting a pedestrian or protecting the passengers inside. This question left me deep in thought. If a human driver faced such a dilemma, their decision might be based on intuition, emotion, or moral judgment. But what about AI? Its choice depends entirely on the design of its algorithm. This realization made me understand that the ethical issues surrounding AI are not just technical challenges—they are philosophical reflections.

1. The “Values” of Algorithms: Who Decides?

AI’s behavior is driven by algorithms, and the designers of these algorithms often determine their “values.” For example:

• A company’s hiring algorithm may prioritize candidates with certain backgrounds because it was trained on historical data.
• Social media platforms’ recommendation algorithms may amplify extreme content because they find it more engaging for users.

These examples show that AI is not a neutral tool—it carries the intentions of its creators and reflects societal biases. The key question then becomes: Who has the authority to decide the values embedded in these algorithms? Is it the developers, corporations, or the public? And if it’s the latter, how do we involve ordinary users in the design and oversight of algorithms?

2. Bias and Fairness: How to Avoid “Bad” Algorithms?

Bias is one of the thorniest issues in AI ethics. Since AI models are typically trained on large datasets, and these datasets may themselves be biased, algorithms can easily replicate or even amplify these problems. For instance:

• In healthcare, some AI diagnostic systems perform poorly for minority patients because the training data lacks diversity.
• In the judicial system, crime prediction algorithms may disproportionately target low-income communities due to reliance on past arrest records.

To avoid “bad” algorithms, we need to take the following steps:

• Transparency : Make the workings of algorithms public and subject them to public scrutiny.
• Diversity : Ensure training data includes people from different genders, races, and cultural backgrounds.
• Accountability Mechanisms : Establish clear responsibility systems so that when algorithms fail, specific individuals can be held accountable.

3. Moral Dilemmas: Can AI Make the “Right” Choice?

Returning to the earlier example of the self-driving car, can AI make the “right” choice? In fact, there is no simple answer to this question. Different cultures and philosophical schools define “right” differently. For example:

• Utilitarians might argue for choosing the option that minimizes harm.
• Moral absolutists might insist that certain actions (such as harming innocents) are never acceptable.

This complexity makes the ethical design of AI incredibly challenging. We cannot write perfect rules for every scenario, but we can minimize risks through the following approaches:

• Contextual Design : Adjust algorithm priorities based on the needs of different scenarios.
• Multi-Stakeholder Involvement : Invite ethicists, legal experts, and the general public to discuss the moral framework of AI, ensuring that algorithm designs reflect diverse values.

4. Public Trust: How to Win Hearts and Minds?

No matter how advanced AI technology becomes, if the public loses trust in it, it will never truly integrate into society. In recent years, some tech giants have faced widespread criticism for misusing AI technology. For example:

• A social media platform was accused of using algorithms to spread misinformation, influencing election results.
• An e-commerce platform’s pricing algorithm was found to engage in price discrimination, leading to consumer dissatisfaction.

These incidents remind us that the development of AI must prioritize public interest. Only when people believe that AI serves humanity rather than manipulates it can it gain true acceptance. To achieve this, we need to take the following measures:

• Education and Awareness : Through outreach programs and educational resources, help the public understand how AI works and its potential societal impacts.
• Open Dialogue : Encourage participation from all sectors of society in discussions about AI ethics. For instance, host public forums or online discussions where people from diverse backgrounds can share their perspectives.
• Policy Safeguards : Governments should establish reasonable regulations to ensure that AI development aligns with societal values, while holding violators accountable for unethical practices.

5. Future Possibilities: From “Good” Algorithms to a “Good” Society

If we can successfully define and implement “good” algorithms, AI will become more than just a tool—it will be a force for social progress. For example:

• In healthcare, AI can assist doctors in making more accurate diagnoses while avoiding racial or gender biases.
• In the judicial system, AI can help judges make fairer decisions, reducing human errors that lead to injustice.
• In education, AI can provide personalized learning plans for each student, narrowing the gap in educational resources.

However, all of this hinges on embedding ethical principles into the design and application of AI. As philosopher Immanuel Kant once said, “Humans should be treated as ends, not means.” We must ensure that AI always serves human interests, rather than becoming a new instrument of power.