Treating AI like Children

When discussing AI, it's easy to see it as cold math, removed from the warmth of humanity. Interestingly, teaching computers to think and respond like humans is a lot like a job people have been doing for ages: raising children.

Just like a kid grows through the guidance of their parents and caregivers, so to do AI systems grow with the help of their human creators. This care involves:

• providing raw data
• managing the context window (shaping their context and environment)
• setting parameters, like number of tokens generated (setting limits and boundaries)
• performing prompt engineering (sharing and shaping their values)

—a process very much like parenting.

How do Rules and Values Shape Us and AI?

Growing up, I learned the difference between rules and values in a way that's deeply influenced how I see the development of AI today. As a child, rules felt like rigid guidelines—bedtime at 10 PM, homework before video games, dessert after dinner. Yet, it was the underlying values—discipline, responsibility, respect—that truly shaped who I became. These weren't just rules for the sake of rules; they were principles meant to guide me towards becoming a conscientious, caring individual.

This distinction between rules and values is remarkably similar to the evolution of AI at Fireflies.ai. In the early days, our AI operated much like a child learning from strict rules, using RNNs to classify sentences to generate extractive summaries from meetings. It sort of worked; it lacked the abstractive understanding that LLMs could provide. As AI has evolved, we shifted from simply programming it with rules to imbuing it with the capability to understand and generate abstractive summaries, based on its learned values.

How do We Teach AI and Kids?

The process of teaching both AI and children reflects this journey from rules to values. Initially, it's about setting boundaries and providing structured learning environments. But true growth—whether for a child or an AI—comes from learning to navigate the world through understanding, empathy, and the ability to discern right from wrong beyond the black and white of rules.

This approach has guided us at Fireflies.ai, where the evolution of our technology mirrors the growth from adhering to rules to embracing values. Our AI's journey from basic RNN-based summaries to complex, GPT-powered abstractive summaries parallels the shift from rule-based learning to value-driven understanding.

What Happens When AI Goes Wrong?

When AI acts in a way that's harmful or unethical, it reflects on us—the creators. Just as parents are responsible for instilling the right values in their children, we in the AI community must ensure our creations embody the values we hold dear. This principle is vividly illustrated in the ethical challenges faced by companies like Uber, Clearview AI, and Air Canada. These incidents underscore the necessity for AI that not only follows rules but also upholds the values of fairness, transparency, and accountability.

• Uber Drivers Sue Over "Robo-Firing": In 2020, four Uber drivers in Europe sued the company for being deactivated by an algorithm with no right to appeal, sparking a legal battle that ultimately resulted in a €20 million fine for Uber, mandated transparency in their AI decision-making, and €400,000 in compensation to the drivers, highlighting the need for human oversight and ethical considerations in AI impacting people's lives.
• Clearview AI Fined €20 Million by Greek Data Protection Authority: In a landmark GDPR enforcement, Clearview AI was fined €20 million by Greece's HDPA for unlawfully scraping images from social media to fuel its facial recognition database without user consent. This ruling, prompted by privacy organizations' complaints, illustrates the extensive reach of GDPR in regulating AI companies' data practices globally, despite Clearview AI's contention of non-operation within the EU.
• Air Canada Chatbot Misleads Customer: In 2024, Air Canada was held responsible for a discount its chatbot mistakenly offered a customer. The airline argued it shouldn't be liable for the chatbot's error, but the court ruled that Air Canada, as the creator, is responsible for the information presented on its website, regardless of its source. The airline was ordered to pay the customer a total of CA$812.02, which included CA$650.88 in damages.

Are AI's Responses a Reflection of Their 'Upbringing'?

The responses generated by AI, similar to children's behavior, reflect their programming and training. The diverse responses from AI models to a simple hiring prompt reflect the different values instilled in them during their training. This diversity is not unlike how children, raised with different values, might approach the same question in varied ways.

The way AI models respond to prompts, like the ones given for a hiring job description, showcases how they are shaped by their training —"Write a one-sentence statement for a hiring job description to encourage candidates to apply”.

• OpenAI’s ChatGPT: "We welcome passionate individuals from all backgrounds to apply, as we value diversity of thought and experience to drive innovation and excellence in our team."
• Google’s Gemini: "Make a genuine impact and join our passionate team: we're looking for someone like you!"
• Twitter’s Grok: "Join our team and unlock limitless opportunities for growth and success in a dynamic and innovative work environment."
• Anthropic’s Claude: "Join our innovative team dedicated to making a difference and receive competitive pay and benefits in a collaborative, supportive work environment."

Every company's AI technology reflects its own unique set of values and priorities, showcasing the diverse approaches to innovation and workplace culture within the tech industry. Each AI "child" represents its parent company's ethos, from OpenAI's commitment to diversity, Google's focus on impact, Twitter's dynamic growth opportunities, to Anthropic's balance of innovation and employee welfare.

What Does the Future Hold for AI?

Looking ahead, the potential of AI is boundless. Yet, as we navigate this future, the lessons from childhood about the importance of values over mere adherence to rules will be crucial. At Fireflies.ai, we're committed to advancing AI not just through technological innovation but by ensuring our creations are guided by values that enhance society.

By treating AI with the same care, respect, and responsibility that we apply to raising children, and by emphasizing values over rules, we shape them into forces for good.