Salman Khan, in Brave New Words; How AI Will Revolutionize Education (and Why That’s a Good Thing), looks forward to many ways that AI might enhance learning. AI can act as a Socratic tutor by discerning strengths and gaps in a student’s responses and asking customized questions that can nudge her forward. It can emulate historical and literary figures and engage her in a debate. She can chat with Benjamin Franklin, Abraham Lincoln, Sojourner Truth, or Mark Twain. Khan says that she can design experiments with a simulation of Marie Curie or co-write a Federalist Paper with a simulated James Madison or Alexander Hamilton. She can discuss the kinematic equations with Isaac Newton and natural selection with Charles Darwin. Khan also says that generative AI can help people learn to play musical instruments, suggesting practice routines and fingering techniques. He also wrote that AI can integrate fields by showing how math works with art, how writing works with science, and how history works with economics.
But AI has not just one bias, but many which can reinforce each other and thus be very difficult to overcome, including:
- The overall perspective of humanity–who we are, what is best for us, what our goals are, what a meaningful human life is, and our highest potentials for well-being. A lot of AI researchers are concerned about the alignment problem—aligning AI’s goals with humanity’s. But AI-based systems have narrow views of the diversity of human cultures and our possibilities for synthesizing them into larger perspectives of ourselves. The content in the large language models (LLMs) they use overrepresents Western cultures and includes relatively little content about the riches in thousands of others, including Ohlone.
- How we use the senses and how cultures emphasize them. The anthropologist Constance Classen has studied a large variety of ways in which cultures process the senses. The Ongee in the Andaman Islands, in the Bay of Bengal, have ordered their world by smell. They have felt that aroma is a basic vital force in the universe and a key aspect of the self. People try to maintain equilibrium between the odors in their environment and between people. Some indigenous societies in the mountains of peninsular Malaysia have words for more types of aromas than English speakers do. Life in jungles with profuse growth that blocks vistas of distant places has encouraged sensitivity to scents from the wide variety of plant forms, including their stages of freshness and decay. This environment contrasts dramatically with the clear views that ancient Greeks had of the sea, coasts, and mountains. The Tzotzil in southern Mexico have felt that heat is the most basic force in the cosmos so that each thing and person contains a certain amount of heat energy. The sun is its main source. The most important people in society are associated with the hot rising sun, and the least significant with its more tepid state as it sets. Its heat thus orders social relations as odors arrange them for the Ongee. The Desana around the Amazon in Colombia have believed that color is the main vital force in the universe; the sun generates life by mixing and matching colors’ energies. Yellow is associated with male virility, red with female fertility, green with vegetal growth, and blue with transitions. The cosmos consists of different layers of colors, with yellow on top (the sun’s creative light), blue next (the transitional zone between the star and the earth), red on the earth’s surface, and green underneath. People, animals, and plants receive their color energies at birth, which return to the sun when they pass away. According to Classen and David Howes, the Dogon in Mali closely associate speech and smell, believing that speech has material properties that include both sound and odor. Both originate in vibrations and thus have a common source. The Dogon sometimes speak of “hearing a smell” and classify words by odor. Good words smell “sweet” and bad words smell “rotten.” People thus process senses according to rich associations of ideas, which reflect their cultural landscapes.
- Perception. People in different cultures perceive their surroundings in diverse ways. A previous article here explored profound differences between Caucasian-American and Chinese perspectives.
- The emotions we feel. There are deep cultural differences between them as well.
- Categories and taxonomies. Today’s blends of cultures make monolithic identities outdated. Most people in the U.S. don’t fit into all-inclusive categories.
- Word embeddings. Large language models that AI uses give the most weight to words with similar meanings, but many meanings are culturally constructed. Westerners usually classify a banana and a strawberry as closer than either is to a chimpanzee. Chinese more often consider a chimp and banana closer because they often think in terms of relationships and holistic contexts rather than distinct objects. A chimp eating a banana is a tangible story.
- What we pay attention to. What we find interesting can continuously expand as we explore more cultures.
- Rewards. A lot of AI research has focused on dopamine rushes, but there are several other neurotransmitters. Serotonin is not sudden and orgasmic like dopamine often is, but more lasting, subtle, and conductive to contemplation and feelings of connectedness with others. Cultural explorations are more oriented to this kind of pleasure. It doesn’t well up all at once; it gets better and better with time.
- Models. David Christian, in The Alignment Problem; Machine Learning and Human Values, wrote that we are in danger of losing our place in the world to models and seeing the world as a formalism. In other words, focusing on the map instead of reality.
- The concepts we treat as most basic.
These biases can strengthen each other. They can be used to train new AI-based systems, and that will further reinforce them. They don’t fully express most of the world’s cultures, including the sensual, natural, and spiritual wealth in Laotian traditions (below).
Joy Buolamwini, in Unmasking AI; My Mission to Protect What Is Human in a World of Machines, wrote that commercial face-classification algorithms are much more accurate at identifying white males than black females. They were trained on datasets of online photos, and since white males dominate the news, a disproportionate number of pictures have been of white men. She wrote that “algorithmic injustice” can be entrenched in employment, housing, education, and criminal justice.
Ethan Mollick, in Co-Intelligence; Living and Working with AI, wrote that a popular text-to-image AI model called Stable Diffusion amplifies stereotypes about race and gender, depicting higher paying professions as whiter and more male than they really are. CEOs and judges are disproportionately white and male, and 70% of fast-food workers had darker skin tones in the models even though 70% of American fast-food workers are white. Buolamwini’s work has deservingly gotten a lot of attention; she was able to meet with President Biden. But the visual bias she details only scratches the surface. Even if it’s corrected, AI will still have a long way to go before it can convey the subtleties and variety of African cultures.
AI also has a long way to go before it can represent the variety of cultures with white majorities. Ben Shneiderman, in Human-Centered AI, noted that the AI community is oriented to rationalism, but many European cultures favor emotion. Slavic cultures value free expression of feelings and find Anglo cultures cold and mechanical. A Russian told me that she finds them “constipated.” The gentlemen below were warming up a March day in Prague, a great party city.
A man from Italy once told me, “Anglo-Saxon culture is the worst culture in the world” and said that many of its people don’t like to be touched. Since Italians are known for being tactile, I can imagine how he would have found that jarring. Being of Irish descent and having traveled in Ireland and savored its music, literature, and speech, I can attest that its people have often preferred imagination and expressions of emotion over stone-cold logic. People in all these cultures have sometimes characterized English people as excessively rational and private, but the English are diverse too. A lot of Northerners have characterized Londoners in that way, seeing themselves as more focused on personal relationships than business. But this is a crude stereotype, since London is one of the world’s most diverse cities.
People have typically experienced reality through their culture’s prism and this can now be increasingly designed by AI. The AI industry has been consolidating into a small number of corporations, including OpenAI, which builds large language models and has created ChatGPT, and Nvidia, which provides hardware platforms for AI. AI requires huge data centers, which are extremely costly. Yuval Harari wrote that we need to commit ourselves to building institutions with strong self-correcting mechanisms, but several of the most powerful institutions in the industry are highly concerned about maximizing profits. Tristan Harris, co-founder of the Center for Humane Technology, says that they’re in a race to rollout, trying to roll out the next technology as fast as possible to achieve market dominance. Understanding the intricacies of African, Slavic, Native American, Southeast Asian, Tibetan, and all other cultures seems far down on their lists.
We can constantly look At, With, and Beyond our AI systems and our uses of them. Keep asking AI out-of-the-box prompts so that it compares multiple perspectives. Instead of passively accepting what an AI system initially says, ask it more questions. You can give it many creative prompts, including:
- That’s not deep enough. Please tell me more.
- Can you give me an alternative perspective?
- What are the best books, articles, and educational videos about (the topic in question)?
- Please compare X with Y (e.g., Preah Ko with the ancient Cham culture).
- Please compare it with Chinese (or Indian, or ancient Greek, or Yoruba) culture.
- What would another writer say about this? Please be Achebe. Now be Pablo Neruda. Now be Lady Murasaki. Now be a Zen koan writer. Now be a Rigvedic priest. An Amazonian shaman. A Yoruba Babalawo.
- Your answer is biased towards vision, but many cultures emphasize other senses as well. What would an African who focuses on sound, motion, and rhythm think?
- You’re assuming that distinct objects and static categories are most fundamental in reality. Please come up with a different perspective, which is based on other cultures’ assumptions. Then question those assumptions and come up with yet another perspective.
- Please list possible analogies, critique your list, and add three more analogies.
- Can you give me a higher perspective, which integrates more cultures?
- What are different cultures’ ideas of paradise? Heaven? Happiness? Can you come up with new ideas?
- There are countless cultural fusions in the world, including Indian and African, and Chinese and Islamic. What insights can you provide about them?
- Please critique your answer. What are its shortcomings? What are its biases? How can it be improved?
Then step back from AI and read a book about another culture, take a nature walk, meditate, dance, or discover a style of music you’ve never heard before. Expand your inquiry in as many directions as possible beyond the current data set. Instead of being passive recipients of content AI gives, we can engage with it in romantic dialogs that can uncover ever more facets of the topic and connections with other topics.
Yuval Harari writes about humans no longer dominating the world but instead relying on data and algorithms to make their decisions. He worries that after all humanist illusions are destroyed, people will only want to merge into flows of massive amounts of data. But each culture has depths that AI still cannot fathom. Each converges from many types of experiences that are unique to it, including language, the natural environment, historical patterns, art, perception, emotional patterns, and general ideas about the world. Mark Coeckelbergh, in AI Ethics, writes that the ways people think have many sides. In addition to data analysis and abstract patterns, it’s based on embodied, interpersonal, emotional, artistic, and environmental experiences. All converge in unique ways in each culture.
By looking At, With, and Beyond an increasing number of cultures, we can ultimately enter any at will, then zoom out and enter others, and then synthesize higher perspectives of our world. We can explore one way of apprehending reality after another with the ease of an eagle in flight. My books show how high we can fly, and the articles here give a lot of glimpses. We can thereby retain our human identity in many ways, including imagination, out-of-the-box thinking, deft shifts in perspective, dreams of what’s possible, empathy with all cultures, and ever-expanding love.