An Update on the Ethical Landscape of Generative AI

When the modern AI era was first spooling up, we heard a lot of talk from the landed gent— er, business elites about developing new ways of living to reorient society toward a future with less work to do. Perhaps we could put in fewer hours per week, be supported by universal income, and generally let machines take over the boring stuff to give us space to do more special human things. 

Since then, tech culture has leaned toward, depending on your ideological sympathies, a rather authoritarian or um, angry-fatherly point of view:

Now, we need to get back into the office in person no matter the ill effects, work for 60 hours a week, embrace masculinity, stop thinking about historical prejudice and maybe fire everyone who ever did.

It seems like a good time to update our thinking about the ethical landscape of Generative AI. I’ve found it impossible not to feel as if I’m missing the forest for the trees, so I’m going to ground this analysis in some foundational questions: 

• What ails us as a society?
• What kind of future do we need to build to serve as a remedy?
• What best practices should we adopt in our current work to help make that future a reality? 

These aren’t just ethical questions from my point of view. I believe the best long term bets in tech product strategy can be modeled from the assumption that society tends toward equilibrium between cultural and economic stability. 

For now, I’ll spare you more prose and suggest to you these brainstorming prompts: 

Societal issues adjacent to AI

• Isolation and the need for community
• Practical impacts of historical prejudice meeting the reality of degraded quality of life for majority classes
• Costs and incentives of maintaining a broadly-educated society
• Capture of innovation by a sclerotic capital class

Present Day impact of AI on each issue

Isolation

Reduces the need for communal transmission of information, elevating a non-human intermediary as the source of truth.

Historical prejudice 

Freezes culture in time by sourcing informational interpretation from the past. This can be mitigated, but these corrections are often untenably awkward in visual contexts. 

Some untenably awkward AI generated visualizations from Google Gemini circa February 2024.

Broadly-educated society 

Encourages students to outsource foundational thinking to models, forcing educators to rely on in-class short form analysis projects vs. in-depth take home material. Also encourages legacy knowledge businesses, such as Nature, to orient their business toward generating lower quality but higher volume automated reports.  

Capture of Innovation

The emergence of high-end, low-cost models and open source AI implementations may wind up making fools of everyone from Open AI and Nvidia’s largest investors to the U.S. government. 

The questions “Why create this risky social thing if Meta will clone it and crush me?" and "why innovate affordable physical products if Amazon will clone it and crush me”? predate AI. 

There are clear benefits of AI in expanding technical access to build out useful software – I recommend you read Kevin Roose’s latest NYT tech column for that. “Vibe-coding” (I hate that term, Kevin, but we will defer to your aura here) can make communal coding experience projects like Glitch and Replit feel truly revelatory. 

That said, we’re still in a fairly desperate search for the next killer AI app. The fact that Google spends millions on ads where you can make it look like your dog jumped 3 feet higher is fairly unassailable proof of this. 

[According to the top industry marketers, lying about your dog’s athletic ability is one of the best selling points for consumer gen AI] 

The lists I’ve put together above tell us something important about how the next major AI product implementations might look. Here’s how we can complete this exercise in a way that improves our strategic thinking about AI products: 

First, make your own list of societal issues adjacent to AI. I personally focus on the tech education space in my present work, but your take on that may have a different focus. Then, define the ways AI is currently influencing those issues (I’d recommend doing so in a longer form than what I offer above. Really try to get into the weeds and challenge your preconceptions.) Finally, create a broad-brush outline of the kinds of products – however tricky or undefined – that would allow AI products to address societal problems in a more productive way than we’re seeing today. 

Here’s an example in miniature, based on my lists above. It seems as If I personally believe AI products that solve for societal-scale issues might: 

• [Isolation] Use the sum total of prompt activity to facilitate communal knowledge sharing, particularly based on distance, real-world or interest connection. 
• [Historical prejudice] Allow creation and easy distribution of scalable technological solutions specific to certain geographic communities or cultural groups. 
• [Broadly educated society] Build rule-based research assistance systems that allow humans to probe large sets of data for specific bits of information and the craft an expression of that information that relies entirely on their own analysis of the data 
• [Capture of innovation] Enable sustainable tech development that serves relatively small groups of people with intricate, experience-based solutions that work so well, major $$$ players won’t be able to make scaled imitations that exchange fidelity with reliability and ease of access. 

My new thesis on the ethics of AI applications is that most unethical conduct we’re seeing today is best thought of as the tailpipe exhaust of concepts that cannot culturally scale. Taking a step back to consider the shape of this troubled era, I've found, helps to clarify our priorities.