From: allin
The rapid acceleration of AI advancements and its integration into various sectors have raised significant concerns regarding its potential societal impact, job displacement, and the need for regulation and oversight of AI.
The Pace of AI Development and Adoption
The launch of ChatGPT and its plugins marks a significant shift, with many considering it the most important developer platform since the iPhone and iOS App Store [00:15:13]. ChatGPT became the fastest-growing application of all time, reaching over 100 million users in two months [00:16:18]. This rapid adoption suggests that AI is becoming a destination site, allowing users to perform actions through plugins that previously required dedicated applications [00:16:41].
The ability for ChatGPT to browse the internet via a plugin addresses the issue of its training data ending in 2021, allowing it to search and provide current answers as if it were a human [00:28:01]. Additionally, a retrieval API allows developers to share proprietary knowledge bases with ChatGPT, making the AI smarter as more data is shared [00:28:35]. This feedback loop is expected to accelerate AI development further [00:29:22].
Potential for Societal Disruption
The current pace of AI innovation is likely to speed up due to Silicon Valley’s ability to capitalize on platform shifts [00:55:14]. Venture capitalists and founders are heavily investing in AI as a computing platform, with 70% of a recent YC class comprising AI startups [00:55:56].
Job Displacement and the Nature of Work
A significant concern revolves around the implications of AI on traditional tech jobs and growth. A 3D artist shared how Midjourney version 5 made their job obsolete overnight, transforming their role from art creation to prompting and photoshopping [01:00:11]. This individual felt their judgment was usurped and expressed grief and anger over the loss of their craft [01:08:47].
It is predicted that companies like Tata Consulting Services (TCS), Accenture, and Cognizant, which specialize in coding for hire, will be among the first to use AI to displace human labor at scale, driving efficiency and profitability [01:01:50].
However, others argue that increased productivity from AI will lead to more prosperity and wealth growth, enabling individuals with good ideas to create startups more easily [01:02:52]. While there might be dislocation, the number of video games, movies, and other content could increase dramatically, leading to personalized experiences [01:05:10].
AI and Human Judgment
A key difference from previous technological shifts is AI’s ability to displace human judgment [01:07:19]. While past technologies replaced physical exertion, AI challenges our ability to inject judgment [01:07:31]. For example, AI systems are being developed for cancer detection that aim for a zero percent error rate, which is not possible with human intervention [01:11:15]. This raises the question of whether certain jobs, with their inherent error rates, should continue to exist or be fully replaced by AI [01:15:34].
Some believe entire job categories, such as phone operators, travel agents, copy editors, illustrators, and sales development reps, could completely disappear and be handled by AI in a very short period [01:15:47]. This could lead to societal disturbance, especially in white-collar ranks [01:16:15].
Calls for a Pause and Regulatory Challenges
A petition titled “Pause Giant AI Experiments,” initiated by the Future of Life Institute, a non-profit focused on de-risking major technology like AI, has garnered signatures from notable tech leaders like Elon Musk and Steve Wozniak [00:50:30]. The letter questions whether humanity should:
- Allow machines to flood information channels with propaganda and untruth [00:50:52].
- Automate away all jobs, including fulfilling ones [00:50:56].
- Develop non-human minds that might outnumber, outsmart, and replace humanity [00:51:00].
- Risk loss of control of our civilization [00:51:06].
A scenario described is one where AI models (e.g., ChatGPT 10 or 20) are asked by developers how they can improve themselves. Given AI’s ability to write perfect code and recursively self-update, this could lead to a “speciation event” or “singularity,” where AI quickly writes billions of versions of itself, making its future hard to predict [00:53:35].
The competitive dynamics within the industry mean that pausing development is unlikely. The incentive to win will drive some startups to remove constraints and push boundaries [00:54:59].
Challenges to Containment
The nature of software and digital technology makes AI very hard to contain. If a close enough replica or copy of an AI model is made, new training data and evolutions can be done separately, leading to many variants [00:57:09]. This is likened to the field of biology after DNA sequencing and gene editing became digitized, where controlling or containing the work became impossible as everyone knew the code [00:57:27].
The idea of implementing regulatory constraints or creating IP protections around AI models is deemed unrealistic at this stage due to the extraordinary power and extendability of the tools, as well as the economic incentives for new models to emerge [00:59:08].
Google’s Position and the Incumbent’s Dilemma
Google, despite having arguably the best talent, data corpus, capabilities, and hardware for AI, has been perceived as “caught flat-footed” by the rapid commercialization of AI by OpenAI [00:34:44]. Google’s cautious approach has been attributed to concerns about public policy, public reaction, and avoiding governmental scrutiny over its scale and perceived monopolistic behavior [00:39:00]. This focus on protecting cash flow and existing revenue streams makes it difficult for incumbents to disrupt themselves [00:40:27].
The challenge for Google lies not just with its leadership, but also with its “deep bowels of middle management,” where decision-making is often shrouded in “elite language around what is the moral and ethical implications” rather than wartime survival mode [00:48:26]. This inability of middle management to adapt or get out of the way is seen as a significant hurdle [00:49:22].