AI: AGI, the Mother of all tech changes. RTZ #354

AI: AGI, the Mother of all tech changes. RTZ #354

The Bigger Picture, Sunday May 12, 2024

First of all. a VERY Happy Mother’s Day this Sunday, to the most important better half of our families everywhere. Wish you all the very best with your family celebrations.

Second, for this week’s RTZ ‘Bigger Picture’, and at this point in the AI Tech Wave, I want to address the timing debate over AGI (aka Artificial General Intelligence, Super-Intelligence, Singularity, and a range of other variations).

Since the dawn of computers over half a century ago, through all our science fiction in print and on the screen, and thus in our imagination, AGI, or a smarter than human ‘digital’ intelligence, has settled in our fears and aspirations, decades before its time. The most recent, 7th installation of Tom Cruise’s multi-billion dollar Mission Impossible franchise, with its AI Villain, being the most recent example.

Especially with Microsoft/OpenAI just deploying an ‘Air-gapped’ Azure GPT-4 Government Top Secret cloud for the Department of Defense (DoD). Fiction and Fact, blurring at a furious pace.

And with the exponential recent changes in LLM/Generative AI of late that I’ve talked about here for almost a year now, it’s becoming more urgent to talk about it again. With LLM models and their underlying AI GPU hardware exponentially doubling 10x or more every two years, and depreciating 10x or more in costs in the same time, it’s all the more important to talk about WHEN it may be here.

It’s potentially the Mother of all tech changes in the next few years, building on our collective digital knowledge. So let’s discuss. For now the timing, and in later posts, the countless implications, some potentially bad, but mostly good over time (IMHO).

AGI and its timing It came to the fore again after a series of discussions at conferences and meetings in the Bay area this past week with AI industry luminaries. I’ve talked in recent posts about the exponential AI Scaling Laws that are driving relentless improvements in LLM AI models and the underlying hardware. For humans, linear thinking is much easier than exponential. As I’ve said before, thirty paces down a football field gets us down part of the way, thirty steps exponentially gets us to the moon. With AI, we’re facing a thicket of exponential changes all at the same time.

If one asks AI luminaries when to expect AGI, the point where computers could get ‘smarter’ than humans, the answer varies. From the nect 2-3 years to a decade or more. I put the question to the AI search engine Perplexity, and here’s what I got:

“The question of whether we will achieve Artificial General Intelligence (AGI) by 2025 is a complex one, with opinions varying widely among experts, industry leaders, and researchers. Based on the sources provided, there is a broad spectrum of predictions regarding the timeline for AGI, ranging from the imminent to the distant future.”

“Near-Term Predictions”

“Several industry leaders and tech entrepreneurs are optimistic about the near-term feasibility of AGI:”

– “Elon Musk predicts AGI could emerge by 2025, suggesting that we are on the brink of creating AI that can perform any cognitive task that a human can.”

– “Sam Altman (OpenAI) believes AGI could be reached in four or five years from 2023, indicating a timeline close to 2027 or 2028.”

– “Dario Amodei (Anthropic) suggests human-level AI could arrive in two or three years from 2023, which aligns with a timeline around 2025 or 2026”.

– “Jensen Huang** (NVIDIA) expects that AIs will be “fairly competitive” with humans within five years from 2023, suggesting a timeline around 2028”.

“These predictions are supported by the rapid advancements in computing power and the development of increasingly sophisticated AI models.”

“Mid to Long-Term Predictions”

“Conversely, several experts and surveys suggest a more cautious timeline”:

– “Ray Kurzweil maintains his prediction of human-level AI by 2029, a timeline he has held since 1999”.

– “A survey conducted by the AI Sentience Institute (US) in 2023 places the median probability of AGI’s arrival between 2050 and 2100”.

“Skepticism and Challenges”

“Despite the optimism from some quarters, there is also significant skepticism and caution regarding the timeline for AGI”:

– “The complexity of achieving AGI, which involves not just computational power but also advancements in understanding cognition and consciousness, is a significant challenge”.

– “Historical predictions about AI have often been overly optimistic, suggesting a need for caution in interpreting current predictions”.

– “Ethical and societal implications of AGI, including the potential for job displacement and economic instability, add layers of complexity to its development and deployment”.

“Conclusion”

“Given the wide range of predictions and the significant challenges that remain, it is difficult to definitively say whether AGI will be achieved by 2025. While some industry leaders are optimistic about near-term prospects, many experts suggest a more cautious timeline, with significant milestones expected in the coming decades rather than the immediate future. The development of AGI remains a moving target, influenced by both technological advancements and societal considerations.”

The full answer, with sources, is again, available here. And you can adjust the question to your satisfaction on Perplexity, ChatGPT and many other AI sources.

But from my perspective, we’re likely to see exponential improvements on a wide range of narrow AI tasks (aka ANI), in the next couple of years. OpenAI’s text to video SORA which I’ve written about here and here, are but one major and early drop in the bucket on the road to better sources of artificial or augmented Intelligence.

The AI futurist Ray Kurzweil pencilled in a rough crossover point of machine and human intelligence at about 10 quadrillion calculations a second at under $1000 price point. Given that software then takes a decade or so to catch up, that triangulates to roughly our current decade.

Given the wide range above, I’d say for most mainstream folks, it’s likely good to assume some pretty spectacular improvements and repercussions of AI in the next six years, by 2030.

A nice round number. between half and a full decade. That next six years will likely give us more changes than we’ve seem in the last quarter century. Perhaps more. Brace and Embrace with positive orientation. Stay tuned.

(NOTE: The discussions here are for information purposes only, and not meant as investment advice at any time. Thanks for joining us here)





Want the latest?

Sign up for Michael Parekh's Newsletter below:


Subscribe Here