Exponential AI Development is Here: We are now in a Fast Transition to (Ro)bots that will both Complement and Replace Humans at Work

We need: AI Skills; job retraining: AI Literacy; metacognitive skills; a social adjustment fund; AI tsunami instead of a wave

The world is changing so fast now in the AI space…This is what our friend Ray Kurzweil has projected for some time..I mean we're getting indeed exponential acceleration and as Ray and others have predicted for a long time …Advanced AI technology in other areas is accelerating super, super fast and we don't need the acceleration to be super fast everywhere to get to a singularity; right, you just need it to be super fast in a few judicious places.
– Ben Goertzel, CEO of SingularityNET (November 29, 2023)

It's too early to make 2024 predictions when there's still more than half of 2023 left
​​(in AI's non-linear time scale)
– Julien Chamond, CTO, Hugging Face (December 1, 2023)

I’ve. Never. Seen. Such. Rapid. Evolution. In. Such. A. Short. Time.
It’s very clear that AI is evolving faster than a human can,
this is exponential capabilities in such as short time.
– Jeremiah Owyang,  Blitzscaling Ventures (November 30, 2023)

The technology is improving exponentially fast…faster than even the technologists expected...There seems to be no end in sight…It’s almost accelerating.
–  Eric Brynjolfson, Stanford Digital Economy Lab (November 2023)

TL;DR

• Regardless of whether we get to AGI, its growth is now clearly exponential.

• A walk-through of exponential developments related to media, ending with robotics.

• The are AI impacts we are already starting to see in labor markets.

• The need for AI skills and literacy training, metacognitive skill development, and advocacy for a social adjustment fund.

• References from leading AI scientists and economists who study the issue and are starting to see real impacts are included.

Introduction

Have you been feeling overwhelmed by the pace of AI developments lately? This is not surprising, as we are now in the exponential phase of AI development and it’s unlikely that you’ve ever experienced anything like this in your life. John Nosta explains:

Each layer of learning and adaptation doesn’t just add to the AI’s capability; it multiplies it, enabling the system to make leaps in understanding and application. This represents a fundamental shift in how we approach challenges and opportunities across various sectors, paving the way for AI to become a central driver in our quest for advancement and innovation.

Linear change, characterized by a steady and predictable rate of progress, is evident in the early evolution of computer technology, where each advancement added a consistent amount of improvement. This incremental progression aligns with human experiences like aging, making it intuitive and easy to forecast. For example, the evolution of phones from home phones, to car phones, to mobile phones with simple call-and-text devices modeled on computer typing represents a linear change that occurred over a long period (decades). This is also true of the internet. Here, in this hypothetical but realistic graph, we can see the comparison between linear and exponential growth.

Exponential Developments in AI

As we now see in technologies like AI, exponential development is a concept where advancements accelerate over time, often following an exponential curve. This type of change is initially slow but picks up speed rapidly, leading to significant transformations in a short period of time. The progress builds upon itself in a multiplicative way, with each advancement paving the way for more complex and powerful innovations.

An Example

In what follows, I’m going to walk you through an example of how exponential development is occurring in AI in media, with multiple AI developments building on top of each other. This is just one example; I do not touch on the commonly discussed gains in intelligence or any of the following —

• How AIs can write more than 40% of the computer code that helps develop more advanced AIs

• How AIs are helping to develop faster chips that perform the AI computations that develop the AI models and process the prompts

• How AIs are finding millions of new materials that AIs are now synthesizing to make new materials

• How AI is helping to train AI (RLAIF)

• How AIs are figuring out ways to use synthetic data the models create to further scale the models

• How AIs are helping to develop new models that can reason and plan; what happens when AIs that engage in abstract reason and plan create their own scientific hypotheses and build new AIs.

My only goal here is to demonstrate rapid, exponential growth towards robotics — the practical side of things.

This example starts with the development of natural language conversation/processing (NLP) and finishes with current advances in robotics development. As you read, think about how these are integrating and being built on top of each other (conceptually), leading to exponential change that will eventually produce a replicated human (and probably an AI that is better than a human). Most of these developments occurred over the last year, starting with the release of ChatGPT3.5, which demonstrated significant advances in natural language processing.

*Advances in natural language processing led to the ability of AI systems to have conversations with users and code, which now supports much faster development of all subsequent improvements in AI and its applications.

*Advances in scaling AI language models, plus fine-tuning, along with RLHF, RLAIF, and RAG to reduce hallucinations, mean AIs possess more knowledge than any human and even collections of many humans.

*Developments in text-to-image generation using diffusion models made it possible to produce some cool photos.

Image generation and speech synthesis, which transform text into natural-sounding audio, made the development of avatars possible. 

*Advances in avatar development, computer vision, and speech recognition made it possible to develop avatars that represented actual people speaking in their original voice. 

If a person is well-known, they can even produce text (and then voice) similar to text that AIs trained on that is available online (knowledge).

*Advances in speech recognition, speech synthesis, image recognition, and language translation have made it possible to translate someone’s speaking in their original voice into other languages.

And the language translation abilities keep getting stronger and can now happen in real-time.

*Advances in video generation made it possible to develop Animatable Gaussians, a new avatar representation method that can create lifelike human avatars from multi-view Red, Green, and Blue (RGB) videos.

[I‘m still “blown away” by how real this next individual looks]l

​​

But we aren’t stopping here. Microsoft just unveiled GAIA (Generative AI for Avatar), an innovative technology that is transforming the landscape of digital communication. GAIA stands out from earlier methods by offering a unique way to create talking avatars using only a single portrait image and speech. This technology employs a two-part process that separates movement and appearance in video frames and then produces motion sequences based on speech. 

The distinctiveness of GAIA lies in its integration of a Variational AutoEncoder and a diffusion model. This combination surpasses prior models in delivering avatars with greater naturalness, diversity, improved lip-sync accuracy, and enhanced visual quality.

3D modeling and animation technologies can be employed to produce 3-D videos.

Video editing and composition software, combined with advances in video generation, then seamlessly compiles these elements into a coherent video. 

Advances in building simulated emotions into AI make it possible to have emotionally supportive interactions.

Advances in computer vision and natural language processing have made image-to-text possible.

Text-to-voice means the AI-produced description of that image can be spoken.

Advances in AI and VR make it possible to experience these scenes, including the one on the right with our friendly avatar, virtually.

What world are we moving to?

We are moving to a world where we can have text and voice conversations in at least hundreds of languages with bots that look like real people about any topic of our choosing. These bots will be incredibly knowledgeable, perhaps more knowledgeable than any other humans, and they will likely be able to demonstrate strong emotional intelligence. 

Investor Jeremiah Owyang explains it this way:

In spring 2023, small hackathon teams could use GPT to create a simple text-based chatbot in two days. Fast forward to November, I've judged several hackathons in SF and witnessed astonishing advancements. Several months later, in Fall 2023, small hackathon teams now can assemble "sentient creatures" like a 4-year-old, in two days.

• "Seeing" with GPT-4 computer vision, what’s happening in the real world,

• "Hearing" via voice commands and ambient sounds in the real world,

• "Thinking" through processing the above real-world input,

• "Learning" by accessing the pre-trained data GPT offers,

• “Referencing” exclusive data sets using Retrieval Augmented Generation (RAG),

• "Speaking" with life-like audio voices, that have inflection and tone in any language,

• "Writing" through text communication, in any format or style required,

• "Drawing" by creating images spontaneously.

• “Interacting”: it proactively can engage in dialog, ask questions, or assign AI agents to complete tasks on their own.

With advances in robotics, AIs will become more lifelike and “live” outside of our computers.

Amazon now has more than 750,000 of these friendly “folks” deployed.

And just as the original developments and improvements in deep learning underlie the development of most of these technologies, so does the development and advancement of chips. AI is contributing to advances in both of these areas.

What does exponential growth mean?

Psychologically, it’s shocking and overwhelming. Every day, the case for AI literacy and public awareness grows.

Practically, the simple reality of exponential AI development means the time frame for AIs being able to do more and more of what humans do at work is collapsing faster.  We already have AIs being used in call centers, including for political campaigns, and on factory floors in the form of robots. 

1. This means we need to prepare graduates and other adults to work with AIs.  According to the recent Amazon jobs report, 92% of companies will integrate AI into their workflows by 2028.

   

   Those who are prepared to work with AI will also get big raises (35%+)

   

2. We need to help students develop the metacognitive skills needed to work with both AIs and humans. This is not just about AI.

Developing speech & debate programs is a great way to start :).

3. It means workers will be replaced with AIs, and it means that we need to support the development of social adjustment funds to support workers whose jobs are replaced by AIs. This will be especially difficult in a world where a (US) Congress is committed to not increasing spending.  To get back to speech & debate, individuals will need to develop strong advocacy skills in a world of rapid change and resource distribution that will certainly magnify inequality.

This is not hype, and you do not have to believe me. Leading AI scientists Fei Fei Li (known as the “Godmother” of AI) and Eric Brynjolfson, Director of the Stanford Digital Economy lab, make the same points: The “AI World” is developing faster than even the technologists thought it would, that there will be significant economic dislocation at least in the short-term, and that public awareness of these developments needs to take place immediately.

Still not convinced?

I think what people are expecting is a wave. We are going to get a tsunami.
– Vinod Khosla, Billionaire Investor, December 1, 2023

AGI in 3 Years

Michael Simmons

Job Losses Due to AI: The elimination of jobs resulting from the prolific use of AI could leave an entire generation of college grads scratching their heads as the market for skilled labor disappears. This on top of the trauma from the COVID-19 pandemic could be a formula for civil unrest. How do we retrain or upskill in time? Can we realistically make some version of UBI work? These are all questions that will become more pressing as the pace of AI development accelerates.

Peter Diamandis in his newsletter (December 3)